Fable 5 is pitched to execute multi-day knowledge work, not merely speed it up. For agencies and publishers built on billable hours, the harder question is no longer what the model can do, but what it does to the business model.

Two months ago, Anthropic told the world it had built a model too dangerous to release. On Tuesday, it released it anyway (well sort of).

Claude Fable 5, launched on 9 June, is the first of Anthropic’s “Mythos-class” models, a new tier sitting above Opus, until now the company’s flagship. The underlying Mythos model captivated Wall Street and government officials when it was unveiled in April, largely because of its facility for identifying security flaws in software. Anthropic said at the time it had no plans to make it generally available, restricting access to Project Glasswing, a collaboration with Amazon Web Services, Apple, Google, Cisco, Microsoft and JPMorgan Chase, among others, who used it to find vulnerabilities in their own systems before anyone else could.

Fable 5 is the public version of that technology. It uses the same underlying model as the restricted Mythos 5, with one structural difference: classifiers detect requests in certain categories (cybersecurity, biology and chemistry, and attempts to extract the model’s capabilities through distillation) and route them to the older Claude Opus 4.8 instead. Anthropic says the safeguards trigger in fewer than 5% of sessions, and that an external bug bounty programme involving more than 1,000 hours of testing found no universal jailbreak.

For marketers and publishers, the cybersecurity story is background noise. What matters is what the model does the rest of the time, and the terms on which it does it.

From assistant to employee

The capability claims follow a now-familiar pattern. Anthropic says Fable 5 is state-of-the-art on nearly all tested benchmarks, with its lead over previous models growing as tasks get longer and more complex. It can work autonomously for longer than any prior Claude model. Third-party testimonials point the same way: Stripe said the model compressed months of engineering work into days, and that it completed a codebase migration that would have taken a team more than two months.

The autonomy claim is the one that should concentrate minds in agencies and publishing houses. The previous generation of AI tools made individual workers faster at drafting, summarising and analysing. The pitch for Fable 5 is categorically different: delegate the deliverable, not the task. A market analysis, a campaign build, a content audit across a full archive, handed over, supervised lightly, reviewed at the end.

If that pitch holds up in practice, the economics shift. Agency margins are built on billable human hours applied to exactly this category of work. Publishers’ cost bases are built on production workflows that assume humans at every stage. A model that genuinely executes multi-day knowledge work changes what a lean team can ship, and consultancies are already constructing offers around it. One US firm claims Fable 5 plus a data source and an orchestration layer can replicate most of what six-figure personalisation platforms deliver, at a fraction of the cost, pricing real-time personalisation into reach for mid-market companies that were previously locked out.

Treat the vendor enthusiasm with appropriate scepticism. Every model launch since 2023 has arrived with testimonials about transformed workflows, and the gap between demo and deployment remains where most AI budgets go to die. But the direction of travel is consistent, and the price signals intent: Fable 5 and Mythos 5 cost $10 per million input tokens and $50 per million output tokens, less than half the price of the Mythos Preview. Anthropic is pricing this for scale.

Read the small print

Two commercial details deserve more attention than the benchmarks.

First, access. Fable 5 is included in Pro, Max, Team and seat-based Enterprise plans only until 22 June. From 23 June, it will be removed from those plans and require usage credits, with Anthropic citing unpredictable demand and promising to restore it as a standard feature “as soon as possible”. A twelve-day free window followed by a metered model is a soft launch dressed as a giveaway. Teams piloting it this month should budget on the assumption that the marginal cost from July is real.

Second, and more consequential for anyone handling client data: Anthropic will require 30-day retention on all traffic to the new models, even for enterprises that previously held zero-retention agreements. The stated rationale is safety monitoring. The practical effect is that agencies running confidential client briefs, publishers processing unpublished editorial, and martech vendors passing customer data through the API all need to revisit what their contracts promise. Zero-retention clauses have been a standard reassurance in enterprise AI procurement for two years; Anthropic has just shown how conditional that reassurance is.

The gatekeeper’s dilemma

Underneath the launch mechanics sits a bigger question. Fable’s release comes as Anthropic prepares to enter the public markets, and days after the company urged major AI labs to establish a coordinated “brake pedal” on frontier development, warning that systems may soon improve themselves without human intervention. Critics have noted the tension: the warning-heavy framing casts Anthropic as both the source of the new capability and the gatekeeper deciding which governments, companies and researchers get to use it. The Verge reported that unauthorised users accessed Mythos after its limited rollout, an awkward incident for a company whose brand rests on responsible deployment.

None of this means the capabilities are overstated. It does mean the industry should be clear-eyed about the structure taking shape. The most powerful general-purpose business tool on the market is now sold by a company that decides, unilaterally and revisably, who gets which version, what questions it will answer, and how long your data sits on its servers. For publishers, there is a sharper edge still: the same model class that promises to streamline editorial production also powers the answer engines that are eroding the search traffic those publications depend on. The tool and the threat are the same product.

The sensible posture for the next quarter is neither evangelism nor abstention. Pilot the model on real work during the free window, with non-confidential material until the retention question is resolved contractually, and measure output against your current cost of production rather than against the previous model’s output. The benchmark that matters is not Fable versus Opus. It is Fable versus your payroll, and that comparison is now close enough that someone in your finance department will run it whether you do or not.

The industry still runs on an assumption that has gone largely unquestioned for two decades: that if you know enough about who someone is (their age, income bracket, browsing history, purchase patterns), you can predict what they want and when they want it. Demographic segmentation, behavioural lookalikes, and predictive models built on historical data: all are variations on the same premise.

Tyler Boyd thinks that premise has a fundamental flaw. “Demographic foundation is additive,“ he says. “It may be interesting, but it’s not defining.“

Boyd is CEO of Youtility, a US and UK based company that has been building what it calls a decision science platform, an AI driven system designed to identify not just who a customer is, but also what emotional state they are in now, in terms of engagement. The distinction sounds subtle, but the commercial implications are not.

The problem with past Behaviour

The standard critique of performance marketing is that it optimises against history. Boyd’s version goes further. Past-behaviour models, he contends, are working from data that may be years out of date — profiles built on PII and credit scores that haven’t kept pace with how someone’s life has actually changed. You got married, had a child, moved house, and changed jobs. Your marketing profile still says single, urban, early career. The segment hasn’t moved; the person has.

​

“If you’re a hungry guy and you want the two-for-one special, you want a deal seeker. Well, if you just had a steak, you’re no longer a deal seeker. That’s where your mental state is.“

Tyler Boyd, CEO, Youtility

​Mental state is dynamic. Someone is not permanently a deal seeker, permanently risk averse, or permanently in the market for a mortgage. They move in and out of those states depending on circumstance, context and recent experience. Advertising that treats people as fixed types will always be misfiring at the margins, and at scale, those margins represent significant wasted spend.

​

Youtility’s platform attempts to model those shifting states in real time. The system, in which Navier is the model and Stokes is the interface, named after the fluid dynamics equations that describe changing flow, has been built on over 230 million customer profiles. It scores individuals across 24 core emotional dimensions, grouped into eight defined categories, producing what Boyd describes as an individual level emotional profile rather than a segment level approximation. “We do not know anyone else,“ he says, “who can go to the individual level and tell you about the individual emotional drivers for a person.“

Think-alike versus look-alike

The practical output of this is what Youtility calls think-alike audiences: a deliberate contrast with the look-alike models that dominate programmatic advertising. Look-alike modelling finds users who resemble your existing customers in observable demographic and behavioural terms. Think-alike modelling finds users who share the same underlying mind state: the same emotional drivers and motivational profile.

Boyd is emphatic about why these matter in practice. Many organisations, he notes, are trapped by pre-defined demographic segments (“Rising metropolises”, “up and comers”, “young professionals”) that may describe their existing customer base accurately but constrain their view of who their next customer could be. “The 80-year-old grandmothers: a few of them are deal seekers. You haven’t talked to them about a new product because you didn’t think they’d ever want it. Well, actually, maybe they do.“

The commercial implication is that think-alike targeting does not just find more of who you already have; it finds people who think the way your best customers think, regardless of whether they fit the demographic profile you would have specified.

What the data says

The case for targeting emotional states is not purely theoretical. Youtility has campaign evidence across energy, banking and consumer goods, and the performance differential between a first campaign pass and a refined second pass is striking.

​Working with a leading UK bank, the company produced a 74% increase in click-through rates and a 17% increase in sales on an energy switching campaign. A more recent campaign for Monese, a European fintech with Barclays and PayPal among its investors, targeting existing customers for micro-loans of between £50 and £200, produced a 91% increase in click-through rates and a 24% increase in sales on the first iteration. When Youtility refined its emotional targeting based on how those customers had behaved (who clicked through but stalled at the FAQ page, who dropped off, who came back), the second round produced a 154% increase in click-through rates and a 285% increase in sales.

A standard A/B test would produce variants of the same message; Youtility’s system identifies that a customer who hesitated has moved from deal seeker to cautious and adjusts the message accordingly. Boyd’s description of what a conventional LLM-based approach misses is pointed: “All English language based models give you a derivation of a different way to say the same thing. We give you: Is this person a deal seeker? Is this person cautious? Has this person changed from a deal seeker to cautious based on the actions they took in response to the first email?“

The activation layer

Emotional profiling only generates commercial value when it connects to campaign execution. That is where Chicago based Fyllo comes in.​

James Ramelli at Fyllo describes the partnership in straightforward terms. Youtility defines the mindstate cohort. Fyllo works out how to find those people in the wild and get a message in front of them. “We’re taking that definition of mindset and actually finding the people through their addressable messages,“ he says. In practice, that means translating a cohort of emotionally profiled users into a list of devices (laptops, phones, connected TVs) and identifying which channels and formats will reach them most effectively.

The channel strategy component is not trivial. Ramelli’s point is that knowing a user’s emotional state should inform not just the targeting decision but the medium and timing of delivery. A financially anxious customer might be receptive to a savings product via a mobile push notification at a specific time of day, and less receptive to the same message in a thirty-second CTV slot. Youtility’s model can specify the recommended channel alongside the emotional profile; Fyllo executes across it.

​What Boyd is describing is a closed loop: response data from each campaign feeds back into Youtility’s model, sharpening the emotional profile for the next round. “Customers who are more inclined to convert at a higher rate come through,“ he says. “We’ve been talking to them in that tone and tenor since they saw an ad for the first time.“

The publisher case

The use case that should interest publishers most is also the one Boyd is loudest about. His pitch to the industry is direct. “Publishers have been sitting with an unused data source on their balance sheet since day one,“ he says. “It’s been growing, and it’s been considered goodwill, not a revenue generator.“

The argument is that a publisher’s first party engagement data, covering what subscribers read, when, for how long, where they drop off, what they come back to, what they never click, is precisely the kind of behavioural signal that Youtility’s model can interpret as emotional state. That interpretation can then drive more targeted advertising inventory, more effective subscription conversion messaging, and more precise retention campaigns. Youtility claims it needs only a handful of data points to produce an initial customer profile, with accuracy improving as the feedback loop accumulates.

For publishers facing the dual pressure of declining third party cookie signals and eroding referral traffic, the proposition is worth taking seriously. First party data has been the industry’s answer to both problems for several years; Youtility’s case is that, used to model emotional states rather than just behavioural patterns, it is meaningfully more valuable than most publishers currently extract from it.

Where this sits in the market

Youtility is a medium sized company (it has raised $4.2 million to date and is targeting a seed plus round), making large claims in a market full of them. The AI targeting space is noisy, and executives who have been through multiple cycles of targeting innovation are right to be sceptical of anything that sounds like a repackaging of what already exists.

What distinguishes Youtility’s argument is the specificity of the methodology and the granularity of the performance evidence. Think-alike audiences built on emotional profiling, scored at the individual rather than segment level, with a feedback architecture that refines targeting in real time: that is a meaningfully different proposition from Look-alike modelling with an AI wrapper. Whether it scales as claimed across a wider range of verticals and campaign types is still to be tested. But the early numbers are striking, and Youtility is, on this evidence, well ahead of the field it is competing in.

​

1. AI training is a theft the industry has been too polite to call that

A.G. Sulzberger of the New York Times made an accounting argument: an AI model is built from talent, compute, energy and data; the first three are paid for lavishly, the fourth simply taken. The six leading AI firms are worth roughly $11 trillion between them, yet the licensing deals struck with publishers amount to less than half of 1% of that. His prescription: defend copyright, license only on sustainable terms, and act collectively rather than one frightened newsroom at a time.

Read the full report: The New York Times calls AI’s founding act a theft

2. The Guardian’s defences are formidable. Whether they hold is a different question

Katharine Viner runs The Guardian, a paper with no proprietor, no shareholders and £125m in recurring reader revenue, 83% of it now from outside the UK. By the standards of the industry, an extraordinary position. And yet the threat she named without hedging was AI’s capacity to sever the direct link between the Guardian and its readers. Her answer is SPUR, the publisher coalition launched in February with the BBC, the FT, Sky News and the Telegraph.

Read the full report: ‘Lean Into Who You Are’: Viner on the Guardian’s Bet Against AI

3. Collective action or go it alone: SPUR versus Le Monde

WAN-IFRA announced a partnership with SPUR and 30 new members moments after Louis Dreyfus of Le Monde explained why he was not signing up. Dreyfus has cut his own deals with OpenAI, Perplexity and Meta, arguing collective bargaining moves too slowly. David Buttle, an architect of SPUR, countered that a deal is the illusion of control: the platforms are paying not to be sued, not for access.

Read the full report: SPUR Goes Global: WAN-IFRA and FIPP Back Publisher Alliance

4. OpenAI showed up, listened carefully, and committed to nothing concrete

Varun Shetty and Tom Rubin from OpenAI came with a coherent message: OpenAI is not hostile to publishers, having backed more than 160 newsrooms through a WAN-IFRA accelerator and a Lenfest Institute cohort. On the value exchange, though, the answers stayed vague: no revenue sharing, publisher dashboards still only “on the list”, and aspirations where a concrete commitment was asked for.

Read the full report: OpenAI Supports Publishers Everywhere except the Balance Sheet

5. Google is still sending traffic. The exchange rate is deteriorating.

Trigger an AI overview and the top five organic results lose around 40% of their clicks, with no shelter for evergreen content equivalent to the Top Stories box. Buttle drew the structural line: Google at least trades content for traffic; ChatGPT carries no equivalent obligation, and publishers still counting page views are using the wrong instruments.

Read the full report: Google is still sending you traffic. For now.

6. Newsrooms say all the right things. 61% spend nothing on training.

FT Strategies’ Future of the Newsrooms Study, drawing on 450 responses from 86 countries, argues newsrooms are entering a “community era” where original journalism and direct reader relationships are the edge commodity AI cannot replicate. Yet the gap between rhetoric and practice is stark: despite the stated urgency, 61% of newsrooms report spending nothing on training. The most valuable hire, the session concluded, is no longer the social specialist but the journalist with deep domain knowledge and strong sources.

Read the full report: Everyone’s Talking About AI Skills. 61% of Newsrooms Spend Nothing on Training.

7. Europe is legislating for media freedom. The money has not followed.

Henna Virkkunen of the European Commission pointed to the Media Freedom Act now being enforced and a proposed €3.2bn Media Plus strand from 2028, the first structured EU support to carve out a dedicated news objective. On copyright, the Commission has opened a call for evidence on the DSM Directive and the text-and-data-mining exception. Catherine Pégard, for the French Culture Ministry, argued the decisive scale for regulation is now the European one.

What Marseille settled, and what it did not

The congress agreed on the diagnosis and nothing else. Sulzberger litigates; Viner builds coalitions; Dreyfus signs deals; Buttle sets standards; none has yet proved sufficient. Publishers left with the question they arrived with: what do you have that a chatbot cannot replace, and will readers pay for it?

At the WAN-IFRA World News Media Congress in Marseille, FT Strategies launched the first instalment of its Future of the Newsrooms Study 2026, a global benchmark designed to track how newsrooms are actually changing, beyond what they say about themselves. The distinction matters more than it might seem.

Lisa MacLeod, Director at FT Strategies, has been working on the study for most of the year. It draws on roughly 450 survey responses from 86 countries, supplemented by interviews, publisher workshops, and an advisory board spanning journalism, academia, product, and media leadership. WAN-IFRA partnered on the research; Arc XP sponsored it.

The goal, as MacLeod framed it, was to ask a harder question than the industry usually asks itself: not whether newsrooms have a strategy, but whether that strategy is shaping what gets commissioned, invested in, or cut.

The answer, broadly, is no.

The community era, and the gap before it

The study’s central argument is that newsrooms are entering what FT Strategies calls “the community era.” When generic content becomes cheap to produce at scale, competitive advantage shifts to whatever is harder to replicate. For publishers, that means original journalism, trusted relationships, and loyal communities built around them.

MacLeod’s qualifier was equally direct. “I know this sounds obvious. We’ve heard it many, many times. What is very clear in the data is that that is not happening.” The phrases that fill conference sessions, “audience first,” “meet them where they are,” “value over volume,” are not translating into practice. The study identifies four areas where the gap is most acute: strategy, trust, capability, and skills.

Strategy: the commissioning problem

The first gap is between stated priorities and daily editorial decisions. Newsrooms are increasingly recognising engagement as a strategic metric, an acknowledgement that high-traffic days are not a sustainable business model. Audience engagement, sustainability, and reach all rank among the top stated goals for 2026.

A quarter of newsrooms, however, report that editorial priorities remain largely reactive, driven by news judgement rather than any wider organisational plan. The commissioning process is where this shows up most clearly.

MacLeod described commissioning as “the unlocker for value in newsrooms,” but the data suggests it is being left locked. Commissioning is still predominantly destination-first: journalists file for a single channel, and repurposing happens afterwards. Data is used to evaluate coverage after publication, rather than to shape decisions before it.

“There’s a big joke that the Olympics sneak up on us every four years,” she said. The joke wasn’t flippant. The inability to plan for predictable events is a symptom of a deeper problem: strategy exists at the organisational level but does not consistently inform what gets made.

Newsrooms with stronger alignment between strategy and editorial decision-making are slightly more likely to involve audience-facing roles in setting long-term priorities: heads of audience engagement, platforms, or community functions. Only one in ten currently includes a dedicated strategy role in that process.

Trust: the workflow that does not match the rhetoric

The second gap sits between what newsrooms say about audience relationships and how journalists spend their time.

The study mapped journalist time across the production cycle. Most of it still goes on production: writing, editing, fact-checking, formatting, publishing. Pre-publication planning and post-publication audience engagement receive a fraction of the time, despite years of investment in automation tools designed to relieve that burden.

MacLeod was pointed about it: “When newsrooms talk about deeper audience relationships, their workflow is still optimised for getting the story out, not learning from what happens after.”

On storytelling formats, the data shows a strong preference for explainers, fact-checks, and background pieces. MacLeod offered a sceptical read. She suspects the enthusiasm for explainers is partly a proxy for wanting to reach younger audiences, and that the assumption young readers need things explained to them is worth examining.

Capability: technology that is not trusted

The third gap concerns whether newsrooms have the tools and structures to deliver on their ambitions. Most are not confident that their current technology stack is fit for purpose or future-proof. Twenty per cent say it is not fit for purpose at all.

On AI, the most common measure of success is time saved. MacLeod does not find this encouraging. The barriers to wider AI adoption are not primarily about the tools themselves: they are about culture: scepticism among staff, unclear use cases, and a lack of strategic direction. Framing AI investment primarily as a mechanism for reducing headcount is, she argued, too narrow. “The challenge is really to move beyond seeing AI only as an assistant for speeding up tasks.”

Skills: confidence today, anxiety about tomorrow

The fourth gap is temporal. Newsrooms express reasonable confidence in their current skills. That confidence erodes sharply when they look three years ahead.

Most newsrooms are still organised around traditional desk structures, politics, business, culture, sport, rather than around audience behaviour, which is increasingly niche and format-driven. The challenge is less about hiring new skills than rethinking how existing ones are organised.

The skills most prioritised for future readiness are hybrid ones: editorial judgement combined with technical fluency, audience understanding, and operational capability across formats. The “bridge role,” someone who can move between editorial and product, between journalism and technology, keeps surfacing as the figure newsrooms say they need but struggle to build at scale.

The study also found broad support for some form of differentiated compensation for journalists whose personal brands and audience relationships generate disproportionate value, including higher base salaries, performance bonuses, and, in some cases, revenue share.

61% spend nothing on training.

The finding MacLeod returned to in her closing summary was the one that sat most awkwardly alongside everything else. Despite the acknowledged pace of change, the stated urgency around AI, and the widespread concern about skills readiness, 61% of newsrooms in the study report spending nothing on training.

“That just represents a really big disconnect between where we think we should be and where we’re going.”

The session also featured Gard Steiro, Editor-in-Chief and CEO of Norway’s VG, who offered a practical counterpoint: a detailed account of VG’s experimental unit VGX, a small fast-moving team operating outside the constraints of the main newsroom and building AI-native products designed to inform what the wider organisation does next. Jens Pettersson of NTM Sweden chaired proceedings and drew both threads together around the recruitment question. The emerging consensus: the most valuable hire is no longer the social media specialist or the video producer, but the journalist with deep domain knowledge, strong sources, and the willingness to work differently.

The FT Strategies study will be repeated annually. Whether the numbers move will be the test of whether this year’s conversations translate into anything more than that.

Publishers churning out commodity content and living off Google clicks are, in Barry Adams’s phrase, in serious trouble. He was one of three search specialists at WAN-IFRA, and the least diplomatic. Adams, an independent SEO consultant whose clients include the New York Times, came with a blunt read on AI overviews and where the traffic is heading. Clara Soteras Acosta, an SEO and digital strategy consultant and author of the SEO Playbook for Publishers, brought fresh data from Google I/O and a framework for acting on it. David Buttle, founder of DJB Strategies and a former director of platform strategy at the Financial Times, brought supply-side research spanning ten-billion-page referrals from eight major UK publishers. They disagreed on plenty. On one thing they didn’t: publishers treating this as a new crisis haven’t been paying attention to the old one.

Plan for zero

What would you do differently if you knew the Google traffic was going to zero? Soteras Acosta opened by asking publishers to build as if it already was. Not because anyone expects it to, but because it forces the right discipline. Planning for zero, she argued, is the only assumption that forces a publisher to stress-test everything it has been taking for granted. Even a slow, steady single-digit annual decline is enough to threaten a model with nothing else underneath it. The question is not whether to prepare for reduced search traffic, but whether you have already left it too late.

Soteras Acosta was just as careful, though, to push back on the doom-mongering. The “Google search is over” takes that have circulated for the past two years are inaccurate. Google remains, by a large margin, the dominant source of referral traffic for most publishers. ChatGPT, despite its growth, sits at a fraction of Google’s scale. Google isn’t dying, but the exchange rate of content for clicks is deteriorating, and publishers who haven’t built anything else are exposed.

AI overviews: the 40% problem

Adams walked through the mechanics. Trigger an AI overview on a query and the publishers sitting in the top five organic results lose around 40% of the clicks they’d otherwise get. In some cases it’s 10%; in others it reaches 70%. The average sits around 40%.

The partial protection for news publishers is real but narrowing. When Google detects a Top Stories box is relevant, it generally does not show an AI overview alongside it. But the shelf life of that window is shrinking: Adams cited research showing an AI overview appeared for a breaking news story just four hours after publication. For evergreen and informational content, there is no equivalent shelter. Adams was direct: publishers producing commodity informational content and relying on organic clicks are in serious trouble.

“Google has been telling us what it was going to do for fifteen years. What AI has done is accelerate that process. Now Google can actively replace that content with an AI.”

Barry Adams

Discover, social, and the shrinking window

Soteras Acosta’s presentation went beyond AI overviews to address where discovery is happening. Google Discover remains the primary traffic channel for many publishers: volatile, yes, but still significant. Her data from three publishers of different sizes showed Discover outperforming search for certain content categories. The risk is that the Discover feed is no longer just publisher content. In the US, Facebook and Twitter are appearing in Discover positions. In the UK and France, YouTube and Twitter sit at the top. The window through which publishers can reach audiences is closing on multiple fronts simultaneously.

She also flagged a structural shift in how Discover may be ranked going forward. Google signalled in December that social engagement metrics could feed into Discover placement, which means a publisher’s visibility in the feed may increasingly hinge on how its content performs on platforms it doesn’t control. That has direct implications for editorial strategy. Content that does not travel socially may simply not surface. Soteras Acosta was not alarmist about it, but she was clear: publishers need to understand which of their content generates social conversation, and treat that as an editorial input rather than a distribution afterthought.

AI Mode and the 1 billion number

Google claimed one billion monthly active AI Mode users in the figures released around Google I/O. Soteras Acosta presented those numbers; Adams interrogated them. His read was that most of those users clicked “show more” in an AI overview and were taken to AI Mode without intending to be there. Before Google made that switch automatic, deliberate daily AI Mode users numbered around 75 million, a small fraction of Google’s two billion daily users.

What the usage data does show, and what Soteras Acosta emphasised, is that search behaviour is changing regardless of which interface delivers it. More than one in six AI Mode searches are now multimodal, meaning image input alongside text, growing at over 40% month on month. Follow-up queries within a session are growing at the same rate in the US. People aren’t just searching more; they’re searching differently, in longer, chattier back-and-forths; and publishers built for single-keyword, single-click traffic simply aren’t built for that.

ChatGPT: a different problem

Buttle drew a sharper structural line. Google’s whole model rests on a value exchange: it uses publisher content to rank results, and in return sends traffic. That relationship gives publishers leverage, however diminished. ChatGPT has no equivalent obligation. OpenAI has signed a small number of licensing deals, but Buttle’s view was that those are principally about litigation risk management, not a genuine commitment to publisher value. The implication: ChatGPT’s growth is bad for publishers not just because it takes share from Google, but because it does so without any reciprocal traffic obligation at all.

The competition cuts the other way too. The more dominant Google stays, the less reason it has to keep up a meaningful exchange with publishers at all. Soteras Acosta noted that European fragmentation makes a coordinated publisher response harder: the UK’s nascent coalition approach has some precedent, but individual markets are operating under different regulatory frameworks and at different stages of AI overview rollout. France, for now, has had limited exposure to AI overviews. That is a lag, not a long-term protection.

The KPI problem

Soteras Acosta’s most pointed argument was about measurement. Publishers are still evaluating performance in terms of page views and unique users. Those metrics made sense when traffic was the engine of the business. They are the wrong instruments for what publishers need to build now. Engagement, time on site, retention, community conversion: those are the numbers that matter now, not because they’re more virtuous, but because they’re what’s left standing when referral traffic dries up.

She gave two examples of publishers that had reoriented around this. One Argentine economics publisher created a board game that teaches families how household economics works in everyday life, community engagement as a direct editorial product. The New York Times Cooking section, facing the same evergreen traffic erosion as everyone else, shifted its content strategy toward original formats and creator partnerships. Neither is a template. Both illustrate the same underlying logic: the value a publisher creates for its audience must exist independently of whether Google sends anyone to find it.

What GEO means

Adams closed on generative engine optimisation, the consultancy world’s repackaging of SEO for the AI era. His verdict was blunt: GEO is SEO with a 50% markup. The fundamentals have not changed. Authoritative, well-structured, non-commodity content that serves genuine user intent is what ranks in organic search, gets cited in AI overviews, and surfaces in answer engines. Publishers who have been doing that consistently are in reasonable shape. Those who do not have a structural problem that no amount of rebranded optimisation will fix.

Buttle’s framing was complementary. The publishers that will survive are those that have become destinations rather than waypoints: brands with a defined audience, a clear editorial identity, and talent that creates a direct relationship between the publication and its readers. The mechanism for getting there is not a new SEO strategy. It is an editorial and commercial decision about what the publication stands for, and for whom. That decision, Soteras Acosta suggested, is the one publishers can no longer put off.

Barry Adams is an independent SEO and audience growth consultant. Clara Soteras Acosta is an SEO and digital strategy consultant and author of the SEO Playbook for Publishers. David Buttle is founder of DJB Strategies and a former director of platform strategy at the Financial Times. All three spoke at the WAN-IFRA World Congress in Marseille. Clara’s blog post from this session can be accessed here.

The session had been billed as one of the most anticipated at the congress, and the shadow of the New York Times lawsuit, whose publisher, A.G. Sulzberger, had addressed the conference the previous day, hung over it throughout, even though both OpenAI speakers declined to discuss it directly.

Varun Shetty, Vice President of Media Partnerships at OpenAI, and Tom Rubin, the company’s Chief of Intellectual Property and Content, came to the congress with a coherent message: OpenAI is not a hostile actor towards publishers. It has invested in the relationship, is building responsibly, and wants credit for showing up in conversations that it implies other AI platforms are avoiding. Whether that message landed depends on how much weight you give to goodwill versus commercial terms.

The case for the defence

Shetty opened with a biography. Both he and Rubin have backgrounds in the news industry. Shetty spent time at the New York Times, and Rubin worked as a journalist before becoming a media lawyer and serving 12 years on the board of the Reporters Committee for Freedom of the Press. The subtext was clear: these are not Silicon Valley disruptors who regard journalism as an obstacle. They are, they argued, genuine believers in the mission.

The institutional commitments they cited were real. Through the WAN-IFRA partnership, OpenAI has supported more than 160 newsrooms through an accelerator programme. A new cohort of publishers has just entered a six-month focused development programme through the Lenfest Institute, which OpenAI is co-funding. These are not trivial investments, and Shetty was right that comparable commitments from other AI labs are harder to point to.

On training data, the issue on which publishers most disagree with OpenAI, Shetty was concise and unapologetic. OpenAI considers the use of publicly available content in AI training to be consistent with copyright law, as US courts have recognised. He acknowledged that publishers disagree. He declined to let that disagreement end the conversation.

What publishers are asking

The questions from moderator Ezra Eeman, AI in Media Lead at WAN-IFRA, pressed on the points that matter commercially. The value exchange question came up repeatedly: if not traffic and not direct financial deals, then what exactly? Shetty’s answer remained consistent: audience exposure, quality traffic that converts at higher rates, and the opportunity to build on OpenAI’s technology. He acknowledged that the answer sounds vague at the top level, and on the evidence offered it is.

On traffic specifically, Shetty offered something that landed between candid and convenient: anecdotal evidence from publisher partners suggests that the traffic OpenAI sends is higher-intent and more likely to result in subscriptions, even if the volume is lower than expected. He was explicit that more statistical work is needed to substantiate this. That publishers are being asked to trust anecdote while adjusting to structurally lower referral traffic is not a point the session dwelt on.

The search product itself came up repeatedly. Shetty pointed out that ChatGPT Search launched just over a year ago and that the current experience, while imperfect, is an early iteration. He described it as a product that highlights trusted journalism, cites sources, and provides click-through opportunities, while acknowledging that news search accounts for only a small part of how users interact with ChatGPT overall. The implication, present throughout, was that publishers should take a long view, at a moment when many of them cannot afford one.

Share

The partnership model and who it reaches

One of the more pointed exchanges concerned who OpenAI’s existing deals reach. Most publishers in the room do not have a licensing agreement with the company. Shetty was straightforward about the logic: when launching a search product, you start in markets where you have the most users and secure relationships with a small number of trusted publishers to get it off the ground. You cannot start everywhere.

For publishers outside that initial cohort, the answer for now is primarily AI implementation work: helping newsrooms use the technology internally to improve editorial operations and audience products. Shetty placed considerable emphasis on this and cited several examples: publishers using AI to simplify and adapt content for different readers’ needs, to make archival material more accessible to journalists, and to structure complex data from public documents into reader-friendly formats. The pitch, essentially, is that publishers who are not yet in the search product can get real value from transforming their operations in the meantime.

Rubin added that OpenAI has active dialogue with a long list of publishers who are not formal partners, and that the door is open. When asked whether the partnerships team would grow to handle more of that demand, Shetty confirmed it would.

What was guarded

The session was not a cross-examination, as Eeman noted from the outset. Several questions went unanswered in any useful way. How much news-related activity is there on ChatGPT? Shetty said the volume is relatively small and that the company is working to release data in a form it can stand behind. Would OpenAI support emerging machine-readable standards for content permissions, beyond the existing robots.txt protocol? Something to discuss further. Publisher analytics dashboards? On the product team’s list.

Regarding advertising, Shetty confirmed that ads appear in the free ChatGPT product and are clearly distinguished from model responses. There is currently no arrangement for publisher content to appear alongside ads in a revenue-sharing model. When Eeman pressed on whether that constituted a fair value exchange, Shetty’s answer was brief: not at this point.

The Pulse footnote

A journalist in the room caught Shetty using the past tense to describe a product called Pulse, an AI-driven personalised digest that would surface content based on a user’s stated interests. Had it been discontinued? Not exactly, Shetty said. Pulse has always been compute-intensive and gated to a subset of users, and the company is currently making significant decisions about compute allocation. It was a minor exchange, but it carried a larger point: the gap between what OpenAI signals publicly and what it can deliver is a live issue, and publishers asking about it are right to do so.

The frame that didn’t shift

At the close, Eeman offered OpenAI the chance to make a concrete commitment for the year ahead. The responses were aspirational: Shetty wants to have evolved the search experience and be in a deeper conversation with publishers about technology adoption; Rubin wants to see meaningful innovation in reader experience. These are intentions, not commitments with a shape.

The session confirmed something the publishing industry has been navigating for three years. OpenAI is, by the standards of its category, a genuinely engaged interlocutor. It shows up, it listens, and it has put real money into newsroom support programmes. It also controls the architecture of a product on which publishers increasingly depend for audience discovery, and it has decided, so far, that the value exchange does not include revenue sharing or publisher analytics as standard. The relationship is real, but its terms still belong, for now, almost entirely to one side.

For three days in Marseille, the WAN-IFRA World News Media Congress kept circling one question: when publishers face the AI platforms, are they stronger together or alone?

On Wednesday morning the organisers gave their answer. WAN-IFRA announced a strategic partnership with SPUR, the industry coalition pushing for fair terms in AI content licensing and confirmed that 30 publishers had joined. The announcement landed moments after a panel in which one of France’s most successful AI dealmakers had spent the better part of an hour explaining why he was not signing up.

That dealmaker was Louis Dreyfus, CEO of Le Monde, welcomed pointedly from the stage as the new members were named. Le Monde has signed licensing agreements with OpenAI, Perplexity and Meta, and Dreyfus has gone on record urging other publishers to follow. His doubts about collective action were not hostile, but they were a sharp counter-argument.

The case for acting together

SPUR is the institutional bet on the opposite view. David Buttle, who founded DJB Strategies and previously ran platform strategy at the Financial Times, has been building it on a deliberately narrow premise. It is not a licensing body. It does not negotiate on anyone’s behalf. It sets the standards, transparency requirements and terms of trade that would let an AI content market function at all.

“The key relationships that matter now are with each other,” Buttle told the panel. Publishers entered the search and social eras fragmented, presented no unified demand, and lost control of their supply chains. A new distribution paradigm is forming around AI, and his argument is that the industry has one more chance to set the terms before the market hardens around it.

Doug McCabe, Chief Strategy Officer at the Guardian, made the historical version of that case the previous afternoon, and made it more forcefully. The industry, he said, is perfectly capable of collaboration when it chooses to be, and pointed to the infrastructure behind broadcast news as proof. The internet broke the habit. Each platform shift produced the same atomised scramble for traffic while the advertising economy migrated to Google and Facebook, and journalism was quietly downgraded along the way.

“What worried me most about the global internet era is that the industry as a whole has lost a kind of collective sense of agency.”

Doug McCabe, Chief Strategy Officer, The Guardian

McCabe put a number on the cost of inertia. Modelling done with Enders Analysis suggested that, over five years, a coordinated set of terms could be worth three or four times more to UK journalism than doing nothing, and that he called the conservative estimate.

The case for going it alone

Dreyfus had heard all of this before and was unmoved enough to say so. His account of the OpenAI deal was pointedly unglamorous. After ChatGPT launched, a French trade body had concluded it could not even find an address to write to. Dreyfus went to a conference, put his question to an OpenAI executive from the floor, sent an email afterwards, and closed a deal inside two months.

“You need reliable information for your LLM. We produce information. 83% of my revenues come from my readers. How can you expect me to provide reliable information without that source of revenue?”

Louis Dreyfus, President and CEO, Le Monde

The numbers he reported were the most contested moment of the session. ChatGPT, he said, was converting Le Monde subscriptions at 30 times the rate of Facebook and 173 times the rate of discovery traffic. The room did not take it on trust. Madhav Chinnappa, the session’s moderator, told him plainly that there was scepticism about the figures, and pressed on whether the underlying traffic was meaningful at all. Dreyfus conceded it was marginal and rested his case on the quality of the audience rather than its size, betting that even a halving of current conversion would leave Le Monde well ahead.

His wider point was about pragmatism. Collective bargaining, in his experience of the French negotiations with Google, moved too slowly and forced constant returns to members for sign-off. Someone large must move first and set the template. He was candid that this had made him unpopular, but unrepentant: without deals, he argued, the only alternative is litigation, and most publishers lack the resources to see a lawsuit through.

Or whether either of them is right

The most contrarian position on the panel belonged to its moderator. Madhav Chinnappa, a Reuters Institute fellow and former Google executive of 13 years, set out to disagree with everyone. His contention was that journalism’s value to the foundation models is low, because news is not central to products that are increasingly built for enterprise customers, and because the commoditisation that has dogged publishers for two decades applies here too. A single wire deal, he suggested, might give a model all the general news it needs.

Niamh Burns, senior research analyst at Enders, did not dispute that the value exists, but warned how little of the apparatus to capture it does. Advertising formats on ChatGPT shift constantly, subscription pricing is volatile, and the pot publishers are negotiating over barely exists yet.

Buttle’s toughest rebuttal was pointed at Dreyfus. A deal, he argued, is the illusion of control rather than the substance of it. Whether or not a publisher signs, its content is still taken, by the platforms directly or by third-party scrapers, and still used in ways that erode its business. The money on the table is real and taking it is rational. But the platforms are not paying for access so much as paying not to be sued. Until usage itself becomes the unit of value, with an audit trail showing how often a publisher’s work is served to a user, the market will reproduce the asymmetries of search.

The asymmetry nobody disputed

If the panel disagreed about tactics, a parallel session on collective responses left no doubt about the scale of the problem. Matt Rogerson, the FT’s director of global public policy and platform strategy, had shared a platform with McCabe the previous day. Here he walked through a regulatory picture that was improving in fragments and worsening in aggregate. The UK has paused a contested copyright opt-out, which would have let AI firms train on published work unless rightsholders actively objected, and is consulting on transparency. The CMA (the UK’s competition regulator) and the European Commission were due to set out remedies on Google’s content scraping during the congress. Brussels will not conclude its AI and copyright review until 2027.

Set against that, Rogerson argued, is a geopolitical shift that has narrowed what any regulator can do. The second Trump administration has shown it will use trade leverage to blunt international AI rules, and a UK government chasing tech investment has been making concessions it may regret. His more provocative thought was that the AI investment story may be turning: questions about energy costs, unmet promises and a tax base hollowed out by the redundancies the models are supposed to cause could yet take the shine off.

From Argentina, Martín Gonzalo Etchevers of the Argentinian News Media Association gave the asymmetry its sharpest geography. More than 600 million Spanish speakers, 19 countries, and on his account not a single comprehensive AI agreement among them. Advertising markets thinner than Europe’s or America’s, referral traffic falling away, and individual publishers with almost no leverage to bargain. His figures, that AI searches overwhelmingly end without a click and that attribution traffic has collapsed, are a trade body’s advocacy rather than independent data, but the direction is not seriously contested by anyone in the industry.

The bet, and the doubt

There is a flaw in Buttle’s model, and Dreyfus put his finger on it. Collective action rewards the small and asks the most of the large; in any such effort, as Dreyfus noted and Chinnappa agreed, the biggest players get the least back. That is precisely the calculation Dreyfus is running. He says he is watching SPUR closely, has friends among its founders and a breakfast with Buttle behind him, but cannot yet see what membership gives a publisher that already has direct relationships with the platforms and the will to use them.

Whether SPUR can answer that question will decide what it becomes. Thirty members and a WAN-IFRA partnership are a statement of intent, not a shift in power. The coalition needs the publishers who least need it, the ones with leverage to spend, and those are the publishers, like Le Monde, who have the least obvious reason to join. A coalition of the willing-but-weak would not move the platforms. A coalition that drew in the Dreyfuses of the industry might. On the evidence of Marseille, it has the argument. It does not yet have the man it most needs to convince.

This article draws on three sessions at the WAN-IFRA World News Media Congress, Marseille, June 2026.

The Guardian may be the best-protected newsroom in the English-speaking world. No proprietor to please, no shareholders demanding a dividend, a trust whose sole purpose is to keep the paper alive in perpetuity, and a reader-revenue base most publishers can only envy. Katharine Viner spent an hour explaining how those defences were built. Only once did the harder question surface, and she answered it candidly: whether any of it protects against the threat now coming.

The session sat within a framework set by Stig Kirk Ørskov, WAN-IFRA’s chief executive, who opened the congress. Press freedom, he argued, is a human good as much as a democratic one: line up the World Press Freedom Index against measures of public trust and wellbeing and the correlations hold. His precondition was blunt. Journalism cannot stay free if it is not financially sustainable, which turns a fair exchange of value between publishers and the AI firms ingesting their work into an existential question rather than a commercial nicety. That precondition framed everything Viner went on to say.

Viner has edited the paper for eleven years to the day, as her interviewer, AFP global news director Phil Chetwynd, reminded her. She took the chair a year before Brexit and a year before the first Trump term. “It’s been a decade of absolute turmoil,” she said, then widened the frame past media economics: more people now live under autocracies than democracies, the climate crisis is accelerating, and the West is logging its own figures on loneliness, poor mental health and young people locked out of housing and entry-level work. Her instinct throughout was to place journalists inside that picture rather than above it. “It’s important for journalists to see ourselves as citizens, not as a sort of different group of people sitting on top of society.”

The third crisis

Viner borrowed her diagnosis from the novelist Naomi Alderman: the information flood is the third great rupture in how humans handle knowledge, after the invention of writing and the advent of the printing press. The symptom is saturation, not scarcity. “We are flooded with information, flooded with facts,” she said, and the disorientation that follows, things you assumed true turning out false, things you dismissed as fake turning out real, is compounded by the sheer volume of deliberate manipulation. She cited the claim that most of what is now online is synthetic, and therefore probably fake. Chetwynd supplied a live example from the Iran crisis: the White House’s X account and the Iranian government’s, pumping out fabricated material in parallel, a point Viner took up at once. “How can you blame people for being confused about what they see?”

The conclusion is as commercial as it is editorial. Where the ground shifts underfoot, a recognised source of trust becomes a scarce asset. Facts alone do not close the gap. “You can’t just slam them on the table and say, well, here are the facts, because people will always provide another set of facts.” The Guardian’s answer is to wrap reporting in story and context, and to meet readers in the format they use rather than lecturing them: a reader who only watches video gets nothing from a 4,000-word essay. News avoidance, the industry’s running anxiety, is not something Viner says she sees in the Guardian’s own numbers.

Refusing the bait

The discipline shows most clearly in how the paper covers Trump. Viner traced the policy to a single morning in 2016, when she woke to find the Guardian, the BBC and The Times had each separately fact-checked the same speech overnight. “The whole world is going to spend the next four years fact-checking this guy. That is a waste of our collective time.” The Guardian’s response was to report less of what Trump says and more of what it does to people’s lives: “people as well as power,” a phrase she returned to repeatedly. The alternative is familiar to anyone who covered the first term: a newsroom reduced to a distribution channel, fed fresh lines morning, noon and night. Chetwynd recalled AFP’s Washington desk dismissing the bleach-injection remark as absurd, then having to cover it once it became consequential. Ignoring him is not an option either, which leaves the newsroom to weigh each remark on whether it changes lives or merely makes noise.

Getting off the platforms

The same refusal to be used drives the Guardian’s social-media policy. The paper no longer expects its journalists to be active on social platforms, a reversal of the era when it pushed them on. Viner’s reasoning is partly about judgment: once a reporter is composing for an audience of critics, “you start hearing voices in your head telling you should have done it that way, and then you make bad decisions.” The Guardian has left X altogether, on the grounds that staying “just feeds a far-right algorithm to make somebody else rich and change politics.” No individual is compelled to follow, but the direction is set. She has also raised spending on physical protection for reporters, citing the number of media workers killed in Gaza as evidence that the press vest no longer offers the cover the industry once assumed, “a sign of the times.”

The model that holds

Behind all of it sits the Scott Trust. No proprietor extracting cash or influence, no shareholders pressing for margin. Viner is blunt that this is the best arrangement an editor could ask for and offers two illustrations of what it buys. The first was a libel case the paper fought for six weeks in the High Court and won comprehensively, with the trust’s full backing, a result she believes has measurably improved the climate for investigative reporting in Britain, where the law tilts towards the claimant. The second was the reader-revenue model itself. When the Guardian began asking readers in 2016 to pay for journalism they could previously get for free, the idea was ridiculed across the British press and within the building. The trust backed it, and readers grasped it faster than the critics had.

The figures Viner cited make the case. Recurring reader revenue has gone from nothing a decade ago to roughly £125m last year across print and digital, she said, before one-off contributions; the digital share alone rose 22% to £107m in the year to March 2025. The model has also gone global: by her account, 83% of the Guardian’s revenue now comes from outside the UK, almost none of which existed a decade ago. What makes the model durable, in her account, is that it is not a transaction. “You’re not a consumer in the traditional sense; you are more part of our community. It’s not a transaction, it’s a choice.” She claims this makes it more resilient than a paywall, while conceding the obvious limit: the share of regular readers who pay stays small. The current work is adding transactional on ramps for people who prefer a straight exchange: the Feast recipe app, the Guardian Weekly, the news app, while keeping the website free to all. That last point she frames as a matter of mission. With paid news becoming the default, a free open site is a genuine differentiator.

The one fear she names

Viner is not against the technology itself and offered a concrete case where it earned its place. Working with University College London, the Guardian analysed a century of rhetoric on immigration in the British Parliament and found that the language of the past five years had been more aggressive and more anti-immigration than at almost any point in the preceding hundred. The study, she said, would not have been possible without AI. The journalists most excited by the tools, she added, tend to be the ones who push the rest of the newsroom forward.

For all that, Viner named a single threat without hedging. AI’s capacity to sever the link between the Guardian and its readers, the disintermediation AG Sulzberger had warned about in the congress keynote, is what worries her most. Everything the paper has built rests on a direct relationship with its audience. An answer engine that satisfies the reader before they reach the site dissolves that relationship at the root. Her response is to lean harder into the channels she controls and the community she has spent a decade cultivating.

Here the two newsrooms part company. Where Sulzberger opened his keynote with a call to arms, Viner’s posture is quieter and more collective. On litigation and licensing, she was deliberately vague. The Guardian is party to some class actions, she confirmed, though not at his scale, and it has struck deals to secure compensation for its work. “You do different things at different moments with different organisations.”

The collective answer is SPUR, the Standards for Publisher Usage Rights coalition. Launched in February by the Guardian alongside the BBC, the Financial Times, Sky News and the Telegraph, it is a non-profit set up to build shared, machine-readable standards and licensing frameworks for how AI systems use journalism. The aim is to let developers reach reliable reporting through paid-for channels while publishers keep practical control of their content and are paid fairly for it. Its founders have written to publishers and broadcasters worldwide inviting them to join, and Viner expected an imminent announcement of further global members. This is the publisher-built version of the value exchange Ørskov had described, where Sulzberger turned to litigation, Viner has turned to the coalition. It arrives as the UK government weighs its own position on AI and copyright. Viner was sharper on that: “Governments around the world seem so desperate for growth that they seem to think the words AI equals growth.” The creative industries, she suggested, keep getting left out of that sum.

She was strikingly un-snobbish about where audiences now find news. Influencers, she argued, have done what newsrooms have struggled with: built close relationships with their audiences and learned their platforms cold. The opening she sees is to fuse that with editorial standards: a Guardian news influencer whose information readers can trust, delivered in a form that fits the platform. The format mix is already moving: the Guardian publishes two novels’ worth of words every 24 hours, a proportion she admitted is “out of whack,” strong in audio, decent in video, with room to do far more for younger audiences.

Asked for one piece of advice for editors, Viner gave the line that doubles as the Guardian’s entire strategy: lean into who you are. “The values of who we are is what differentiates us in this moment.” For a small publisher, she added, a community built around a clear identity is a sound basis for a business.

It is good advice. The Guardian’s defences, the trust, the reader community, the editorial discipline, the open site, were built against the threats of the past fifteen years: proprietorial meddling, the collapse of display advertising, the erosion of public trust. Against those they are formidable. Whether a community survives when AI bricks up the front doors the audience walked through, search and social and the open web, is the bet Viner is now placing the institution on. She knows it. With Ørskov’s precondition still unmet, the honest reading of the hour is that the best-defended newsroom in the language does not yet know the answer, and neither does anyone else.

A.G. Sulzberger does not often raise his voice. The chairman and publisher of the New York Times argues in the measured cadence of a man who expects to be quoted accurately. So when he took the stage for the opening keynote at the WAN-IFRA World News Media Congress in Marseille, where I was sitting in the auditorium, and accused the richest companies in history of an “original sin”, the phrasing was deliberate. The sin is theft: the appropriation of journalism, books, music and film to train artificial-intelligence systems, carried out, in his words, “without permission or compensation”, and at a scale no previous technology has attempted.

None of this is new from him. The New York Times has litigated a version of the argument against OpenAI, Microsoft and Perplexity for two and a half years, at a cost Sulzberger put above $20m. In Marseille the setting sharpened it. He had come to tell publishers from some 60 countries that their response to AI has been “too quiet, too passive, and too fragmented”, and that polite resignation now leads only to extinction. From where I sat, the hall heard him out in a sombre quiet. The New York Times is the great commercial success of the subscription era, with over 12 million paying subscribers, by far the largest number by some way, and a warning from the industry’s clearest winner is harder to dismiss.

What the engineers pay for, and what they take

Sulzberger’s primary argument was an accounting one. An AI model, he said, is built from four ingredients: the talent that designs the algorithms, the computing infrastructure of chips and data centres, the energy to run it, and what the industry blandly calls “data”. The first three are paid for, lavishly. No AI chief executive proposes that the best engineers work for nothing; pay packages run into the millions. None would steal chips from Nvidia or tap a power line. Investors are absorbing hundreds of billions in losses to build the plants.

The fourth ingredient is taken for free. And “data”, he noted, is a word that flatters the theft by making creative work sound like a commodity, lying around for collection. The justifications change with the occasion: innovation demands it, facts cannot be owned, licensing is too slow and too costly, fair use permits it, national security requires it lest China win the race. Each one, he argued, falls apart under scrutiny. A chatbot can recite facts only because it has absorbed the protected language and structure that carry them, and drafting a licence costs a great deal less than building a power plant or data centre. A country gains nothing in a technology contest by dismantling the intellectual-property regime that underwrites its own creative economy.

The six leading AI firms are worth some $11 trillion between them, by his reckoning, more than three times French GDP; private AI investment in the United States ran to roughly $350bn in 2025 and is climbing. Against that, the licensing deals struck with publishers’ amount, on the reported figures, to less than half of one per cent of the money flowing to the people who produced the training material. The money is plainly there; withholding it from creators is a choice, not a constraint.

The ingredient nobody wants to cost

If data is cheap, quality journalism is the part of it that AI firms most quietly prize. Sulzberger cited an admission from OpenAI that it would be “impossible” to train today’s leading models without copyrighted material, and an engineer’s blunter line that a model’s success is “determined by your data set, nothing else”: you are, in the old machine-learning saw, what you eat. Five of the ten most-used sources for training popular large language models are news publishers. In one major training set, he said, the Times was the single largest source of proprietary data, ahead of the Guardian and the Los Angeles Times. A Microsoft executive has conceded that premium content measurably improves a model’s answers.

That output is expensive to produce. Sulzberger said the Times published close to half a million pieces last year, in articles, photographs, video and audio, at a cost of more than $2bn, with correspondents in all 50 American states and 155 countries, more than 70 of its journalists in Ukraine. A chatbot values that material precisely because people reported, verified and edited it, people who can be sued, sacked or sent to a war zone. The AI industry has nonetheless maintained that it owes nothing for any of it. Meta was trained on a corpus of pirated books. Perplexity scraped sites that had explicitly refused it. OpenAI has lobbied Washington for legal immunity. Sulzberger reserved a pointed line for the firm with the cleanest reputation, telling the hall that “even Anthropic”, lauded for its ethical posture, had in his account been unwilling to pay for the journalism it uses.

Litigation is the obvious recourse and a poor one. The New York Times’s case has run for thirty months, and cost more than many newsrooms earn in a year. AI firms know that almost no one else can afford to follow it, which leaves the law to be tested by whoever has the deepest pockets.

Sulzberger makes the case as both plaintiff and dealmaker. The New York Times sued OpenAI and Microsoft in 2023 and has not settled; in 2025, it signed its first AI licence, with Amazon. OpenAI has separately struck paid deals with other publishers, among them News Corp and Axel Springer. The pattern complicates his arithmetic and confirms his warning at once: the marquee titles are being paid something, while the smaller newsrooms he was addressing are offered nothing. His quarrel is less that money never changes hands than that it does so on the buyer’s terms, for a fraction of the value, and after the work has already been taken. There is also a circularity to his position. Sulzberger calls the practice theft; but whether it is theft or fair use is precisely the question his own lawsuit exists to settle, and no court has ruled yet. He is, in effect, asserting the verdict while the trial is still running.

The click that no longer comes

The deeper threat lies in substitution rather than training. For two decades, the open web ran on a lopsided bargain: search and social platforms took a growing share of advertising but sent readers back to publishers, who could sell subscriptions or show ads against the traffic. Google set the terms and kept sending the clicks.

That exchange is being withdrawn. As Google answers more queries directly, the link that used to deliver a reader is, by his estimates, roughly 10 times harder to earn than it was a decade ago. Rival AI products are stingier still: one study put their referral rate at 96% below a Google search. The largest newspapers tracked by Comscore have lost more than 45% of their traffic on average over four years, with the slide expected to continue. Sulzberger quoted Microsoft’s own head of AI monetisation, conceding that the old value exchange “does not translate cleanly to an AI-first world”, then read out the launch copy for its AI search, which invites users to skip the links and “talk through whatever you’re curious about”: an honesty about the model that its own commercial behaviour does not extend to publishers.

The revenue picture behind all this is stark. Newspaper advertising income has fallen by around 80% over twenty years, according to his figures, while Meta alone now earns eight times what all newspapers on earth make from ads combined. Subscriptions were the lifeboat. A reader who can extract the same reporting free from a chatbot has little reason to pay for it, and around a third of AI scrapes, he said, breach explicit restrictions, paywalls among them. Licensing cheques and per-scrape micropayments will not, for most publishers, replace what is lost; the majority expect no meaningful AI income at all.

Confidently wrong

The case becomes harder to wave away when the products fail at the one task they claim to perform. Sulzberger cited European Broadcasting Union research findings that leading AI assistants misrepresented the news in close to half of all answers. When the activist Charlie Kirk was killed last year, Perplexity’s bot suggested the White House statement was fabricated and Grok insisted he was alive. Because the systems handle uncertainty poorly, they tend to be wrong with conviction and neither track nor correct their errors. Microsoft, he noted, had launched Copilot with a disclaimer telling users to rely on it at their own risk.

Then he turned to what comes next. Amazon Web Services reckons most online content is already machine-generated, perhaps nine in ten before long. Fake local-news sites already outnumber real ones, cheap to erect as the genuine article grows harder to sustain. Sulzberger borrowed Margaret Atwood’s image of the author “murdered by my replica”. The hollowing of original reporting starves publishers and corrodes the shared factual record on which a society agrees to disagree, and on which the chatbots themselves depend.

The list of demands

His prescriptions were practical and unusually concrete. Defend your copyright and litigate when you must. License, but only on terms that reflect sustainable value and leave you some say over how the work is used. Lobby legislators with a short common list: bots that identify themselves, transparency over how content is used, liability for defamatory output, copyright protections strengthened not weakened. Above all, act together, and with the music, film and book industries facing the same theft, rather than cutting deals one frightened newsroom at a time.

The second half of his advice turned inward. Adopt AI inside the newsroom aggressively and under human control, because the laggards will be flattened anyway. Build a direct relationship with readers, so the platforms cannot ration the audience. Recommit to original reporting, the investigations and eyewitness accounts, because it is the one product a chatbot cannot manufacture and the public cannot get elsewhere.

In the last great disruption, Sulzberger admitted, The New York Times swallowed Silicon Valley’s line that “information wants to be free”, forgetting that Stewart Brand’s original aphorism had a second clause: information also wants to be expensive, because it is valuable. He asked the room not to be naive twice.

He stopped short of an answer, because it is not his alone to give: either the value of journalism returns to the people who go out and find it, or it stays with the firms that take it. The wave already breaking, he warned, was the small one. The New York Times, with its lawyers and its licensing deals, will probably ride it out. Whether the rest of us in the hall can do the same was the question he left us holding.

Caroline Orange-Northey spent nine years at Amazon Ads. That is not a neutral starting point. Amazon built the performance advertising infrastructure that most of the industry still runs on: last-click attribution, conversion tracking, and direct response as the default measure of platform value. She understood it, sold it, and believed in it. Then she joined Pinterest.

Caroline is someone who came from the place that defined how advertising is measured. Her central argument, six months in her role as Managing Director, is that the measurement model is wrong.

Eighty Billion Searches and an Unbranded Majority

The scale of Pinterest’s search activity catches most people off guard. The platform processes 80 billion searches a month, 96% of which are unbranded. Users are arriving with clear commercial intent but without a fixed brand allegiance. For a marketer, that is as close to an open goal as digital advertising gets.

“There’s some alchemy in there,” Orange-Northey says, “because they actually have the opportunity to form brand preference before it’s set.”

Gen Z now accounts for over half the platform’s audience, a demographic shift she describes as one of the bigger surprises since she joined. The platform she had associated with aspirational browsing, home interiors and wedding planning, is ageing down faster than its external reputation suggests.

FOBO and the Cart Abandonment Problem

One of the more useful frameworks Orange-Northey brings to the intent debate is what Pinterest internally calls FOBO: the fear of better options. Research from the University of Cambridge suggests humans make approximately 35,000 decisions a day. Simultaneously, 70% of online shopping carts are abandoned before purchase.

Her argument: these are not separate phenomena.

“Consumers have a confidence problem. Pinterest quintessentially solves that: not with a tool, but the entire experience is curation. You’re literally whittling down your choices to ones that feel most like you, and then you can make a more confident purchase decision off the back of that.”

Caroline Orange-Northey, MD of Pinterest, UK & Ireland

The commercial intent on Pinterest, she argues, is structurally different from that of other platforms. “Attention is often fragmented, but intention is usually whole.” The problem is that the advertising industry has built a measurement model that can’t see exploratory intent at all.

The Measurement Gap

If that intent argument is compelling, the industry’s measurement infrastructure consistently works against it. Last-click attribution, still the dominant model for performance budgets, places all value on the final touchpoint before conversion. Pinterest, sitting upstream in the consideration phase, rarely gets the credit.

Orange-Northey cites an analysis suggesting that 95% of Pinterest’s impact is invisible to advertisers who look only at last-click data. Brands are almost certainly underinvesting in the platform while crediting other channels for sales that Pinterest helped generate.

The Omnilux case makes the point sharply. The beauty technology company switched from last-click to view-based attribution through Northbeam and saw a sevenfold increase in Pinterest-attributed conversions. The media spending had not changed. The measurement model had.

Visual Search at Scale

Visual search queries grew 44% year-on-year on Pinterest, a figure Orange-Northey uses to argue that brands need a fundamental rethink of how they present themselves: not a creative refresh, but a change in the underlying logic of their content approach.

Her other instruction is to maintain creative standards throughout the funnel. Pinterest’s internal research suggests brands routinely drop creative quality between upper-funnel awareness and lower-funnel conversion, a habit Orange-Northey identifies as both an industry problem and a Pinterest-specific missed opportunity.

On the competitive question, whether Google Lens or TikTok’s visual discovery represents a threat, Orange-Northey is measured. Visual search has always been core to Pinterest in a way that it is not for either competitor. “It’s always been in the DNA,” she says, and that capability advantage is compounded by something harder to replicate: a decade of data about what people find visually compelling and what those signals reveal about taste.

The Taste Graph: Targeting Built on Behaviour, Not Identity

That data is what Pinterest calls the Taste Graph, and most advertisers misunderstand its unique value.

It is an algorithm built on 80 billion signals (searches, saves, clicks, image interactions) and continuously updated to track how individual tastes evolve over time. Someone who saved coastal interiors six months ago and is now pinning maximalist colour palettes is not the same person they were, and the Taste Graph moves with them.

In an industry still working through the operational consequences of cookie deprecation, that distinction matters. Most of the targeting infrastructure advertisers rely on runs on one of two things: third-party behavioural data, which is increasingly restricted, or declared identity signals, which are static and easy to game. The Taste Graph is built from first-party, real-time behavioural data generated by users actively engaged in discovery, precisely when purchase intent is most legible.

For a performance media buyer, the practical implication is that Pinterest’s targeting layer does not require third-party cookies, does not depend on users accurately self-declaring their interests, and updates continuously rather than sitting as a fixed audience segment. A campaign targeting homeowners in the market for a kitchen renovation does not need to rely on cookie-based retargeting if the Taste Graph already knows, from first-party signals, that a user has been actively curating kitchen content for three months.

The signal Orange-Northey singles out as particularly underappreciated is saves, including ad saves. When a user saves an ad, they are flagging commercial intent at a moment of active consideration. That signal feeds back into the Taste Graph, sharpening subsequent targeting. It is the kind of closed-loop feedback that performance marketers build elaborate workarounds to approximate on other platforms.

Pinterest Performance+: The AI Layer That Runs on Top

The Taste Graph provides the intelligence. Pinterest Performance+, the AI-powered campaign management suite launched in Q2 2024, provides the infrastructure to act on it at scale.

Performance+ is not a single tool. It is an end-to-end campaign layer covering targeting, creative optimisation, bid management and outcome measurement, all underpinned by machine learning that draws on the Taste Graph’s behavioural data. The practical effect is that campaigns running through Performance+ can be set up in roughly half the time of standard Pinterest campaigns, while consistently delivering stronger results.

Performance+ now carries 30% of Pinterest campaigns. The headline performance figure is a 24% higher conversion lift versus standard campaigns, and Orange-Northey is direct about the scepticism that number attracts.

The case studies bear that out. Celebrity Cruises ran an always-on campaign using Performance+ creative refresh, persona-interest targeting, and keyword trending. It beat its cost-per-click target by 14% and doubled its close-up rate relative to the travel benchmark.

A Levi’s campaign ran Performance+ targeting against video completion and trend objectives, beating the benchmark by 27.4% on completion and lifting Pin awareness 2.6 points.

Levi’s has since extended the partnership into the Summer 2026 Trend Report, becoming the first UK brand to co-launch a trend report with Pinterest. The timing reflects what the platform is seeing in search: “denim shorts outfits women” is up 430% year on year, with “micro denim shorts outfits” (+453%) and “denim jorts outfit” (+330%) climbing alongside it. Levi’s has built its most comprehensive shorts range to date around that signal, anchored by the 501® Original Short and the longer, looser Baggy Dad Jort, and paired the launch with an “Off-Duty Summer Denim” edit and a Premiere Spotlight on Search.

What Performance+ is doing, in structural terms, is closing the gap between the quality of Pinterest’s underlying targeting data and advertisers’ ability to use it. The Taste Graph has existed in some form since the platform’s early years; the question was always whether the campaign tools were sophisticated enough to translate that data advantage into measurable outcomes. That translation layer is now in place, which is what makes the current period commercially significant for advertisers who have historically treated Pinterest as a supporting channel rather than a primary one.

The Levi’s Test

The toxicity debate around social media is a live commercial consideration for any platform with a large Gen Z audience. Orange-Northey’s position is that Pinterest doesn’t meaningfully belong in that category: users build boards for themselves, not for an audience, and the social performance mechanics that drive so much of the documented harm in adolescent social media use are largely absent.

The most compelling endorsement of that positioning did not come from a Pinterest press release. It came from a Levi’s 501 Day video in which Emma Chamberlain, one of the most commercially influential Gen Z voices online and with an audience in the tens of millions, sits down with Hailey Bieber over Chamberlain Coffee and, at around the four-minute mark, they start talking about Pinterest.

There’s no Pinterest representative in the room. They’re in a Levi’s studio, on a Levi’s production, promoting Levi’s jeans. What follows is an unscripted exchange about how obsessed they both are with Pinterest: how it is the happiest corner of the internet, a place apart from the noise and performance anxiety of every other platform they use. No prompting, no brand brief, no talking points.

Chamberlain and Bieber are two of the most commercially astute Gen Z voices online, people who have built careers understanding what content works, on which platform, for which audience. They know exactly what they are saying and where. And what they choose to say, in a Levi’s studio, is how much they enjoy using Pinterest.

That is not something money buys. It is something a platform earns over the years.

Emma Chamberlin and Hailey Bieber discuss Pinterest (4 minutes 30 into the interview)

The Gap Worth Closing

Pinterest is doing more for advertisers than its share of their media plans suggests. The Taste Graph provides first-party targeting infrastructure that most of the industry is still trying to build from scratch. Performance+ gives advertisers the operational tools to act on it. The measurement problem is real but solvable, as Omnilux demonstrated, and brands that have solved it consistently find the channel under credited in their attribution models.

The headline thesis is worth restating. The industry is measuring the wrong thing. Last-click attribution was never built to capture the contribution of platforms upstream of conversion, and Pinterest is the clearest case in point. Brands still running performance budgets on a last-click basis are not making rational choices about Pinterest. They’re making choices; a broken measurement model is making it for them.

Orange-Northey’s task is to make that case to an industry that has spent years treating Pinterest as a nice-to-have. She has the data, the case studies, and the experience to do it. She also now has Emma Chamberlain and Hailey Bieber.

Pinterest will be at Cannes Lions Festival this year from 22nd - 25th June. If you are attending and want to meet the Pinterest team register at Pinterest Cannes.

To respond to its report on young news audiences, Understanding Young News Audiences at a Time of Rapid Change (covered in a recent Media Stack article), the Reuters Institute convened four practitioners. The consensus on the problem was complete. The divergence on what to do about it was the substance.

Moderated by Mitali Mukherjee, the Reuters Institute’s director and one of the report’s authors, the panel featured Tristan Werkmeister, Reuters’ first social-first video reporter; Brian Hioe, founder of New Bloom in Taiwan; Muhammad Adnan Butt, chief executive of the Pakistan-based distributor Dot Republic Media; and Christian Esguerra, the Filipino journalist whose Facts First podcast has more than 550,000 YouTube subscribers and who appears at number 12 in the Reuters Institute’s ranking of Filipino news figures audiences pay attention to.

Reuters: trust as the asset, AI as the multiplier

Werkmeister’s framing was the simplest: trust is what Reuters has that creators don’t. The strategic conclusion is to put Reuters journalists themselves on camera, with the institutional advantages (beats, geographic footprint, brand) intact, rather than handing the work to producers who never spoke to the sources.

The interesting part is the tooling. Werkmeister has built a chatbot trained on Reuters’ stylebook, trust principles and editorial guidelines. It writes vertical-video scripts from scratch or gives feedback on drafts, down to knowing that Reuters spells “U.S.” with a period. The effect is speed: when news breaks, text reporters who have never produced social video can get a draft close to the eventual published version, and the journalist who sourced the story can front the camera. The editorial checks remain. The bottleneck has been engineered out: script production by people not trained in vertical formats is no longer where new video stalls.

Worth holding onto: Werkmeister said Reuters sees its strongest engagement metrics in younger demographics, but older ones are not far behind. Vertical video is a distribution strategy with a youth-skewed payoff, not a youth strategy.

Pakistan: rebuild from the pipeline up

Butt’s argument was the most uncomfortable one for legacy newsrooms. Dot Republic, which he said holds around 60% of Pakistan’s digital content distribution market and serves, by his own account, more than half a billion video views a year in the US alone, runs a newsroom whose average age is 22 to 23. In his framing, vertical is not a format, not 16/9 flipped to 9/16, but a different product: produced by different people and optimised by machine learning across the first 24 hours of a video’s life on each platform.

His parallel was music. Lo-fi, he noted, came out of nowhere and became one of the most consumed formats for 18-to-24-year-olds. Entertainment has already gone through this reinvention; news, he argues, is next.

For traditional broadcasters, the implication is direct: the youthful audience cannot be reached without a youthful newsroom and a re-engineered production pipeline. Most legacy outlets have neither. Treating short-form video as an additional output of the existing newsroom misses the point.

Taiwan: craft and platform promiscuity

Hioe’s answer was the most resource-constrained, and probably the most familiar to small publishers. New Bloom, founded in 2014, increasingly relies on reporters who carry their own personal brands. Its storytelling format opens with a recent incident, uses it as a window into a larger issue, then circles back. That is a craft response to news avoidance, not a platform one.

On distribution, Hioe described the position every small publisher will recognise be on whatever platform comes along, because you cannot predict which will work and which will collapse. New Bloom is on Facebook, X, Bluesky, Mastodon, Threads, and Instagram. Customising for each is a stretch with a small team; staying on a single platform is no longer viable.

Hioe’s point about trust geography is the one most likely to get missed. The aggregate trust gap between younger and older audiences in the Reuters data is nine percentage points across 18 markets, but in Taiwan, trust in mainstream media is low across the board. “If you don’t study in school, you’ll become a journalist,” he said, paraphrasing the local view of television news. Where mainstream trust is already weak, creators are not displacing a trusted source; they are filling a vacuum. The global prescription on offer, that young audience’s distrust mainstream media and creators are the alternative, doesn’t fit Taiwan, where the mainstream was never trusted to begin with.

The Philippines: become the creator

Esguerra’s case is the one most publishers need to think about. After more than 20 years in newspapers and broadcast media, including at ABS-CBN, he took his political journalism to YouTube and built a half-million-subscriber audience. His framing is that journalists carry ethical boundaries content creators do not, and that this is simultaneously a competitive disadvantage and the entire point.

Mukherjee asked the question every newsroom dread: what happens when the journalist’s personal brand becomes bigger than the masthead? Esguerra’s answer was straightforward. You cannot stop a journalist whose brand has overtaken the organisation from leaving. Collaboration is possible; retention, as newsrooms used to manage it, is not.

The cost is visible in the workflow. Esguerra runs a five-night-a-week live stream as essentially a team of one, leaning on AI to compensate for the absence of producers, editors and researchers. Burnout, Mukherjee noted, is what creators most consistently point to. Nobody on the call had a model for how these scales without breaking the journalist.

The shared thread

What unites the four answers is the editing pipeline. Werkmeister automates it with proprietary editorial standards; Butt rebuilds it around machine learning and a younger newsroom; Hioe rethinks it as a storytelling craft; and Esguerra is asking AI to compensate for not having one at all. None of them treats journalism as the variable. The variable is everything around it: how stories are produced, packaged, distributed, optimised and recovered when an algorithm shifts.

The previous article argued that publishers had spent a decade misdiagnosing a structural problem as an attitudinal one. The webinar suggests the diagnosis is now widely accepted. The harder question is whether legacy newsrooms are willing to make the operational changes that follow, or whether they will keep treating short-form video, AI workflows and creator partnerships as edge experiments while the core operation stays intact. The four practitioners on the call had each, in their own way, picked an answer; most publishers have not.

The UK advertising industry will arrive at Cannes Lions this June with a government-backed commercial mandate, not just a delegation.

UK Advertising, led by the Advertising Association as the official UK representative at the Festival, returns for a sixth year with an expanded growth mission designed to accelerate international expansion for British advertising and marketing services companies. The programme runs across Cannes Lions 2026, which takes place 22nd–26th June.

The mission is coordinated by the UK Advertising Exports Group (UKAEG), the industry body backed by the AA, APA, IPA and the Department for Business and Trade, representing more than 70 UK companies. Support this year comes from Channel 4, Marketreach, The Guardian, Sky Media, Tesco Media, Outernet and DBT.

The commercial logic is straightforward: UK advertising exports were worth nearly £18bn in 2025, and Cannes is the most efficient week in the year to push that figure higher.

“The UK’s advertising exports were worth nearly £18bn in 2025, demonstrating how the demand for UK advertising has never been greater.”

“As we head to Cannes Lions, we are ready to showcase the UK’s creative and technological expertise to the world, building on last year’s success with the UK in the top three most-awarded nations.”

Stephen Woodford, Chief Executive of the Advertising Association

That ranking matters commercially as well as reputationally. A top-three finish at Cannes signals to international buyers that British creative output can compete at the highest level, and UK agencies are increasingly leaning on it in new-business conversations.

The Programme

UK Advertising will be based at the Empower Café, a short walk from the Palais, for the duration of the Festival. The venue will serve as a dedicated hub for the delegation, hosting a programme of invitation-only roundtables across four target markets: the USA, Saudi Arabia, Nigeria and APAC. Each session is built around deal-making and partnership development rather than general networking, bringing senior UK executives into direct contact with decision-makers in those markets.

The week will also include thought leadership panels and the UK Advertising Global Reception, an invite-only gathering for senior industry figures. Previous missions have generated commercial opportunities in markets including Saudi Arabia, where UK creative companies have established a foothold ahead of major Saudi investment in domestic media and entertainment infrastructure.

Aisling Conlon, International Trade Director for UK Advertising at the AA, said the Festival is a moment to compete, not simply attend. “The UK continues to set the standard for creativity that delivers real commercial results, and Cannes gives our companies the opportunity to build international partnerships, open new doors and compete at the very highest level.”

“Cannes Lions is our global stage, a chance for UK advertising to show up loud and proud as one of the top three exporters on the planet.”

Ameeta Virk MBE, The Department for Business and Trade

Next Generation and Jury Representation

Beyond the trade programme, the AA is backing 12 UK creatives competing in the Young Lions competition, the Festival’s talent pipeline for early-career professionals. A further 31 British jurors have been selected to judge work across the Lions categories, a level of representation that carries weight with international clients assessing the credibility of UK creative expertise.

Before the Festival Starts

On 15 June, before the Festival opens, the AA will launch UK Advertising’s new industry campaign at the Outernet London. The event is designed to frame the UK’s presence at Cannes and set the tone for what the delegation intends to achieve on the Croisette.

Full details of the Cannes programme are available at www.ukadvertising.co. To find out how to get involved, contact Aisling Conlon, International Trade Director for UK Advertising, at Aisling.Conlon@adassoc.org.uk.

Publicis Groupe has agreed to acquire LiveRamp, a US data technology company, for $2.167 billion in cash. The deal, announced on Sunday, is the latest step in a deliberate strategy to own the data infrastructure that the advertising industry runs on.

The price of $38.50 per share, a 29.8% premium to Friday’s closing price, values LiveRamp at $2.546 billion including net cash. It follows Publicis’s $4.4 billion purchase of Epsilon in 2019 and its acquisition of Lotame in March 2025. Sadoun, who has been spending around $1 billion a year on acquisitions, described the LiveRamp deal as the equivalent of “two years of bolt-on acquisitions.”

What LiveRamp does

LiveRamp solves a problem every large advertiser has. Their customer data sits in multiple places: CRM systems, websites, apps, loyalty schemes, etc. None of it talks to each other or to the outside world in a consistent way. LiveRamp acts as the link between them. It takes that fragmented data, strips out personally identifiable information, and replaces it with a common identifier, called RampID, that can be used to match audiences, run campaigns, and measure results across publishers and platforms without anyone handing over raw customer records.

Its other major product, Safe Haven, is a so-called data clean room. Think of it as a secure room where two companies, say a retailer and a publisher, can each bring their customer data, run analysis on the combined set, and take away insights without either party ever seeing the other’s underlying records. It allows collaboration that would otherwise be impossible under privacy law.

LiveRamp’s infrastructure is embedded across more than 25,000 publisher domains and integrated with 500-plus technology partners in 14 markets. It has become, effectively, shared infrastructure for the advertising industry.

Why Publicis wants it

Publicis already owns Epsilon, which holds verified records on 2.3 billion consumers, and Lotame, which adds a further database of consumer identities. LiveRamp adds the clean room infrastructure and the connections that let all this work across the web.

The combined stack lets Publicis offer clients something its rivals currently cannot: a single integrated system for collecting a brand’s own customer data, connecting it, and using it at scale, with AI tools running on top.

Publicis CEO Arthur Sadoun framed the deal as a play for what he called “data co-creation”: pooling data securely with partners to feed AI tools. “By building the future of data co-creation, we’re empowering our clients to generate new, exclusive and proprietary data, to build the smartest, most differentiated AI agents,” he said in a statement. LiveRamp provides the secure infrastructure to do it.

For Publicis, this is also a commercial expansion. It is no longer selling media buying and creative services alone. It is selling data infrastructure, identity matching, and AI capability. These are higher-margin businesses, and ones clients find harder to walk away from.

What it means for publishers

Publishers have trusted LiveRamp precisely because it was independent. When a UK publisher used LiveRamp to transact with an agency, including Publicis, they were using a neutral third party with no stake in the outcome. That is no longer the case.

Publishers now face a version of the same question that has defined their relationship with Google and Facebook/Meta for a decade: how much do you want to depend on infrastructure owned by a company you’re also negotiating with? A publisher using LiveRamp’s clean room to collaborate with a Publicis-managed brand is now doing so on a platform owned by the agency on the other side of the table.

Publicis has promised LiveRamp will continue to operate as a neutral platform that works with everyone. That promise may well be genuine. But it is hard to stay neutral when the company that owns you also has commercial reasons to favour its own clients.

The bigger picture

Analysts were sceptical when Publicis bought Epsilon in 2019. The bet paid off as the industry shifted towards data-driven advertising. Sadoun is making the same bet again: own the data, and you own the client.

The question for the rest of the industry, from rival holding companies and publishers to independent technology firms, is whether they can build a credible alternative before that bet pays out again.

The Insider, The Economist’s twice-weekly video show, took six months and 40 pilot episodes to build. It put named journalists on camera for the first time in more than a century, reached 90% brand awareness across the subscriber base within months of launch, and is, by the publication’s own admission, only one half of the answer to a harder question: how to acquire new readers in an environment where Google is sending fewer of them.

​That was the question circling Newsrewired 2026, the biannual journalism conference in London. No theme was declared, but one emerged anyway. Google search is declining as a referral source, social platforms are less reliable than they were five years ago, and AI is absorbing queries that used to produce clicks. The discovery mechanisms publishers spent the last decade building around are quietly eroding, and not everyone has a plan for what comes next.

​Across six sessions, three rough postures emerged. Build new products inside the legacy structure (The Economist’s Luke Bradley-Jones, The Times’s Tom Calver). Go where the audiences already are, on the platforms’ terms (Reddit’s Oliver Wrighton and Rachel Duffy from The Times). Or build something small enough to own outright, audience and all (Jim Waterson, Daniel Ionescu, Flight Story’s Grace Miller). The Reuters Institute’s Nic Newman, presenting research on news podcasting, sat across all three.

The Economist breaks its 100-year anonymity.

Luke Bradley-Jones, President of The Economist Group, opened by describing a product that would have been unthinkable for most of the publication’s history. The Insider is a twice-weekly video show in which Economist journalists appear on screen, debate, and make visible the internal mechanics behind the publication’s collective view of the world.

For more than a century, The Economist ran no bylines. Anonymity was part of the brand. The Insider inverts that. “What consumers wanted,” said Bradley-Jones, “was to get closer to our journalists, to see behind the curtain of the Economist, the workflows and how we really debated the discussions that happen inside.”

​The product took six months to develop before launch. The team produced around 40 pilots, each reviewed by about 20 people in the newsroom each week. Of a staff of 300, around 10 to 15 journalists appear each week. Shows run at 45 minutes. Completion rates are 30 to 40%, high for that length. The show built quickly across the subscriber base, reaching over 90% brand awareness within months of launch.

​The Insider is primarily a retention play, designed to deepen relationships with existing subscribers in a market where acquiring a new one costs significantly more than keeping an old one. But retention is only one side of the problem, and acquisition is the harder, more urgent one. As Google search declines and AI products absorb more of the discovery journey, the routes to new readers narrow.

​The response is a new streaming product, video and audio only, being tested in Canada this summer before a potential global rollout. The format is a daily menu: a 5- to 10-minute morning briefing, a 20-minute analysis segment, The Insider’s deeper dives, and games. Pricing will sit well below the core subscription, with a free tier for pre-subscribers in development. Bradley-Jones cited the New York Times as the model: build habitual engagement first, then ask for the conversion.

​When it comes to AI adoption, The Economist is moving faster than most. Two board members with deep AI backgrounds, one from Google DeepMind and the other Amazon’s former international head, have set the pace. Journalists are using code to automate research workflows, pull data from Snowflake, and generate charts. Agentic AI has taken over increasingly large chunks of product development. A CarPlay app that had sat on the roadmap for years was shipped in five weeks.

Asked what keeps him awake, Bradley-Jones said: “How we continue to acquire customers in an environment where good-enough AI products are increasingly mounting up.”

Kill the guesswork

Grace Miller, Head of Failure and Experimentation at FlightStory, the media agency behind a YouTube channel that has grown from 100,000 to 16 million subscribers, gave a session on experimentation culture.

Most teams, in her view, make decisions on instinct; FlightStory does not. “Anytime someone says vibes, it makes me cringe,” she said. Before any campaign goes live, the agency runs structured tests on content, thumbnails and trailers: what Miller calls killing the guesswork.

The arithmetic is harsh: seven in ten experiments are inconclusive; of the remaining three, two fail and one succeeds. FlightStory’s trailer team spent over a year in inconclusive testing before results began to come in. The argument for continuing was simple: the failures were teaching them faster than any optimised repetition would.

For larger organisations where public experimentation carries brand risk, the practical routes are platform tools that don’t permanently surface test content, synthetic AI audiences for pre-live testing, and incremental budget asks of 10 to 20% to test an idea rather than committing in full. The pitch to sceptical stakeholders is not a bet but a structured question with a planned answer, whatever that answer turns out to be.

Reddit: the slow burn

Reddit now has 125 billion monthly views and 490 million weekly active users, according to the company’s most recent figures. For publishers, it has an unusual property: a substantial share of its users are not active on other major platforms. That is material unduplicated reach.

Oliver Wrighton, who works with publishers and brands on organic Reddit strategy, and Rachel Duffy, who is Head of Community at The Times, gave a joint session on what works. The answer was patience. Reddit is community-led, moderated by volunteers who take the character of their subreddits seriously. Publishers who arrive treating it as a distribution channel tend to get rejected and quit. Those that build moderator relationships first, and establish editorial authority rather than traffic intent, eventually see the return.

​“Publishers have given up because it kind of feels like you’re getting nowhere,” said Duffy. “But it’s short-term pain, long-term gain. When you reach out not even with the intent of sharing content, just to introduce yourself, you’re able to go back to those relationships when you do have something really valuable.”

The Telegraph’s Ukraine coverage demonstrated the approach. Frontline journalism gave the publication something to offer communities as a credible authority rather than a traffic-seeker. Around the 2024 US election.

Reddit Pro, launched this year and free for publishers, provides link tracking, RSS feed sync and community recommendations. Wrighton’s caveat about RSS was clear: bulk posting is a fast route to spam flags. The tool is for discovery, not automation.

Times Data: Is Britain getting better or worse?

Tom Calver, data editor of The Times and Sunday Times, presented Times Data, a live dashboard of 40 indicators tracking whether Britain is getting better or worse, updated daily. The session was not, in fairness, about audience reach; it was about how to do data journalism well in 2026. But the underlying logic touches the same nerve: a habitual, fixed-location product designed to bring readers back without depending on platforms to deliver them.

The ambition was breadth combined with ruthless simplicity. Heavy economic indicators sit alongside the price of a pint, restaurant meals and government approval ratings. Each gets a single line chart and a directional signal, designed to be interpreted in seconds. Every card carries a source note with reproducible methodology, and the page updates regardless of whether the news is good for the government.

Colour-coding proved more politically fraught than expected. Inflation: red. GDP growth: green. House prices? The answer depends on whether you own one. Small boat crossings were harder still. A third grey category was introduced for a handful of measures where a binary verdict would itself have been a political act.

Calver used Claude to brainstorm indicators and locate obscure data sources (it was particularly useful for finding hidden data on the Bank of England website) but wrote the aggregation scripts himself. “The moment it needed to carry the Times name, I needed to trust it, and for me, AI wasn’t there yet.” Scripts run but require a human to push to the GitHub repo before publication, preserving an editorial check.

The dashboard now sits as a rotating data strip at the top of The Times UK homepage. Next steps include an API feed and CMS integration, so journalists can drop live charts into articles with a single command.

Video podcasting: a dual-format medium

Nic Newman of the Reuters Institute presented findings from the Institute’s third major report on news podcasting. Video is not replacing audio so much as expanding the medium to new audiences and contexts, and the production and commissioning decisions that follow are significant.

In the US, most podcast users now consume both audio and video formats each week, twice as many as consume audio only. YouTube dominates as a discovery platform. Spotify has been pushing producers towards video conversion. Apple joined a few weeks before the conference, enabling seamless toggling between audio and video within the interface.

The growth is concentrated in conversational formats: personality-led shows that are cheaper to make and feed algorithmic demand for content. Daily news podcasts remain largely audio and brand-led. Documentary series, expensive to produce, are being squeezed. Publishers are converting their chat formats to video first and leaving the others untouched.

The pull is as commercial as it is editorial. Video opens discovery and deepens the parasocial relationship between audience and host. Online video advertising is substantially larger than audio, and successful podcast studios are now licensing content to television; Goalhanger’s Gary Lineker has a deal with Netflix. According to reporting in The Times his company will earn over £14million for The Rest is Football daily show during the World Cup.

Publishers that treat podcasting as a funnel add-on are structurally disadvantaged relative to models like Goalhanger, which has built around talent, community and fandom, with multimodal outputs (audio, video, newsletters and live events) and multiple revenue streams. The Guardian and the New York Times are both moving towards studio-based models. The add-on approach is becoming harder to sustain.

The indie argument

Jim Waterson, founder of London Centric and former media editor of the Guardian, opened with a project he ran in 2019 and repeated in 2024: recording users’ screens during a general election campaign to see what they consumed. The 2019 data was mainstream-heavy. He called the 2024 data “terrifying”: almost entirely creators, influencers and short-form video, with mainstream sources barely registering.

That prompted him to leave and build London Centric, a London-focused, investigation-led operation with around 40,000 free subscribers and 5,000 paying subscribers. It is profitable, and deliberately AI-free. “The thing that really works is when there’s a human delivering original stuff people can’t get elsewhere. The stuff that flops is when it’s stories or content you could already be getting somewhere.”

Daniel Ionescu, founder of the Millennial Masters podcast and newsletter, started from a different place. His first media venture was local news in Lincoln, a city of 100,000 people where, he argued, sustainable independent journalism is essentially impossible regardless of operational efficiency. Millennial Masters is lean and geographically unbound. Content is split into three types: audience-acquisition pieces, personal-interest pieces, and long-form podcast episodes. Conversion to paying subscribers tends to happen between the second and third type.

​Unlike Waterson, Ionescu has AI built into his production stack from day one: ChatGPT for editing notes, Canva AI for production. On equipment, his advice was direct: half the best podcasts in the world are recorded on an iPhone, and a mid priced microphone can do the same job as a £200 one.

The recurring question

The day kept circling one question that none of the speakers quite stated outright: how do you build a direct relationship with an audience when the platforms that used to do it for you are increasingly unreliable? The answers on offer differed in scale but not in substance. The Economist is engineering one at the level of a global subscription business, with streaming products, video shows and an AI-accelerated product team. Waterson is building one on his own, with 5,000 paying subscribers and a deliberate refusal to use any of it. What none of them are doing is waiting for the old discovery routes to come back.

​

Meta’s advertising business is on course to generate $240bn this year, a 22.3% increase on 2025, according to new research from WARC Media. It confirms Meta’s dominance in social advertising and sharpens a question the company has avoided answering: what happens when a business funds everything from a single revenue stream?

In 2025, Meta’s ad revenue grew 22% to $196bn. Prior to 2023, it had been losing ground: its share of global social spend was declining and annual growth lagged the wider market. The turnaround came through AI. Rather than loading more ads, Meta invested in automated campaign tools and a unified AI architecture that optimised for advertiser conversions without degrading the user experience. The flywheel started turning faster.

Facebook accounts for 60% of Meta’s ad revenue, Instagram 40%. Both are growing at double-digit rates. Kantar analysis suggests the allocation doesn’t reflect the returns: the average campaign puts just 4% of budget into Instagram and 5% into Facebook, yet both platforms over-deliver on brand awareness, association and purchase intent relative to spend. More than 40% of global marketers rank Instagram among the top four platforms for attention.

The CapEx Problem

Meta has announced a $125bn–$145bn increase in annual AI capital expenditure, funded almost entirely through advertising. Unlike Alphabet or Amazon, Meta has no meaningful alternative revenue base. Investors balked: the stock fell 10% after the announcement.

“Meta’s flywheel is spinning faster than ever. The company’s AI-driven automation is transforming how brands connect with audiences, driving rapid growth in advertising spend with Facebook and Instagram. This is enabling further record-breaking levels of investment in AI innovation.”

“Yet investors appear concerned that the flywheel is at risk of spinning out of control, in light of plateauing user growth and mounting pressure to better monetise existing audiences.”

Alex Brownsell, Head of Content at WARC Media

& co-author of the report

The user growth figures support that concern. Meta reports 3.5 billion daily active users across its family of apps, but Q1 produced the company’s first-ever decline in total DAUs, the result of restrictions in Russia and Iran. Latin America and Sub-Saharan Africa lead on high-net-worth engagement across Facebook and Instagram but generate far less revenue per user than North America or Europe. Meta is working on closing that gap; the rollout of Threads ads in Brazil is one signal of intent.

Campaign Performance

Meta’s Q4 2025 model rollout drove a 24% increase in incremental conversions through improved attribution. Analysis by Fospha found cost-per-purchase has improved 4.5% year-on-year. Brands using Meta’s Advantage+ tool are seeing 41% higher blended ROAS and 17% lower new customer acquisition costs compared with manual campaign management.

Creator-linked Partnership Ads are landing: 71% of consumers report making a purchase within days of seeing creator content on Meta’s platforms.

Short-form video is pulling both platforms in the same direction. Reels accounts for 45% of all engagement on Instagram; on Facebook, the figure is 29%, with time spent watching video up 8% quarter-on-quarter.

The Structural Question

The performance case for Meta is measurable and, for most advertisers, still persuasive. The strategic problem is bigger. WARC forecasts growth decelerating to 12.1% in 2027, still substantial but a marked slowdown. Whether advertising alone can sustain the AI investment Meta needs to stay competitive is a question its results have not had to answer. Investors are asking it anyway.

Nine international publishers, Le Monde, El País, Politiken, Corriere della Sera and De Standaard among them, now sell subscription bundles that include access to The New York Times alongside their own journalism, according to Nieman Lab. The arrangement suits both sides. Local publishers add internationally recognised reporting to their value proposition without the cost of producing it. The New York Times Company extends its subscriber reach into markets where direct acquisition is slower and more expensive.

Two of those nine publishers, The Irish Times and De Standaard, sit inside Mediahuis, whose “essential subscription” is the clearest worked example of the strategy. As I wrote for Audiencers, Mediahuis has built a bundle that wraps journalism, e-books, masterclasses, a hiking and cycling app, and access to The New York Times into a single subscription. The early data is compelling: activated users churn 26.5% less, and the bundle is engineered to push subscribers up from the basic tier to a premium one priced at roughly double. The New York Times partnership is one component of that bundle, not a curiosity bolted on top.

That arrangement is the clearest signal of where the market is heading. FIPP’s latest Subscription Report counts 53 million digital-only subscriptions across 206 titles, and the headline 24.7% growth conceals a sharp split. Publishers that have built direct audience relationships, bundled product ecosystems and cross-title access are accelerating. Those still dependent on single-title subscriptions and search-driven acquisition are losing ground, in some cases losing subscribers outright.

The rapid expansion of AI into search has sharpened the divide. Referral traffic that publishers relied on for audience acquisition and advertising revenue is compressing. Publishers that were renting their audiences from Google are finding the lease harder to renew. The response from the strongest performers is consistent: own the relationship, deepen the product, reduce dependence on external distribution.

No company illustrates that logic more clearly than the New York Times Company, which is why the cross-publisher bundles matter: they are central to the strategy, not a side experiment.

The New York Times is no longer just competing. It’s embedding.

The New York Times Company now reports 12.21 million digital-only subscribers and stopped reporting title-level numbers this year. The second decision reframes what kind of business the company believes it is in. The New York Times is no longer a newspaper with digital products attached. It is a subscription business that happens to include a newspaper, alongside Games, Cooking, Wirecutter and The Athletic, bundled into a single proposition designed to make cancellation feel like losing several things at once.

The cross-publisher bundles extend the same logic outward. As AI search compresses referral traffic, the value of a subscription that contains multiple compelling reasons to stay rises. Bundles reduce churn inside a single portfolio. Cross-publisher bundles spread that logic across the whole market and position the New York Times Company as the connective tissue of other publishers’ subscription strategies, not just its own.

The Nordics settled this argument years ago.

While publishers in the US and UK are still debating the merits of bundling, Norway’s Amedia has been running the experiment since 2020, and the results are stark.

Its +Alt bundle, which gives subscribers access to all 127 of Amedia’s local newspapers plus a sports streaming service, now has 433,000 digital-only subscribers. Churn on the bundle runs at 0.7%. Churn on a single-title subscription runs at 16.4%. That difference translates into a subscriber lifetime value 26 times higher on the bundled product.

Norway as a subscription market grew 55.56% in the period, a function of a deliberate structural choice. Publishers there decided early that local relevance and shared digital infrastructure were not in tension; they were complementary.

In Sweden, Bonnier News reached the same conclusion. Its +Allt bundle, combining national titles (Dagens Nyheter, Expressen), more than 40 local newspapers and magazine content into a single subscription, reached nearly 500,000 digital subscribers by the end of 2025, up from around 270,000 in mid-2024. Including print subscribers with bundled access, total reach is over 1.1 million.

Germany and France are also moving. Germany’s subscription market grew 35.83% in the period, the fastest of any major European market. France grew 31.21%. Both reflect the same underlying pattern: publishers deepening their product propositions and reducing their exposure to external distribution.

The UK is following. FIPP’s data shows the market up 22.61%, with Newsquest, DC Thomson and Mail+ among the standout performers on percentage growth. Mail+ grew 177.8% in the period, from a low base, but the direction is clear.

AI search didn’t break the business model. It revealed which models were already broken.

The useful question is not whether AI search hurts publishers. It is which publishers. The FIPP data, read against recent industry reporting, makes the divide specific.

The content most exposed to AI overviews is the content most easily summarised in a sentence: car specifications, factual service journalism, commodity news. Bauer Media Group has said as much directly. When the overview answers the question, the click never happens, and the publisher’s role in the value chain ends with the model’s training set.

The content least exposed is reporting that does not compress. Analysis, investigation, distinctive voice and depth all carry information that survives summarisation, because the value is in the framing rather than the facts. The Telegraph’s SEO director has described search as being in “managed decline” rather than collapse, on the basis that subscribers come back for reporting summaries cannot replicate. The framing fits the data.

The more interesting signal sits in the other direction. The Washington Post reports that readers arriving via large language models spend longer on site and convert to subscriptions at higher rates than traditional search visitors. The volumes are small, but the conversion quality matters: AI search may be filtering out low-intent traffic that was inflating publisher metrics without ever turning into subscribers.

As Colin Nagy argued recently in Monocle, AI hasn’t created a new problem for publishing; it has clarified an old one. Publishers built around search traffic were never really building audiences; they were borrowing them, and the loan is now being called in. The publishers losing ground are not victims of a new technology. They are tenants whose landlord has changed the terms.

US local publishers are paying it.

Lee Enterprises reported that its digital subscriber base fell from 728,000 in March 2025 to 609,000 by December: a loss of 119,000 in nine months. The USA Today Co, formerly Gannett, saw a 26% decline year-on-year, dropping from 1.95 million subscribers to 1.45 million.

Both companies attribute the declines to deliberate strategies focused on price optimisation and improving average revenue per user. That may be true. It also describes companies managing contraction rather than building growth.

The sharpest contrast sits within the same corporate structure. Newsquest, the USA Today Co’s UK division, grew digital subscribers 35% year-on-year to 139,000 over the same period. Same parent company, opposite trajectory.

What the winners have in common.

Bloomberg grew subscription revenue 10% in 2025, taking paying subscribers beyond 707,000. It now produces 800 hours of video each month. Subscriber-only newsletters grew 20% year-on-year. Podcast downloads rose 26%. The pattern is consistent: more reasons to open the product, more often.

Substack, with five million paying subscribers, has built its growth engine around internal recommendations that move readers between newsletters inside the platform. It is less a publishing platform than a subscription marketplace, one that reduces its dependence on external traffic by creating its own internal referral economy.

Immediate Media, with 767,000 digital app subscriptions now accounting for 64% of group subscriptions, is using AI to identify which audiences are most likely to subscribe and when engagement signals are strongest. The objective is habitual use and longer subscriber lifetime value.

The thread connecting all of them is the same one running through the Nordic data, the NYT bundle strategy, and the divergence between US and European local publishers. Subscriptions built on borrowed traffic are fragile in ways that subscriptions built on daily habit, multiple products and direct relationships are not.

Growth in the market is concentrating around publishers who solved for retention before they needed to.

INMA World Congress • Agentic AI Seminar • Hotel Adlon Kempinski, Berlin • 8 May 2026

Michael Smith, CTO of Darwin CX, opened his session with a line he had borrowed from Brian Alvey, CTO of WordPress VIP: publishers, Alvey had told him, are focused on “doing just enough AI to stay behind.” It is what Smith hears from publishers around the world.

It was a useful provocation to open the seminar at the Hotel Adlon, part of the INMA World Congress in Berlin. The four speakers who followed session chair Jodie Hopperton, INMA’s Product and Tech Initiative Lead, represented a spectrum from careful strategic positioning to radical operational deployment. What connected them was a lack of patience for abstraction. The technology is no longer theoretical. The questions are organisational.

Holtzbrinck: getting into the room

The first question for any publisher serious about AI is where to learn from. Holtzbrinck’s answer was to open an office in San Francisco. Katharina Neubert and Claire Furino presented Holtzbrinck’s AI hub in San Francisco, now nine months old, staffed by two people and funded at less than 1% of group revenues. The point Neubert was explicit about: European publishers typically encounter AI companies through their sales functions. The people who build the products and the models are based in the Bay Area, and they are not the same people.

The hub operates across three layers:

• Signal-gathering and synthesis for group leadership;

• Knowledge transfer through an internal residency programme;

• Direct investment, with six companies backed since launch. The residency is deliberately non-technical.

Any Holtzbrinck employee can apply for a 3-month residency in San Francisco to develop a project from idea to MVP, typically in 11 weeks. Twenty-four to thirty projects have gone through the programme, ranging from a neuro-symbolic control layer sitting between MCP outputs and model results, to a book club app designed to surface audience signals that inform commissioning decisions. The latter was built by a staff member who, by her own account, had not worked with AI before joining the residency. Furino, who leads the technical side of the hub, described building the entire management infrastructure (application tracking, C-suite reporting across all seven business units, and an agentic layer for processing inbound requests) in 30 days with two people.

“Within 30 days, Katharina and I, with the help of artificial intelligence, have been able to build infrastructure that five years ago would have required a whole team of software engineers, if not a vendor contract for three or four months.”

Claire Furino, Director of AI Hub Programmes

Key takeaways from the Holtzbrinck session:

• Physical presence in San Francisco yields access to the people building AI products, not just selling them.

• A residency open to non-technical staff accelerates cultural change alongside technical capability; it also changes how individuals think about what is possible.

• Shared tracking infrastructure converts individual project outcomes into institutional knowledge across the whole group.

• Investment positions the hub to generate returns that offset its own cost.

Darwin CX: the operational case

Smith broke AI adoption into three waves and argued publishers needed to read each one differently. The first wave (generative AI for content) was largely paused for reputational risk. The second (transcription, source detection, and editorial review tools) is already in use across major publishers, whether it has been officially sanctioned. The third, agents running business operations, is where the commercial case is clearest, and where most publishers have barely started.

His demonstration centred on marketing workflow automation. The traditional model requires a marketer to raise tickets with the engineering, data, legal, and channel teams before a campaign deploys. An agentic workflow replaces this with a natural-language prompt: describe the campaign goal, and specialist agents execute in parallel across segmentation, compliance checking, channel selection, and deployment. The marketer reviews a summary screen and approves. Darwin CX reports the following from a client deployment it described as live and in production:

• 100% reduction in cross-system navigation.

• 97.5% fewer clicks.

• 14–15x faster campaign deployment.

• Marketer’s time split shifted from 8 hours strategy / 32 hours execution to 30 hours strategy / 10 hours execution per week.

“The masthead is your product. Everything else is plumbing. It’s necessary. But on the business side, AI agents can run and do work for you.”

Michael Smith, CTO of Darwin CX

The implication for publishers is that the right place to start is not editorial. Segmentation, paywall configuration, campaign deployment, lifecycle marketing: these are operational workflows where agents can move now, and the risk to editorial credibility is zero.

Hindustan Times: What full deployment looks like

Before the case study, Verma laid out why standing still is no longer an option. Audience trust is deteriorating: 40% of global readers now avoid news sometimes or often, up from 9% a few years ago. Print and digital revenues are declining across the board. And AI is redistributing traffic in ways publishers have limited ability to influence. When Google rolled out changes to Discover in September 2024, Hindustan Times was among those hit hard. Publishers dependent on the platform are reporting referral losses of between 20% and 70%.

The logic follows directly. If publishers are losing traffic to AI-driven search, the answer cannot be to run more pilots. The speed at which AI can compound internal capability is the only lever publishers directly control.

Hindustan Times Digital has moved past experimentation and is measuring outcomes:

• 97.4% of code is now generated by agents.

• Monthly code output has risen from 300,000 lines to 1.7 million with the same or smaller developer headcount.

• Engineering delivery timelines have improved by 60–70%.

• Sales cycle length has been cut by 40% through AI-generated pitch decks, produced in hours rather than five to seven days.

• Repeat client rates in advertising have improved by 20% through automated post-campaign insight reports sent directly to clients.

• 12 AI tools are currently in production; 10 more from a recent hackathon are being deployed this month.

The organisational model behind these numbers is as instructive as the technology itself. Hindustan Times identified AI evangelists in every department: not technical staff, but functional domain experts who act as translators between capability and application. AI adoption is now linked directly to performance appraisals: staff who are not using AI will not meet expectations. A public leaderboard gamifies adoption, with the most active users recognised at monthly all-hands meetings.

“The question is not if we adopt it. It’s how fast we can execute.”

The governance structure is deliberate: a human reviewer sits at each stage of the agent pipeline, and quality-assessment agents audit the output of other agents. The goal is not to remove humans from the loop but to restructure where they sit in it.

Ippen Digital: the architecture question

Markus Franz, CTO of Ippen Digital and founder of its incubator lab, made the case for a different mental model. Stop thinking about individual AI tools. Agents are the operating system; skills (reusable, composable capability modules) are the applications built on top.

Ippen has built a suite of agents covering content research, SEO, audience analytics, and editorial quality assurance, orchestrated by a conductor agent that manages task allocation among specialist sub-agents. A content research workflow that previously took a journalist two to three hours now takes five minutes, scanning 300 articles and surfacing relevant material with full sourcing transparency.

Key practical observations from Franz:

• Agents should be treated like junior employees: they require onboarding, context, defined personas and explicit guidance on tone, priorities and editorial standards.

• Multi-agent systems require each agent to carry an “agent card,” a defined identity that enables agent-to-agent communication across departments.

• Skills (reusable capability modules) are more durable investments than one-off agent builds.

• Interface design is the current bottleneck: most agent outputs are still delivered as walls of text that editors cannot work with efficiently.

• The adoption curve follows a predictable pattern: exploration, then specialised use cases, then genuinely novel workflows that emerge from the people using the system day to day.

Where the gap sits

The distance between the most and least advanced publishers in the room is not a technology gap. The tools are accessible, getting cheaper, and vendors are moving fast. What separates Hindustan Times’ 97.4% AI-generated code output from a publisher still running manual marketing workflows is not access to better software. It is the organisational decision to treat adoption as non-negotiable, and to build the governance, the incentives, and the measurement frameworks that make it stick. None is waiting to see how it plays out.

The Advertising Association marked its centenary at a reception in central London on 6 May, bringing together 250 senior industry figures. The occasion was billed as a reflection on a century of organised advertising. What it surfaced was the industry’s still-unresolved relationship with the public.

Chief Executive Stephen Woodford used the speech to reframe trust as a commercial concern rather than a reputational one. The argument: advertising’s claim on economic relevance depends on the public continuing to tolerate it.

“Advertising must be recognised for its contribution to growth, rather than as a cost to cut or something to ban. Being trusted helps drive that growth and we must keep earning the trust placed in us by improving the advertising experience and being accountable for the responsibilities we have.”

Advertising Association, Chief Executive Stephen Woodford

The speech had historical grounding. The AA was founded in 1926 because advertising was in crisis: unregulated, occasionally fraudulent, mistrusted as a result. The Association established the Advertising Standards Authority in the 1960s to impose independent oversight, a model since adopted as a regulatory template in markets well beyond the UK.

A century on, the numbers are better but not reassuring. Credos, the AA’s research arm, puts public trust in advertising at a five-year high of 40%. Woodford acknowledged the figure is fragile, and that the trajectory requires active management rather than passive confidence.

AI: Opportunity or Accelerant of the Problem?

AI is the most immediate complication. The industry has absorbed it rapidly, and Woodford’s assessment was notably optimistic given the context. Rather than treating AI as a threat to creative standards or employment, he argued it could be the mechanism by which advertising becomes less irritating: reducing ad bombardment, improving media experiences, making it easier to ensure that advertising is, in the AA’s foundational phrase, “legal, decent, honest, and truthful.”

Whether that ambition survives contact with the incentives of performance marketing is a separate question, but the framing signals where the AA’s leadership wants to position the argument.

The Association also announced a new Trust Action Plan, to be launched at its Parliamentary Reception in July. The details are still to come, but the timing (Parliament, mid-year, post-AI proliferation) suggests a policy pitch as much as an industry commitment.

Media Minister, Ian Murray offered the government’s marker. He called the AA instrumental in shaping the UK’s advertising landscape, pointed to the sector’s role in the Creative Industries Sector Plan, and gave the standard ministerial encouragement for a second century of the same.

The Mackintosh Medals: Two Awards for the First Time

The most substantive moment of the evening was the presentation of the Mackintosh Medals, the AA’s highest individual honour, introduced in 1952 and named after the first recipient, Lord Mackintosh. In 100 years, the medal had never been awarded twice in the same year. AA President Andria Vidler presented two for the first time to mark the centenary.

The first went to Kathryn Jacob OBE, recognised for her leadership across media businesses and her role in building the All-In programme on talent and inclusion, now described as the largest initiative of its kind in any UK industry. That scale matters. Inclusion programmes in the sector have historically been numerous and small, with limited cross-industry coordination, and All In was a structural intervention rather than a gesture.

The second went to James Best CBE, whose contribution has been primarily architectural: deep strategic work within the ASA system, and a central role in building Ad Net Zero, the industry’s climate commitment framework. He is also co-author, with Matt Bourn, of Trusted Advertising, a book that essentially codifies the case Woodford was making from the podium. Together the two awards made a coherent argument about what the AA values: institutional credibility and long-term structural change over profile and short-term visibility.

A Hundred Years, the Same Problem

There is something pointed about an industry body marking 100 years by returning to its founding problem. The AA was created because advertising had a trust deficit. The centenary speech was, at its core, about a trust deficit. The intervening century produced the ASA, Ofcom, GDPR and a proliferating body of self-regulatory codes. The public trust figure still sits at 40%.

That is not an argument against regulation. It may be an argument that trust in advertising is structurally bounded, that no amount of regulatory architecture changes the fact that advertising is, by definition, partial. The more productive framing, which Woodford edged towards, is that the industry’s job is to keep narrowing the gap between what it claims and what it delivers.

Whether that framing translates into anything concrete will become clearer in July, when the Trust Action Plan launches.

Ten years ago, the young news audience was already online-first. Now it is social-first, and the gap is widening every year the survey runs. The proportion of 18–24-year-olds citing social media as their main news source has risen from 21% to 39% over the past decade. News websites and apps have fallen from 36% to 24%. Television, never a serious part of this demographic’s news diet, is down from 28% to 21%.

Understanding Young News Audiences at a Time of Rapid Change, published in March 2026 by researchers Craig Robertson, Amy Ross Arguedas, Mitali Mukherjee, and Richard Fletcher at the Reuters Institute for the Study of Journalism, draws on twelve years of Digital News Report data to map how this shift has played out, and what it actually means. The headline numbers have been widely quoted; the structural argument underneath them less so.

The direct relationship is gone

Traditional media did not die in young audiences over this tracking period; it had already died, mostly before the period began. What the data captures is something else: the collapse of the publisher’s direct relationship with the young reader. The consequences of that collapse are not the same as those of simple audience loss, and the industry has been slow to notice the difference.

When young people encounter news on social or video networks, they do so incidentally, while there for other reasons. They are less likely to recall which outlet published the story. Only 14% of 18–24s say their main way of accessing news online is by navigating directly to a website or app, compared with 28% of those 55 and over. Direct access is the foundation of subscription revenue, email list growth, and brand loyalty. Incidental social discovery produces reach metrics that look presentable in a media kit and builds nothing durable underneath them.

Daily consumption tells the same story: down 15% points among 18–24s since 2017, against 5% for the over-55s. But the numbers understate the structural problem. Young people without habitual loyalty to a specific outlet are less likely to pay for content, less likely to subscribe to emails, and more exposed to whatever platform algorithms decide to surface or suppress next. The upstream risk is familiar: dependence on gatekeepers with no editorial interest in journalism’s survival. What the data shows is how far that shift has already progressed.

A different definition of news

The report carefully separates what young people are not doing, by traditional media’s own metrics, from what they are doing instead. 60% of 18–24s across 48 markets say they feel always connected to the internet. The disengagement is not from media or information. It is from one particular definition of what counts as news.

This is not, as it is sometimes framed, evidence of civic disengagement. It is evidence of a mismatch between journalistic output and audience expectations that has been building for a decade, and which most newsrooms have addressed only at the margins.

Creators are winning on platforms publishers depend on

Young people don’t just get news differently. They trust different people to give it to them. On social media, 51% of 18–24s say they pay most attention to individual creators and personalities for news. Only 39% say they pay most attention to traditional media or journalists. For those 55 and over, those figures are reversed.

The heaviest news consumers, the researchers note, tend to use both creators and mainstream sources; appetite for one does not preclude the other. And the creator landscape is not simply partisan commentary. A significant subset of mostly young creators has built substantial audiences by translating and contextualising mainstream reporting: Hugo Travers (HugoDécrypte) in France, Dylan Page in the UK, the Spanish outlet Ac2ality, which has more than six million TikTok followers, all operate as explainer services. Their raw material is often original journalism produced by organisations struggling to reach the same audience.

Facebook, the dominant social news platform for young people a decade ago, has been comprehensively displaced: 16% of 18–24s now use it for news. Instagram, YouTube, and TikTok have replaced it. The average young person uses 4.6 social or video platforms weekly for any purpose, against 3.4 for those 55 and over. No single platform reaches this audience; the fragmentation is itself part of the challenge.

The uncomfortable read for publishers is that the explainer creators succeeding here are doing a job legacy newsrooms used to consider their own. HugoDécrypte and Ac2ality are not breaking news; they are translating it. That work used to live on the homepage, in the stand-first, in the explainer box. It has migrated, and the audience has migrated with it.

AI is doing the explaining publishers used to do

On generative AI, the age gap is the sharpest the report finds anywhere. 15% of 18–24s used an AI tool to access news in the past week; among those aged 55 and over, the figure is 3%. Young people are not just earlier adopters but more elaborate ones: 48% of young AI news users say they use it specifically to make a story easier to understand, against 27% of older users. They also use AI chatbots to evaluate sources and navigate long-running stories where they lack historical context.

For a growing subset of young users, AI now sits between complex news and accessible understanding. That role used to belong to good journalism itself. NRK in Norway and Aftonbladet in Sweden both report that AI-generated summaries are disproportionately popular with young readers. Young readers are not asking AI to replace reporters; they are asking it to compensate for journalism that assumes prior knowledge they do not have.

This should worry editors more than the AI-summarises-our-content debate currently does. If a chatbot is the layer between a young reader and a complex story, the publisher has lost not just the click but the framing: what mattered, what didn’t, what the reader should make of it. That is the editorial function. Outsourcing it to a model trained on everything is a different proposition from outsourcing distribution to a platform.

The trust gap is smaller than assumed

Young people are routinely described as much less trusting of the news than their elders. The data say something narrower. The gap between 18–24s and those 55 and over across 18 markets in 2025 is 9 percentage points: 37% versus 46%. It has remained broadly stable since 2015, even as absolute trust has fallen across all age groups. In many countries, there is no statistically significant difference across age groups.

News avoidance follows a similar pattern. 42% of 18–24s say they sometimes or often avoid news, against 37% of over-55s, a five-point gap. The reasons matter more than the aggregate. Young avoiders are more likely to say news does not feel relevant to them (21% versus 16%) or that they find it difficult to follow (15% versus 5%). These are complaints about execution and agenda-setting, not fundamental hostility to journalism as an institution. The industry has spent considerable energy responding to a trust deficit that is partly real and partly misdiagnosed, while a more tractable problem, relevance and accessibility, has received less attention.

The distinction matters because the two problems take different work to fix. A trust deficit calls for transparency initiatives, corrections policies, and reader-facing explainers about how journalism is made. A relevance deficit calls for editing harder: choosing different stories, writing different leads, assuming less prior knowledge. The first is a communications project. The second is a craft one, and most newsrooms have spent the last decade getting better at the first.

The structural argument

The report’s authors frame their conclusions around pathways, formats, and definitions of news. Put more bluntly: the news industry has been treating a structural problem as an attitudinal one. Young people do not mistrust journalism in the way a decade of crisis commentary has suggested. They have simply found themselves in a media environment where journalism, as most publishers produce it, is rarely the most accessible, relevant, or engaging thing available. That problem does not get easier as this generation ages.

The habits that older audiences formed around direct news access, such as the morning paper, the evening bulletin, and the bookmarked homepage, were shaped by an information environment that no longer exists. There is no evidence, the researchers argue, that young people will age into those habits as their life circumstances change. The pipeline assumption that has underpinned publishing business models for decades is no longer safe.

What that means in practice is unglamorous. Publishers serious about reaching this audience need to stop measuring success by traffic from platforms they don’t control, and start building the things social and AI cannot easily replicate: distinctive voice, original reporting worth quoting, and direct channels (newsletters, apps, communities) that survive the next algorithm change. None of that is new advice. The data in this report is the clearest evidence yet that running the same playbook for another five years is the riskier option.

This article is based on the report published by the Reuters Institute for the Study of Journalism, University of Oxford. You an view the report by clicking HERE

The Advertising Association and WARC have published their annual Expenditure Report, and the most consequential thing in it is not the £46.7 billion headline figure for 2025 (up 6.4% year-on-year). It is a reclassification that reframes what every previous dataset was measuring.

For the first time, retail media and social media are being reported as standalone channels. They had previously been absorbed into broader categories, making it difficult to track their true scale. Pulling them out changes the comparability of every channel total in the report.

What “Other Online Display” tells you

“Other Online Display” fell 20% in 2025 and is forecast to drop a further 25% in 2026. That category, which includes audio, gaming, video outstream, native advertising, and display embedded formats, now explicitly excludes revenues from social, retail media, TV, radio, and news and magazine brands. Strip those out and what remains is a contracting residual.

The same logic applies to search. Retail media has now been separated from the search total, which means the £17.9 billion attributed to search in 2025 is not directly comparable to previous years’ figures. Search still accounts for the largest share of UK advertising investment at 38.3%, but that number has been quietly recalibrated.

The channels growing quickly

Retail media recorded full-year growth of 17.5% to reach £3.75 billion, and its Q4 performance was particularly sharp, up 30.5% year-on-year during the festive season. That tracks the US trajectory, where retail platforms now compete with traditional media owners for brand budgets, not just performance spend.

Addressable TV grew 37% across the year to £1.84 billion, though total TV revenue declined 1.2% when linear is included. Addressable is taking share inside a shrinking total, buyers are reallocating within TV before they reallocate away from it.

Social media (including YouTube) grew 21% to £11.5 billion, 24.7% of all UK advertising investment, a larger share than TV and published media combined.

Where the money is leaving

Published media (national and regional newspapers and magazines, print and digital) declined 5.1% in 2025, with regional news brands falling 6%. Even online revenues within regional titles dropped 2.2%. National news brands fared slightly better, with online revenues down just 0.9%, but the overall picture for print-heritage publishers is one of structural decline. Direct mail was broadly flat, down 0.3%.

What comes next

The AA and WARC describe this as phase one of a broader update programme. Phase two will address three areas:

• How investment in the influencer and creator channel is captured

• How to categorise spend in the emerging generative AI and LLM advertising channel

• How to distinguish between the investment patterns of large brand advertisers and the estimated 3.5 million UK SMEs that advertise annually.

Stephen Woodford, Chief Executive of the Advertising Association, framed the refresh as a response to a more fragmented market: “This evolution of the AA/WARC Expenditure Report will ensure the industry has the best possible information to guide understanding of investment across the UK advertising and media landscape.”

The LLM channel question is worth watching. At present there is no agreed methodology for measuring what brands spend on AI-native placements sponsored outputs in chatbot responses, model-embedded brand integrations, and similar. Without a standard, AI-native placements stay invisible in benchmarks like this one, even if the underlying spend is already material.

For more information visit www.warc.com/expenditurereport