The first Future Media Series forum opened on 20 October in the Agence France-Presse building on the Place de la Bourse, two days after riots in the capital had shaken the country, and nobody pretended the afternoon was about anything other than money. The slogan on the stage ran “regulation, revenue, renaissance”. The promise was that everyone would leave with a route to a cheque. Four hours later the room had a reasonably clear map of where the cheques come from, and a much less clear idea of whether the map survives contact with AI.

The programme was built for that question: Fabrice Fries, Chairman and Chief Executive of AFP; Damien Geradin of Geradin Partners; Mikaël Hervé of Charles River Associates; Paul Fletcher and Stephen Jones, Australia’s former Communications Minister and former Assistant Treasurer; Pierre Petillault of the Alliance de la Presse d’Information Générale; and Elena Perotti of WAN-IFRA, with Ricky Sutton hosting and Lexie Kirkconnell-Kawana of Impress in the chair. Every session ran under the Chatham House Rule: the names above are the programme’s, and nothing that follows is attributed to any of them.

Competition law got publishers paid. Copyright on its own did not

France and Australia reached the same conclusion by different routes. Spain tried a copyright-based link levy in 2014; Google closed Google News and paid nobody. France wrote the 2019 EU copyright directive into national law and then took the fight to the Autorité de la concurrence, the French competition authority, which fined Google €500m in 2021 and a further €250m in 2024 for failing to negotiate in good faith. Australia skipped copyright entirely and had its competition regulator write a bargaining code after an 18-month inquiry into Google and Facebook.

The logic, as put from the stage, is that a copyright argument invites a platform to dispute the value of every extract indefinitely. A competition argument asks a simpler question: may a dominant company neutralise a legal right by refusing to negotiate? Framed that way, the stock defences stop mattering. We send you traffic. A snippet is worth nothing. Neither answers the question. As for what the competition route is worth, the French press reported the week before an ICC arbitration award requiring Google to pay French magazine publishers around €80m a year. Google is contesting it.

Nothing moved without a prime minister, and nothing stuck without a backstop

Both countries had a head of government in the meetings. In Australia, Google threatened to pull search from the country; the Prime Minister met Microsoft’s Chief Executive, in full public view, and the threat evaporated. Meta then switched off the Facebook pages of Australian news outlets, and a state health department’s for good measure, which united a divided parliament behind the bill. The design feature that made the platforms sign was final-offer arbitration: each side submits a number and the arbitrator picks one. The platforms concluded it would hold up in court, and deals estimated at more than A$200m a year followed.

Then Meta stopped paying in 2024. The government of the day chose not to force arbitration, judging that the small publishers who depended most on Facebook would be the casualties if it pulled news entirely. Instead, it legislated, in August, a charge of 2.75% of Australian digital advertising revenue on the largest platforms, offset at 150% for money spent on news deals and 200% for deals with small and medium publishers. Call it an incentive; the mechanism is tax law. Whether Meta should instead have been designated under the code, and the original leverage tested in court, was openly disputed on stage, with the ease of people who have had the argument before.

France is drawing the same lesson. Five years of good-faith obligations have not produced enough agreements, nor kept all of those signed, and a bill giving the regulator Arcom an arbitration role has passed both chambers of parliament and awaits final adoption. Without a backstop, as a line that recurred across sessions had it, you talk forever.

Strength in numbers, and the deals that give it away

Divide and rule is the platforms’ first move everywhere. In France, the directive had barely become French law before three national titles signed individual deals with Google. The collective came later and did better, on a model of collective bargaining with individual licences: the alliance negotiates, each publisher decides whether to sign. The same sequence is now repeating with AI. A handful of one-to-one deals exist, and the room was warned that some AI pilot agreements hand over an entire archive for a modest fee and a release clause.

The money on the other side of the ledger is now large enough to make unity attractive. The route to a cheque runs through litigation funders, who carry the cost of a competition damages claim in return for 20 to 25% of the proceeds, and a special purpose vehicle that keeps a publisher’s name off the pleadings. Group claims against Google’s ad tech business are in train in the Netherlands (56 publishers, €2.6bn) and the UK (about £13bn, on behalf of the whole industry). The fear of retaliation, raised from the floor, was judged to be fading: there is safety in numbers, and judges read the newspapers. Competition law draws one line: a collective damages claim is fine, a shared valuation methodology is fine, a collective boycott is a cartel.

The platforms alone know what the content is worth

The valuation argument reduced the problem to two asymmetries. Platforms have more bargaining power, because no single publisher is indispensable to them and the reverse is not true. And platforms have more information: they alone know the audiences, the engagement and the revenue that news generates. Valuation, on this account, is an empirical question, and the right counterfactual is what Google loses if all news disappears, not what it loses if one title does. Both asymmetries widen when the product is an answer rather than a link.

The value has moved from links to answers

Every mechanism described during the afternoon was designed to price a link or a snippet, and the argument from the stage was that this world is ending. Summaries and chatbots blend thousands of sources into an answer that replaces the click, shifting the value from an extract to a publisher’s entire output. French publishers have already complained to the competition authority about AI Overviews launching without any negotiation.

Australia’s new incentive does not cover AI platforms at all, and a question from the floor about whether Canberra is softening on copyright to attract data-centre investment drew the reply that Australia’s glass is three-quarters full when most countries’ stand empty, and no denial of the rumour. WAN-IFRA research cited on the day put newsroom AI adoption at 80%, up from 40% a year earlier, with few publishers seeing any route to revenue and most seeing search referrals fall. Opt-outs, nobody on the panel disputed, do not work.

The mood has turned, and that is the renaissance case. Threats of a return to the Stone Age no longer frighten legislators, even if American tariffs still do, and the platforms’ standing with voters, and by extension with judges, has collapsed. The case against is that a decade of regulatory engineering has produced a machine for pricing something that is disappearing, while governments legislate in cycles of years and the thing being priced changes every quarter.

Three-quarters full, then, for a glass that is draining.

TikTok’s UK communications and partnerships teams briefed journalists at the company’s London office on Wednesday morning, taking the room through what British creators are doing, sector by sector.

Lauren La Faci, Communications Lead, opened with the creators themselves: Francesca Rose, who began posting this year and has passed 200,000 followers with wholesome train journeys around Britain; a Transport for London (TfL) employee posting as Tube Bellingham about the realities of the job; Rob doing Rob, a gardener in a sleepy village. Their common quality, in La Faci’s phrase, was “using their Britishness to transcend different communities”.

The fan edit outperforms the trailer by 37 to one

Lily Hall, a Partnerships Manager on TikTok UK’s entertainment team, works with studios, broadcasters and streamers. Over the past year, she said, 12 million posts about film and television have been published on TikTok every day globally, and 99% of them are fan content. Her example was Lionsgate. The studio’s official trailer for The Hunger Games: Sunrise on the Reaping, posted on its own account, had 227,000 views. Its best-performing fan edit of the franchise had 8.4 million. The edit she played had been made by an in-house TikTok fan editor Lionsgate now employs.

“We’ve seen examples of studios such as Lionsgate or HBO hiring TikTok editors as in-house trailer makers after having seen their edits on the platform,” Hall said. Around Spider-Man: Brand New Day, TikTok ran a competition inviting fan editors to recut the official trailer asset, with the director, Destin Daniel Cretton, picking the winners; Spider-Man edit posts grew 104% over three months. Movie-edit posts are up 24% in the same period. “No film or TV show can truly be considered a fan favourite until the TikTok editors have worked their magic on it,” she said.

Hall traced the studios’ relationship with creators from paid promotion, through junkets and red carpets, to behind-the-scenes filming on sets as large as Christopher Nolan’s The Odyssey, and now to casting. George Clarke was on the BBC’s Strictly Come Dancing last year; the football creator John Nellis is in this year’s line-up. Searches for the programme rose 1,329% in the week of its launch compared with the week before.

The X Factor has not aired since 2018 and still draws 3.7m views a day

Love Island generates more posts than any other British programme, 1.4 million globally for the latest series, and was the UK’s top search term on TikTok during finals week this year and last. Big Brother is having its biggest season in years. The X Factor, off air since 2018, still draws more than 3.7 million views a day, and a Saturday Night Live UK sketch parodying its peak years has sent creators back to the 2008 to 2011 series. Horror posts are up 26% in six months; the audience that found the genre through this year’s creator-led hits Obsession and Backrooms is working backwards through Scream, Halloween and Friday the 13th.

Anime, Hall added, now tops the platform’s television leaderboard in Britain, with creators posting about Naruto, Jujutsu Kaisen and One Piece more than any other shows. A community the industry treated as niche is outranking Hollywood on fan conversation.

Fans have moved from promotion into development

The claim Hall pressed hardest was that fan communities now shape the IP itself. Prime Video announced the leads for its Fourth Wing series last week; Emilio Sakraya, cast as Xaden, had been campaigned for by the book’s TikTok fans for months. Dungeon Crawler Carl, originally self-published, went into development on the strength of its online following. Studios are working through romance and young-adult fiction, Heated Rivalry and Off Campus among them, and through the classics: Sense and Sensibility, Wuthering Heights, The Odyssey and a forthcoming Pride and Prejudice, where book, film and history creators converge on the same release. “Whether it’s execs taking note of who fans want to see cast, or originally self-published titles going into development because of this huge online community behind them, it’s been really interesting to see that develop this year,” Hall said.

BookTok now has an official chart and a bricks-and-mortar footprint

Yaz How, TikTok’s Head of Emerging Communities, took the publishing numbers. More than 50 million books recommended on TikTok were sold across Europe in 2025, she said. The official BookTok chart, published monthly with NielsenIQ BookData sales figures behind it, ran from romantasy to crime to comedy in September. Rebecca Yarros’s Threshing Day, a story collection set in the Fourth Wing world, was her example of the community doing more than recommending: fan theories, dream castings and, in one creator’s case, commenters talking her into a book she had meant to skip.

TikTok’s own survey figures: 66% of BookTok users say the community has led them to read or listen to a new book, 76% have tried new authors or genres, and 54% say they have supported a local business or bookshop as a result. Bookshops are using the platform in turn, and their followers, How said, are there for the owners as much as the stock. Jack Edwards, whom she called the internet’s resident librarian, has taken his audience off platform into the Inklings book club and its podcast. His response to Alix Earle’s admission that she had never read a book was a reading list for beginners. How’s gloss on why that converts: “You don’t need to be a literary expert or to have read all the classics to be part of the BookTok community.”

TikTok supplied almost all of the figures, but the companies it is pitching to have already acted on them: Lionsgate and HBO have put fan editors on the payroll, Strictly has cast two creators in two years, and the UK’s BookTok chart now runs on NielsenIQ sales data. Each of those decisions rests on an audience that TikTok’s recommendation system assembled and can reassemble at will, a dependency the morning did not chart.

Thirty minutes before Jessica Sibley walked on stage in Madrid, Time published its first interview with John Ternus, Apple’s new Chief Executive. He had given it to nobody else. The cover, timed to the launch of the $2,000 iPhone Duo, was the last thing she checked on her phone before the closing session of the FIPP World Media Congress, where she spoke with Minette Ferreira of Media24.

That access, she argued, is the asset everything else at Time is built on. The question she has spent four years answering is what to build.

Time 3.0 began on a blank screen, three months before she had the job

During the recruitment process in 2022, Sibley was asked what she would do with Time. She wrote “Time 3.0” on an empty document, and the phrase became a rallying cry before it was a plan. By the day she joined she had a written plan, though, as she put it, “while I thought I had all the answers, I knew I didn’t.”

Her first act was to leave the building. Rather than sit with consultants and a whiteboard, she announced she would meet 100 stakeholders in 100 days, pegged to the magazine’s centenary, and hit the number far faster. Politicians, media buyers, chief executives and celebrities all had a Time story, and told it to her as though it were personal.

The commercial diagnosis was blunter. Time had made the move to digital later and less successfully than its rivals. She chose not to chase them. “Do we look back and try to catch up? I kind of thought of that as chasing trends. Or should we just prioritise our priorities and figure out how to build the best version of ourselves?”

The Google Zero dashboard was built before Google Zero was a slogan

A chief product officer was hired within days of her arrival, and one of the first things the team built was a dashboard showing what Time would look like with no Google traffic at all: if search went away, which businesses still worked? The answer shaped the diversification that followed. “We have a smaller website, but we’ve got the largest social audience in our history,” she said. The site matters less than the brand’s reach, and she argued the same logic now must apply to AI.

The cover gets you the stage, and the stage gets you the room

Time’s events and franchises are now its commercial engine. Nine covers in ten carry a person, and Sibley noticed early that everyone wanted to be on one, even people who had stopped giving interviews. “When you’re on the cover of Time, you then are on the stage,” she said, and the stage is where Time gathers the community it then sells to sponsors.

Her example was Taylor Swift, Person of the Year in 2023. “She doesn’t have to give an interview. She could talk to her fans directly. She knew being Person of the Year on the cover of Time was history-making.” The Time100 franchise has since expanded into AI, health and creators, and the newsroom’s question for every list is the same: is it your year?

Why Time? Her answer was trust, which she called the brand’s number one priority, backed by a century of reputation. Many publishers say as much; few have the cover list to show for it.

The AI strategy started with a text message and a pause

Sibley’s account of Time’s AI journey began on the day ChatGPT launched, when Marc Benioff, Time’s owner, texted to say Sam Altman would call her directly. He did. Benioff’s instruction was short: “I trust Sam Altman. You guys can figure this out.” Then Time stopped and worked out what it was doing.

She described five phases. First, infrastructure: protecting IP, understanding what AI systems were doing with 100 years of journalism, training the newsroom on the tools, and getting a seat at the table with the companies building them. Second, a decision to negotiate with the AI companies over the value of Time’s content. Third, licensing and partnerships. Fourth, products, built with Scale AI, including the Time AI agent that summarises articles, reads them aloud and searches the archive in 13 languages. Fifth, and current, advertising.

Agents want the opposite of what readers want

The infrastructure work revealed a pattern in Time’s own traffic: agents were arriving in volume, and Time concluded the open web was heading toward a point where they would outnumber humans. By Time’s own account, it has already arrived on some days: Mark Howard, its Chief Operating Officer, told Digiday in July that the site sees more bot traffic than human traffic “most days”.

She decided to treat them as an audience. They are agents working for people, answering questions, informing decisions, and increasingly making purchases, and they go looking for trusted sources. What they want from a page is nothing like what a reader wants. “Humans are coming to Time.com, and you’re seeing our beautiful visuals and our covers. Agents want the opposite. They also want speed, but they want a machine-readable version that doesn’t have any of the stuff that humans want. They want structured data.”

So Time, with the ad-tech firm Mobian, began serving agents a text-only markdown version of its pages and, in July, sold the first ads written for them. Ally Bank, sponsor of the Time100 Creators list, ran a conventional campaign for readers and, alongside it, an agent ad: a plain-text set of brand-verified facts about its products and positioning, labelled as sponsored and structured for machines to ingest.

The pitch to brands rests on three measures: visibility, favourability and accuracy. A study of 750 blue-chip brands by Mobian, which sells the product, found what Sibley called a “pretty big” discrepancy between what AI systems say about them and their verified facts, which she put down to outdated or untrusted sources. Time’s offer is to close the gap, with its own trusted pages as the carrier.

Ferreira closed on the convergence of the Congress’s two themes, trust and AI. Whether a page built for machines can carry the trust earned by a page built for people is the question Time is now being paid to answer. Sibley’s advice to the room was the line she had used for Time’s own journey. “Don’t let AI happen to you,” she said. “Let AI happen for you.”

For years Harvard Business Review grew by widening the funnel. Sub-brands multiplied: HBR Ascend for younger professionals, the Women at Work podcast, the HBR Presents network. Articles on how to ask for a pay rise drew heavy traffic.

On the first day at the FIPP World Media Congress in Madrid, Erika Heilman, HBR’s Group Publisher, explained why the title is now pointing itself at a far smaller group of readers. The result is HBR Executive, a premium offering for senior leaders launched in May 2025 and now in its second year.

Traffic turned out to be a poor guide to value

Heilman was careful not to disown the wide-funnel era. “This work was not a mistake,” she said. “It helped millions of people and taught us how distinct audiences behave.” The difficulty was that reach and revenue stopped moving together. Early-career advice travelled furthest and reached the readers least able or willing to pay. Conversion was low, and every property needed its own editorial attention, production and upkeep.

“Traffic can tell you something is useful; it cannot tell you whether the audience values it frequently enough to support the cost of creating it.”

She described the error as “mistaking a strong audience signal for a strong business signal” and called it a common one.

Generative AI answers the easy questions now

Then referral traffic from search and social fell, and generative AI accelerated the decline by separating content from its source. Heilman’s example was a manager asking how to give difficult feedback in a performance review. HBR has published plenty on it, and much of its archive has been used to train AI models. A plausible answer now arrives instantly, with no visit required.

Information, good and bad, “is abundant, but judgement remains scarce,” Heilman said. HBR’s job, as she framed it, has moved from supplying generic answers to helping leaders understand which trade-offs apply to them. “Our ambition is to become part of how leaders decide, not merely part of what they read or consume.”

The premium audience was already paying

The data made the choice to go upmarket easier than it might have been. Some 40% of HBR’s subscriber base was already at Vice-President level or above. “This wasn’t a speculative market; it was an existing relationship hiding in plain sight,” she said.

HBR listed the assets rivals would struggle to copy, applying the “right to win” test from strategist Roger Martin’s Playing to Win: editorial trust, a broad contributor network, access to Harvard Business School’s executive education, and relationships with chief HR and learning officers through its parent, Harvard Business Publishing.

• Segmentation research by an outside strategy firm produced a specific brief. Senior readers said they wanted:

• Trusted, credible information above everything else.

• Insight tied to action, with less emphasis on volume.

• Curation and personalisation of a vast archive, which HBR had long described internally as “value for time spent”.

Delivery inside the workflows they already use.

Peer exchange that offers “real insight and context, not networking for its own sake”. AI for speed and efficiency, with humans supplying judgement, nuance and context.

Further research found executives develop through four routes: premium content, executive education, coaching and peer networking.

HBR already owned the first and, through HBS, had access to the second. Coaching is being explored, possibly with AI. Peer networking is where much of the new work has gone.

What HBR built, and what it cut

HBR Executive launched around two needs its editors use as a filter: “enlighten me” (the signals that matter) and “guide me” (application to the reader’s own situation). Peer connection and a way to rehearse difficult conversations are now being added.

HBR also beta-tested Strategy Lab, a proprietary AI strategy tool built with HBS and designed to be demanding and pedagogically sound. Feedback was good. It was dropped anyway: building and maintaining it was expensive, and the features subscribers would have lost to pay for it made the opportunity cost too high.

What subscribers asked for instead was candid, multi-perspective conversations about real challenges, in closed groups where they could speak with psychological safety. HBR now runs case study discussions and communities of practice that put members in a room with HBR Press authors, HBS faculty and magazine contributors. A white-labelled digital community platform is being researched for 2027.

Because senior leaders’ employers budget for learning and development, HBR added group subscriptions for executive teams, companies and membership bodies. Given the headwinds in direct-to-consumer subscriptions, Heilman called it a necessity. “You really can’t just rely on one customer acquisition channel.”

Telling staff what stops is the hard part

HBR’s staff were proud of the reach-era brands. To bring them along, Heilman used a stop-start-continue frame, crediting advice from HBS Professor Frances Frei:

· Continue is the largest category and the one to stress first, because it lowers anxiety. “Most of what you do is going to stay the same.” At HBR that meant editorial rigour, its frameworks and its expert network.

· Start should be limited to two or three new capabilities, so the team is not overwhelmed.

· Stop “is perhaps the hardest one to say”. Heilman found it easier to present as a resourcing and prioritisation decision, with the discontinued work honoured as part of the journey.

She also had to answer the charge that serving fewer people means doing less good. “Focusing on senior leaders does not mean retreating from impact,” she said. Decisions on strategy, technology, jobs and AI taken in the boardroom cascade to employees, customers and partners.

HBR Executive is on track against its goals, according to Heilman, though she gave no subscriber figures. The next chapter, she said, “will be determined by adoption”: whether leaders use what has been built, and whether executives behave the way they told the researchers they would.

Jon Roberts opened his session in Madrid with a mathematical scenario. A person who searches for something reads one or two articles. An agent sent to answer the same question may read between 10 and 100. So, with no change in how many questions get asked, “we would be consuming 100 times more content,” the Chief Innovation Officer of People Inc. told Stig Ørskov, Chief Executive of WAN-IFRA, who chaired the session, titled “Getting Paid in the Age of Answers”.

The number of questions is not standing still either. Roberts argued that people using AI tools ask narrower, more specific questions than they ever asked before, so publishers need to produce new content to answer them, not just get paid for the content they already have. He described AI as a business with a supply chain. “AI needs chips. It needs power. It needs information.” Chips and power are well funded. Information, in his account, is the third pillar and the only one nobody is paying for.

“This isn’t for us a question about how we make ourselves whole,” he said. People Inc., the IAC-owned publisher formerly known as Dotdash Meredith, treats AI as a growth market and has a three-step method for entering it: block, get paid, then become a partner worth paying.

Blocking everything cost People Inc. nothing

Since January, People Inc. has blocked every non-human system that tries to reach its content, whether an ad tech bot, a Google crawler or anything else, with 38 exceptions covered by specific agreements. Roberts dismissed robots.txt, the voluntary protocol most publishers still rely on, as about as useful as a chocolate teapot. People Inc.’s content delivery network, Cloudflare, does the blocking using the same layer that repels denial-of-service attacks. Roberts said Fastly and other CDN providers now offer the same capability.

Ørskov suggested the company had simply accepted the hit to its reach. Roberts said there was no hit. “We lost no revenue, and we lost no traffic.” Advertising rose, “for some completely weird separate reason”. The block now turns away tens of thousands of unique bad actors and tens of millions of access attempts a day. “It is industrial, what is happening to your site.”

Some bots claim to be Google; others pose as readers, arriving through smart TVs and phones. People Inc. has taken its logs and screenshots to Capitol Hill as evidence of what Roberts called wholesale theft.

The case for blocking is mostly defensive: if someone is robbing your house, Roberts asked, why would you not lock the door? Blocking has also proven good for business. Licensed partners value that their rivals cannot get the same content for free, and some of the companies being blocked now contact People Inc. to ask how they can pay for access.

The test is whether a buyer builds a platform

Roberts sorts AI companies into two categories. “If you’re a parasite, we block you. If you’re a platform builder, we won’t block you.” In his definition, platforms create value that whole industries and economies can grow on; parasites extract it. The line can run through a single company: People Inc. admits Google’s search crawler and blocks Google-Extended, which governs use of its content in Gemini. “Search is a platform. Gemini is a parasite.”

Of the 38 permitted systems, the three largest deals are with OpenAI, Meta and Microsoft. The first two supply ChatGPT and Meta AI, with training allowances and licensed use inside those products. The Microsoft deal works differently: People Inc. is one of the publishers building Microsoft’s publisher content marketplace, which Roberts sees as a route to millions of buyers: any company that builds its own AI product and wants premium, rights-cleared content. He said the group’s health and finance titles are already in heavy demand from AI systems, because getting health information wrong is a mistake those systems cannot afford.

People Inc. also blocks the start-ups building marketplaces for scraped web content. One of them, Exa, raised $250m this year at a $2.2bn valuation. “Anybody who says there’s no money in content for AI is wrong,” Roberts said. Venture capitalists, he argued, do not put a quarter of a billion dollars into a single round unless a company has found its market. Exa has customers; what it lacks is suppliers.

Publishers are waiting for AI’s Spotify

He reached for the music industry. Napster took the product; Spotify, whatever one thinks of its terms, is a platform others can build on. Publishers have no equivalent for AI. When Ørskov asked whether he would rather be paid for access or use, Roberts said he did not mind much, provided a functioning marketplace emerges. The two models need different contracts, but either would give publishers a reason to invest. He reckoned the companies in the room could put at least a billion euros into new content if a platform justified it.

When Ørskov raised SPUR, the publisher coalition launched in Britain this year, Roberts agreed it was one of the groups trying to set the terms for that market. SPUR, he said, is in favour of an AI future, just not a parasitic one. His complaint is that the AI companies have shown no ambition to build the platform that would make such a market work.

People Inc. sells a better product than the scrapers take

Step three, being a good partner, is for Roberts a product question. Crawling, he argued, is a very bad way to get information. Most of the HTML built for human readers, he said, is useless to an agent. Pages are crawled only because the web is “kind of the USB connector of content”.

People Inc.’s alternative is an API that returns articles as Markdown, roughly 5% of the weight of the full page, which Roberts said makes them 95% cheaper for a model to reason over. Each piece carries clean metadata on when it was written and what matters. Latency is about 60 milliseconds, against seconds for a scraped page. Rights are handled at source: images are not sent, because People Inc. does not hold the rights to pass them on. “That’s not trying to tax AI companies,” Roberts said of the API, which People Inc. sells as a paid service rather than giving away. Only the earliest of the big partnerships still relies on crawling.

Subscribe now

Small publishers have more leverage than they use

Ørskov’s last challenge was whether any of this works without People Inc.’s scale. Roberts conceded that no single publisher, however large, can shape the market alone. A specialist title’s content is harder to copy than a national’s, though, and that is leverage. “Do not think that because you are small you don’t have leverage,” he said. “You only have leverage if you start to use it.”

The first step is the one People Inc. has already taken: until a publisher blocks, the door is wide open. The second is to bargain collectively, through the RSL Collective, which backs the Really Simple Licensing standard, ProRata and similar groups.

Asked what publishers should do when they get back to the office, Roberts reached for the newsstand. Publishers, he said, are putting their product on the stand and letting everyone steal it. “Step one: stop them stealing it.”

Any publisher can block. Whether blocking ever pays depends on a marketplace nobody has built yet, and Roberts can't offer a date for when AI companies will get around to it.

None of the three case studies Lennart Schneider brought to Piano Academy in Amsterdam publishes news. His examples came from a restaurant app, a language app and a book-summary service, and he argued that publishers, among the oldest sellers of subscriptions, now have most to learn from businesses that came to the model late.

Schneider spent more than six years at Die Zeit, where he helped build the Freunde der ZEIT loyalty programme, before setting up as an independent subscription consultant. He hosts Subscribe Now, a German-language podcast that interviews subscription operators across sectors, and Haufe published his book of the same name in June.

The reader who already pays is the cheapest acquisition channel

NeoTaste launched in Germany in 2021 and sells a subscription to restaurant deals, typically two main courses for the price of one. It now operates in more than 50 cities and says it has more than 2 million users. Much of that growth comes from existing users. “Their users are their best marketers,” Schneider said.

The programme is built to take as little effort as possible. A subscriber generates a personal referral link with one tap and sends it to a friend, who lands on a personalised page and joins for free for a month. The referrer also gets a free month, and an account page shows which invitations have been accepted. The new member can then start the loop again.

Schneider took the finer points from a podcast interview with Stefan Bader, Chief Executive and Co-founder of Cello, a Munich company whose referral software runs programmes for NeoTaste, Miro and Riverside. The most common error, Bader told him, is burying the programme in a menu nobody opens. It belongs on the first screen of the app. The best moment to ask is straight after a good experience, which Bader calls a moment of delight. For NeoTaste, that is the meal itself. For a publisher, Schneider suggested, it could be the end of an article or a finished game.

Rewards range from free months to cash and, for AI tools, usage credits. Each business has to test what works for it. The rule Schneider passed on from Bader was that “the rewards have to feel fair and balanced on both sides.”

Duolingo grew by counting who came back the next day

Referral brings people in; Duolingo’s case was about keeping them. By 2018, its daily active users had stalled. The company looked to mobile gaming for ideas and settled on a single product metric: current user retention rate, or CURR. A current user is active today and was also active at least once in the previous seven days. CURR is the share of those users who return the next day. Over the next four years, daily active users grew 4.5 times. Duolingo’s shareholder letter for the second quarter of this year put CURR at an all-time high of 84%, up about a point on a year earlier.

Schneider’s argument for paying attention rests on the lag built into subscription businesses. A product change or CRM campaign may move churn months later, by which point the cause is hard to isolate. Engagement scores shorten the wait, which is why publishers already use them: the Financial Times’ recency, frequency and volume model, median time spent, or the ratio of daily to monthly active users. Duolingo’s version demands a daily return.

Every user is assigned each day to one of seven mutually exclusive states. Active users are new, current, reactivated after more than a week away, or resurrected after more than 30 days. Each movement between states has a conversion rate, and each conversion rate can be the target of an experiment. With search traffic falling, Schneider argued, publishers face the same arithmetic. “We have to do more with less, and that means that we have to re-engage our existing user base,” he said.

Duolingo runs hundreds of experiments a year to that end. One prompts users to install a home-screen widget that changes with their state and the time of day, and grows more distressed the later it gets without a lesson completed. Schneider also showed slides from the Duolingo Handbook, the company’s published account of how it works. By his account, only about half of Duolingo’s experiments ship, because it tests outlandish ideas alongside obvious ones; “sometimes better ideas come from the worst ideas,” as he put it. The company works to cut the time from idea to test to release, and insists the product stay fun, since learning a language is hard work and enjoyment is what brings people back. “We need to start focusing on habits, and habits mean daily routines that bring people back to our apps and our websites,” Schneider said.

Telling trial users when they would be charged sold more trials

The third example, from Blinkist, the Berlin book-summary app, concerned the free trial. Users start one, forget to cancel, get charged, then leave angry app-store reviews and ask for refunds. Blinkist’s data showed that a third of trial cancellations happened on the first day, before users had given the app a chance.

Its redesigned trial screen set out a timeline: full access today, a reminder two days before the seven-day trial ended, and the date the charge would land. To receive the reminder, users had to switch on push notifications. Trial sign-ups rose 23%. Cancellations fell because users waited to see what the app could do. Complaints dropped by about 55%. Push opt-ins rose from 6% to 74%, giving Blinkist a direct line to three in four trial users. The format, Schneider said, has since been copied across subscription apps.

His advice was to look beyond other publishers when redesigning a paywall. Mobbin, a library of screens and user flows from apps and software products, makes it easy to see what the leading subscription businesses do. “We have to be transparent. We have to show them how it works,” he said.

Subscribe now

Publishers will need another industry’s vocabulary

Schneider ended by telling the room to skip the next publishing conference in favour of a SaaS or app event, and to follow the people who run those businesses through their podcasts, newsletters and LinkedIn posts. The point was partly linguistic: “they are sometimes using very different words when they are talking about the same things.”

The designer behind Blinkist’s redesign later wrote that the company had been making money from users who forgot to cancel. Publishers with auto-renewing trials and cancellation pages that take some digging collect a share of that money. Blinkist gave its share up and sold more trials. Few finance directors will sign off on that trade based on another company’s A/B test.

The first day of Piano Academy in Amsterdam brought product news, chiefly that Stripe Checkout is coming to the Piano platform. Four practical sessions, from a German regional publisher, the BBC’s commercial arm, a Dutch data agency and Piano with Stripe, made the same case: the revenue publishers are missing sits in audiences, subscriptions and data they already hold.

The most loyal readers are also the best advertising inventory

pd digital runs the digital operation for Mediengruppe Pressedruck, publisher of the Augsburger Allgemeine and a cluster of regional titles in southern Germany. Jürgen Risinger, its Head of Subscription Technology and UX, began with a diagnosis most of the room shared: search traffic down sharply, CPMs following it, and no appetite for compensating with more ads. “Our traditional approach of monetising our digital products does not work anymore, or at least it’s not as effective as it has been in the past,” he said.

The response, built during a relaunch of the group’s news sites on a single platform, is what pd digital calls revenue-driven design. Piano Audience sorts each visitor into one of five tiers:

• Unknown.

• Low loyalty (one visit day in the past 30).

• Mid loyalty (two to four visit days).

• High loyalty (five or more).

• Subscribers.

A server-side call passes the segment to the CMS, which serves the matching template. Unknown and low-loyalty visitors, many arriving from Google Discover, get heavier ad loads and nationally sold campaigns. As loyalty rises, ad load falls, the advertising becomes more relevant and subscription offers grow more prominent. “Why bother a first-time user coming from Google Discover with a subscription offer?” Risinger asked.

Piano’s Ad Revenue Insights, which attributes advertising income to content and users, supplied the measurement. The ad operations team reported:

● 20% better ad visibility.

● 7% higher average CPMs.

● 11% more ads per page view.

● 32% more revenue per page view.

Subscribers spend more time on the sites each day and generate around 11% more page impressions.

High-loyalty users are about 4% of visitors and produce 28.5% of digital advertising revenue. Subscribers are 1% of monthly visitors, 10% of advertising revenue and, once subscription income is added, about half of digital revenue. Some articles with 50,000 page views, Risinger said, earn far more than others with 100,000.

Relaunches usually pit newsroom, ad sales and reader revenue against each other over placements. At pd digital, tests now settle those arguments. “If you have good data, you can prove that the design works,” Risinger said. The concept is shortlisted for Piano’s new awards, decided on day two.

Breaking news was costing the BBC subscribers

Joe Iles and George Agate of BBC Studios reported on progress since the BBC.com subscription launched for US audiences in June last year. It is now live in the US and Canada across web, iOS and Android.The BBC moved cautiously, starting with about 10% of audiences eligible and running a likelihood-to-subscribe model before switching to Piano’s dynamic paywall, always the goal for a business that relies on both advertising and subscriptions. In a 50/50 test, the dynamic version put the paywall in front of 160,000 more visitors with little cost to advertising, finding converts among thinly engaged users through contextual signals.

During major news spikes, page views at times doubled while subscriptions fell by 25%. Readers landed on the live page and went no further. The BBC’s fix was editorial. A redesigned homepage now wraps the live page in paywalled analysis of the same story. Since the change, page views and conversions have risen together.

A meter of about four pages works for most users. Some 42% of active subscribers converted in the iOS or Android apps. In Canada iOS is the largest subscription surface, ahead of the web. Bringing propensity models to the apps is next.

The BBC also places no Meta or Google pixels on its sites. Events collected through Piano go to Snowflake, and the BBC decides which go back to the ad platforms through conversion APIs, stripping metadata such as URLs. “We choose what events we send to the advertising platforms. We choose when we send them,” said Agate. A high revenue-score event can now feed Meta’s and Google’s optimisation without a pixel, and the funnel is running at a positive ROI.

After the move to Stripe Billing, Smart Retries lifted payment recovery so sharply that Iles called the chart “completely bananas”. Before, the BBC recovered 40 to 60% of failed payments within two weeks; now, Iles said, it barely loses subscribers that way.

A third of churn is a billing failure nobody owns

Failed payments were the subject for Tim Rowell, Piano’s SVP Global Publishing, and Jana Aleksic, Stripe’s Customer Success Manager for Piano. Across Piano’s clients, 33% of churn is involuntary: expired cards, replaced cards, declines on security grounds. Even after grace periods, 2.4% of monthly subscribers are lost to payment failure each cycle, roughly a quarter of the cohort a year.

Rowell blamed how publishers are organised. “Acquisition is the primary focus. That’s what receives the budget. That’s what’s reported on in the weekly meeting,” he said. Payments often sit in finance as a vendor relationship, with nobody owning the numbers day to day.

Aleksic argued that each decline reason needs its own response, with dunning emails reserved for cases where the customer must act. Billing, messaging and access must also move together, or a recovered subscriber still meets a paywall. She left four questions to put to payment providers:

● What are your authorisation rates, from first attempt to final, ideally by market?

● Why do renewals fail, and which causes can you influence?

● What share of initially failed payments do you recover?

● During recovery, do billing, messaging and access stay in step?

“I honestly don’t think many of our clients can answer these questions off the cuff,” Rowell conceded.

Piano’s own migration to Stripe Billing now covers about a million subscriptions, with one failure per 10,000 switches. Early data from the first clients moved shows the share of failed renewals recovered rising from 15.4% to 55.1%.

When everyone buys the same AI, the input data is the only edge

Janus de Visser, Founder of Cloud Nine Digital, took the argument into marketing automation. Google says 80% of its advertisers use one of its AI tools. Citing the IAB’s State of Data 2026 report, de Visser said half of respondents are already scaling AI in marketing measurement, yet 75% say the tools underperform on trust, rigour and efficiency. Tools such as Meta’s Advantage+ and Google’s AI Max take spending decisions inside systems no outsider can inspect. If competitors run the same black boxes, de Visser argued, the data going into them is the only lever left.

One client lost €5m in revenue over five days when one layer of its data set-up failed and the algorithm kept spending. A theme park’s campaign nosedived when add-to-cart tracking failed on a single mobile device and browser combination, removing 30% of those events.

“If you automate the decision, I believe you have to automate the validation of the data you send into the machine that makes the decision,” de Visser said. He dismissed the one-off audit as “the illusion of the truth”, since a set-up can look different the day after.

None of the speakers said who inside a publisher should own segment templates, renewal recovery or tag health..

TikTok used Advertising Week New York to unveil a set of AI advertising and commerce tools. The headline consumer feature is Shopping Assistant, a conversational shopping agent. But the change with the widest industry reach is that advertisers can now run TikTok campaigns from inside AI tools such as Claude and Snowflake.

TikTok says more than 200 million people in Europe use the platform. It cites a study it commissioned from Public First, which put the economic value generated by businesses advertising on TikTok across the EU at €31bn in 2025, a figure that counts direct, indirect and induced output. In Kantar’s Media Reactions 2026 report, marketers ranked TikTok first among platforms where they plan to increase ad spend in 2027.

“AI is transforming how advertiser content is created and scaled,” said David Kaufman, TikTok’s Global Head of Monetisation Product Partnerships.

“With our suite of solutions, we are bringing together AI, automation, and creativity to help advertisers improve performance while reducing effort. We are enabling advertisers to create work that feels relevant, move at the speed of culture, and drive meaningful business impact.”

Campaigns from inside AI assistants

TikTok for Business MCP, built on the open Model Context Protocol, lets developers build AI agents and tools that plan, launch and manage TikTok campaigns. TikTok lists it as a connector on Claude, Kimi, Manus, Perplexity, Replit, Snowflake, Tencent WorkBuddy and the IAB Tech Lab Agent Registry. A buyer working in a general-purpose assistant, or an analyst working in Snowflake, can set up and run a TikTok campaign without opening Ads Manager.

The MCP was launched at TikTok World in May, and an Agentic Hub marketplace of ready-made Skills built on it followed at the end of June, with HubSpot, Wix, Innovid and Kochava among the early developers. Google, Meta, Pinterest and Snap now offer MCP access of their own. TikTok says the number of advertisers using its MCP to activate and manage campaigns has risen by more than 200% since launch. It did not give a baseline.

Search, creative and shopping

Search Ads Campaign, powered by Smart+, brings TikTok’s automation to search, handling keywords, creative, bidding and budget across queries on the platform. TikTok frames this as capturing intent from users who search there for products and ideas, intent that advertisers currently pay Google to reach.

Smart+ Creative Selection pulls advertisers’ own assets, creative generated by TikTok’s Symphony tools and creator content into a single approval workflow, where advertisers decide what runs. GMV Max, the automated TikTok Shop ad product, will start factoring affiliate commissions, coupons and referral fees into its optimisation, so sellers can bid against return after costs.

Shopping Assistant sits across several TikTok touchpoints and gives real-time guidance on product details, shipping, sizing, availability and purchasing within one conversation.

Two further products announced in New York will not launch in Europe for now. Buy Direct is a one-click checkout that lets users buy from a brand without leaving the For You feed, built with partners including Salesforce, Shopify, Shoplazza and Stripe. Agentic Leads uses AI-guided conversations to capture and qualify prospects, both in TikTok direct messages and in lead forms on advertisers’ own sites. TikTok gave no European timetable for either.

SAP’s headline from its 2026 Global Customer Loyalty Index is that rewards-based loyalty has fallen below 40% for the first time in the survey’s six years, from 48% to 38%.

Part of that fall may reflect a change of sample. Last year’s survey covered the US, UK, Germany, Australia and Taiwan; this year’s 10,002 respondents came from the US, UK, Germany and Hong Kong. SAP does not publish the rewards figure by country, so the effect cannot be checked, but in the three markets surveyed in both years overall loyalty barely moved, rising two points in the UK to 69% and falling two in the US to 71% and one in Germany to 65%.

The report, published last month by SAP Engagement Cloud (formerly Emarsys) and fielded by Opinion Matters, is more useful on a question that doesn't depend on last year's sample: who is making the recommendation. Some 35% of consumers say they prefer AI recommendations to those of friends and family. In the US, according to SAP’s announcement, the figure is 44%.

A companion survey of 2,295 enterprise technology buyers, the Global B2B Buyer Loyalty Index, finds that 69% would switch supplier if an AI agent found a better option.

Shoppers are letting AI do the recommending

Across all consumers surveyed, 38% use AI to help make purchase decisions and 33% have bought a product an AI recommended. For 15%, a suggestion from a chatbot such as ChatGPT or Gemini has made them more loyal to a brand. Fewer now say they show loyalty by recommending a brand to friends and family: 42%, down from 48%, though that comparison carries the same sample caveat. SAP’s “silent loyalty” category, customers who stay with a brand they would not endorse publicly, covers 58%, the highest share in the series, though on the new sample.

Paying for subscriptions, which SAP counts as a loyalty signal, rose from 14% to 17%, and engaging with brand content from 17% to 21%, subject to the same caveat. The report advises consumer brands to cultivate both, which sets them chasing the habits subscription publishers have spent a decade trying to build.

SAP also reports that 78% of consumers who use AI to shop consider themselves loyal to brands, against 69% overall, and presents this as evidence that AI strengthens loyalty. AI shoppers are also more than twice as likely to describe their loyalty as trend-driven (21% against 9%). People who shop with AI may simply be keener shoppers to begin with, and the survey has no way of telling the two apart.

Consumers want AI to shortlist and a human to call when it goes wrong

On SAP’s “AI control spectrum”, only 4% of consumers would accept fully autonomous purchasing, while 60% want shared control with the AI, 16% would let it lead under supervision and 20% want to keep full control themselves. More than half (56%) want to be able to reverse an AI’s decision, and 48% want a human to be available, although 50% would rather a brand simply got it right without their involvement.

Gen Z is the most willing to delegate: 44% would let an AI assistant buy for them. It is also the quickest to walk away: 45% have dropped a brand over AI, against 30% of all consumers and 23% of Gen X.

Why is less clear, because SAP explains that 45% in three different ways. Its key findings put it down to “their use of AI”, suggesting AI steered them towards other brands. Its generations page cites “better AI practices” by rivals. Its press release blames brands’ own use of the technology.

Mark Ritson, founder of the MiniMBA, a former marketing professor and a columnist for The Drum and AdWeek, took the view in SAP’s announcement, saying the cohort is “proving to be the least forgiving when brands get it wrong”.

Enterprise buyers would let an agent end a supplier relationship

In the B2B survey, 77% of buyers consider themselves loyal, up from 71%, with the UK rising furthest, from 69% to 78%. That sample changed too, from 3,008 buyers in five markets to 2,295 in four. Much of the loyalty looks like inertia: 64% agree that switching supplier is too complex and expensive.

Yet 69% would switch if an AI agent found a better option, and 71% would trust AI to choose the most popular option over automatically renewing with an incumbent. Some 78% are comfortable letting AI handle routine purchases on their behalf, against 33% of consumers who trust AI to complete a purchase. Business buyers’ most-cited condition is a human fallback, required by 34%, and 72% expect discounts to matter less as agents take on more of the buying.

AI is becoming the first stop for supplier research

Among the enterprise buyers surveyed, 71% use AI to research or shortlist vendors, and 30% say an independent AI system was their most common source for researching suppliers this year. Some 88% trust AI-generated supplier recommendations, although only 37% trust them completely. That leaves most buyers with reason to check the AI’s answer, and an opening for a trade title they trust.

Subscribe now

Supplier research is what B2B publishers sell to technology vendors: an audience of buyers who are still deciding, reached through advertising and events. If nearly a third of those buyers now rely mainly on an AI system, vendor budgets are likely to follow them there. SAP is already telling suppliers to make sure the material those systems read about them is accurate and consistent. Tom Goodwin, a marketing expert and author, quoted in the report, argues that brands “become more important” as AI takes on purchasing decisions, because reputation is what makes a recommendation credible. Trade publishers have that reputation. Few have found a way to be paid for it when the reader is an AI system.

The Advertising Association’s LEAD conference will return to the Queen Elizabeth II (QEII) Centre in Westminster on 28 January 2027, run jointly with the Institute of Practitioners in Advertising (IPA) and ISBA, the body for UK advertisers, under the theme “Shaping advertising’s future”. The venue is a few hundred yards from Parliament, and LEAD is one of the main occasions on which the industry makes its case to policymakers.

Confirmed speakers include:

• Tim Davie, Co-Chair, Creative Industries Council

• Baroness Hazarika MBE, Times Radio Broadcaster

• Andria Vidler, President, Advertising Association

• Charlie Rudd, CEO, UK Creative Practice, Publicis

• Damon Reeve, CEO, Ozone

• Kate Rowlinson, CEO, WPP Media

The event will be hosted by Nina Hossain, Broadcaster and Journalist.

Ozone tops a sponsor list that includes Google and Meta

Ozone is headline partner, with Google, News UK, Meta and TikTok at gold level. Ozone, which is owned by UK publishers, competes for advertising budgets with Google, Meta and TikTok, which between them take most of the UK’s digital advertising spend. Its Chief Executive, Damon Reeve, is among the speakers.

Woodford sets a broad theme

Stephen Woodford, the AA’s Chief Executive, framed the event around pressures from every direction:

He added: “We want to look ahead, ask the difficult questions and equip our industry with the ideas and insight it needs to move forward.”

The release does not specify which questions. Likely candidates include AI answer engines diverting traffic from publishers, the competition regime’s treatment of the platforms and the cost of further advertising restrictions, each a live issue for at least some of the event’s sponsors.

Future Leaders joins LEAD at the QEII

Future Leaders will take place in the same venue immediately afterwards. It is aimed at professionals with four or more years’ experience across agencies, brands, media owners and technology companies. Woodford described it as a space “for the next generation of talent who will be shaping the industry in the years to come”.

Tickets for both events are on sale, with discounts for AA, IPA and ISBA members. You can buy tickets here for LEAD 2027 and Future Leaders.

No government speaker has been announced yet. The first line-up is drawn entirely from the industry.

RTL AdAlliance, the international sales arm of RTL Group, has opened an office in Spain. The announcement, made in Madrid and Luxembourg on 29 September, pitches the move as a way for Spanish advertisers and agencies to buy premium video across Europe through a single point of access, combining linear TV, broadcaster video-on-demand and connected TV under what the company calls its Total Video portfolio.

The inventory is broad. It includes broadcasters such as RTL, ITV, M6 and RAI; BVOD platforms including France.tv, RTL+ and Videoland; and CTV supply from HBO Max, Samsung TV Plus and Waipu TV. The company claims more than 150m daily TV viewers, 37m addressable TV households and over 4bn online video ad impressions a month across its network.

Atresmedia is helping a foreign seller into its own market

The launch is, in the release’s words, “closely coordinated with and supported by” Atresmedia Publicidad, sales house of one of Spain’s two dominant commercial broadcasters. Atresmedia already sits in the RTL AdAlliance portfolio, which sells its inventory to international advertisers. What is new is the reverse flow: Atresmedia helping an outside sales house sell other broadcasters’ airtime to Spanish buyers.

On its face, this is odd. National sales houses guard their domestic agency relationships jealously. The logic becomes clearer once you ask where Spanish money goes when a brand wants to reach France, Germany and Italy at once. For most, the answer is Google, Meta or Amazon, which offer one contract and a reach claim that ignores borders. Stéphane Coruble, RTL AdAlliance’s Chief Executive, put the problem bluntly to VideoWeek in January: “When you see that 85 to 90 per cent of the business is going to just a few players, you need to act.” A Spanish brand buying ITV or M6 through a broadcaster consortium is a brand not spending that budget on YouTube. Atresmedia can live with that.

José Miguel García-Gasco Martínez, General Director of Atresmedia Publicidad, describes the arrangement as “one access point, premium reach, and greater efficiency, while reinforcing Atresmedia’s role as the trusted local partner.” The final clause matters most to Madrid. Atresmedia is lending its relationships on the understanding that it remains the gatekeeper.

The Spanish numbers explain the timing

Daniel Bischoff, RTL AdAlliance’s Chief Operating Officer, frames the opportunity in two halves: “Connected TV ad spends are growing rapidly, while broadcast TV remains an integral provider for premium reach.”

Both halves hold, though the proportions matter. IAB Spain puts CTV advertising at just over €174m in 2025, up 48.4% on the year and more than double its 2023 level. InfoAdex put television investment at €1.78bn, down 4.4%, yet still the largest medium with a 28.5% share of controlled spend. CTV is growing fast from a small base; linear is shrinking slowly from a large one. A sales house that bundles both, and that already carries Atresmedia’s inventory abroad, is hedged against either trend.

Europe’s broadcasters are sorting themselves into blocs

The Spain launch fits a wider pattern. RTL AdAlliance added France Télévisions and Austria’s ORF to its roster during 2025, and RTL Group is buying Sky Deutschland. On the other side, MFE-MediaForEurope, the Berlusconi-controlled owner of Mediaset España and Mediaset Italia, took control of ProSiebenSat.1 last year and is assembling its own cross-border broadcaster group.

That leaves Spain’s two commercial broadcasting heavyweights in different camps: Atresmedia aligned with RTL’s sales alliance, Mediaset España inside a rival pan-European group. The Madrid office is as much a competitive statement as a service to agencies, and Mediaset España’s sales team will read it that way.

The office looks light on local names

Artur Kobryn, RTL AdAlliance’s Vice-President of International Sales, will lead sales in Spain, alongside Mattia Badino, head of advertising; Michela Bruno, head of agency partnerships; and Simone Rondena, advertising manager for Southern Europe. The release names no Spain country manager, and says the company will keep working with its “established long-time partners and agents” in the market.

Read together, this suggests a lean presence that leans on Atresmedia’s local weight. That is a sensible way to test demand without high fixed costs. It also means the new office depends on a partner whose core interest is selling Spanish inventory, and whose goodwill is conditional on remaining the trusted local partner.

The larger unanswered question is whether the demand exists in Madrid at all. Pan-European video campaigns are mostly bought by multinationals through the global agency groups, whose trading decisions tend to be made in London, Paris or New York. Spain has export-minded brands in fashion, food and tourism that ought to want European reach. Whether they will route that spend through a Madrid desk rather than a central holding-company deal, or simply keep buying from the platforms, is something the office will discover quickly. The press release does not attempt to answer it.

A typical B2B purchase now drags on for 272 days and pulls in 22 stakeholders. Many of those people never speak to a salesperson, and many sit well outside the department that will use the product.

The recently published report, How to build more Buyable brands, builds on work LinkedIn has done with Bain & Company on what it calls Buyability. The idea is that a buying group will only commit when its members feel confident, they can justify the choice. The new study tests that idea against 700 B2B campaigns.

Procurement and finance have as much say as IT

LinkedIn and Bain surveyed 515 technology buyers in software, banking, manufacturing, healthcare, logistics, energy and retail, and asked how influence over a purchase was shared. Specialists from the team that would use the product, whom the report calls target buyers, held 51%. People from procurement, finance and legal, the “hidden buyers”, held 49%.

Yet most B2B marketing is aimed at the first group. Lead-generation programmes target the IT director or head of operations and nurture them toward a sale. The finance director, who can kill the deal, rarely hears from them at all. The report cites 6sense data showing that fewer than 1% of leads make it from the top of the funnel to the bottom, partly because buyers are usually more than halfway through their process before they contact a vendor.

It doesn’t help that suppliers are getting harder to tell apart. Dentsu’s Superpowers Index found that the perceived gap between winning and losing vendors shrank by two-thirds between 2021 and 2025.

Buyers want to be able to defend the decision

LinkedIn and Bain analysed more than 1,000 interviews with B2B buyers, then surveyed 750 buyers to find out what made them feel confident about a purchase. The top answer, at 34%, was that they could defend the decision if it went wrong. Confidence that the product would work came a whisker behind, at 33%, followed by managing the downsides (29%), agreement within the group (28%) and an easy decision (21%).

Buyers are also watching their own backs. Two-thirds said they would prefer “products that provide peace of mind without career advancement” over choices that offer “business growth but increased career uncertainty”.

Jann Schwartz, Senior Director of Marketplace Innovation and Strategy at LinkedIn, is quoted in the report: “Across all the different individuals and functions that comprise a buyer group, avoiding the downside of failure is more important than enjoying the upside of success in a B2B purchase.”

Past experience and recommendations count for more than price

The research also looked at what tips the balance when two vendors seem equally good. For 37% of buyers, it was positive first-hand experience of the supplier. For 25%, it was a recommendation from someone they trusted. Product performance and price were decisive for only 8% each.

Negative word of mouth hits even harder. A colleague with first-hand experience who would not recommend a vendor was more than three times as influential as price. Video featuring “a customer similar to us” was four times as persuasive as video featuring “a market leader”.

And it pays to be known by the whole group from the start. In LinkedIn and Bain’s survey, 81% of buyers said everyone or almost everyone in the group knew the brand they eventually bought when the process began. Only 4% bought something that just the recommending department had heard of. Separate 6sense figures cited in the report show that 95% of contracts go to a vendor that was on the buyer’s first shortlist.

6sense found that 94% of B2B buyers used large language models such as ChatGPT during their buying journey in 2025. The report argues that the things that reassure buyers, such as customer stories, peer endorsements and press coverage, also help brands appear in AI-generated shortlists.

LinkedIn groups its findings into three areas, which it calls the 3Rs: Recommendations, Relationships and Relatability.

“Buyability gives us a seven signal framework that marketers can use to strategically enhance campaign outcomes because it is built on insights into what helps buying groups feel confident to buy. The campaigns that win are the ones that make buying decisions easy to defend by focusing on areas that customers can relate to and experiences and voices they can trust.”

Mimi Turner, Head of Marketplace Innovation at LinkedIn

Most campaigns use only one or two signals

To test the model, WARC and LIONS Advisory reviewed 700 B2B campaigns from 2010 to 2025, including entries and winners at Cannes Lions, the Effies and the WARC Effectiveness Awards. Each was checked for seven signals linked to the 3Rs: customer testimonials, backing from industry experts or peers, backing from celebrities or influencers, long-term relationships, category leadership, customer insight and the brand’s working style.

The average campaign used 1.6 of the seven. Nearly a third (30%) used just one, and 22% used none. Even campaigns with budgets above US$1m averaged only 2.25. The biggest difference between campaigns with three or more signals and those with fewer was the use of customer testimonials: 64% against 16%.

Imaad Ahmed, Thought Leadership Director at LIONS Advisory and WARC, says:

“As we found, intentional use of Buyability signals is still an opportunity for B2B marketers. When well-stacked within a campaign, Buyability signals can make revenue, ROI and brand health uplifts far more likely.”

Campaigns with three or more signals were 63% more likely to report higher ROI, 2.1 times as likely to report extra revenue and 91% more likely to report gains in consideration, preference and purchase intent.

The commercial results are an early read

To its credit, the report is upfront about the limits of its data. The numbers behind the headline ratios are relatively small: 13% of campaigns with three or more signals reported an ROI uplift, compared with 8% of the rest. For incremental revenue, the figures were 23% and 11%. The authors point out that award entries are written to make a case, that the sample only covers campaigns already seen as successful, and that many entries reported little hard commercial data. They describe the commercial findings as preliminary.

There’s also a gap between how the research was announced and what the report says. The press release says B2B marketing now needs to prove that it directly generates sales. The report finds that campaigns with more signals did no better than others at turning existing demand into sales. Their advantage showed up in longer-term measures, such as brand fame and customer affinity. My read is that Buyability is mostly a brand-building tool, and brand-building takes time to show up in sales.

The buyer research is on firmer ground. B2B purchases are group decisions, and the people in the group want to be covered if things go wrong. Most B2B marketing, and the way it gets measured, hasn’t caught up with that yet.

Steve Davies knew a story was coming. When he spoke to The Media Stack, the Chief Executive of the Advertising Producers Association had been told that news about one agency group’s in-house ambitions was imminent. A journalist had it. He could not discuss it. Within the week Campaign published extracts from internal WPP Production guidance telling staff to “actively convince” clients to send all their production work to WPP and, where a client insisted on three bids, to consider supplying all three from inside the group. The stated goal was to avoid traditional triple bidding by “positioning WPP Production as the default, trusted partner”.

The APA called it “a brazen attempt to pull the wool over their clients’ eyes”. WPP said the extracts were selectively quoted from a comprehensive document and that its process is completely transparent. Within a fortnight, Campaign reported that the document had been taken down. WPP declined to comment on its removal. The argument is still running, and Davies had made his side of it, at length, before there was a document to point at.

“The same person turning up in a different outfit every day”

Davies runs a trade body for around 250 independent production, post, editing, music and sound companies, and he doesn't ask for protection from the market. “Red hot competition is just part of the world we live in, so we can’t complain about that,” he said. When an agency takes a script to three production companies and the directors it wants, they compete on treatment and price, and the client can go back to its favourite and ask it to match the cheapest. “The client’s got all the power. I don’t think clients ever need to worry about value. The market delivers value.”

What he objects to is distortion. Agencies are entitled to build production units, he said, but an in-house unit that hires three freelance directors, assembles three treatments and presents the client with a choice is not running a competition. “That’s not an open market competition. It’s the same person turning up in a different outfit every day, saying you can have me, or me, or me.” A genuine triple bid involves three companies with no connection to each other. He has been saying so since the IPA’s 2024 client research; the leaked document briefly put it in writing.

The withdrawn guidance went further. According to Campaign, it proposed that every production below €500,000 go directly to WPP Production, and that where a client still insisted on an external triple bid, staff should either take part with safeguards, presenting WPP’s proposal before outside companies were briefed, or stay out and work on the client’s procurement team instead. Campaign also reported unease among WPP’s own staff at being asked to abandon trusted third parties.

WPP disputes the reading. James Brook-Partridge, Head of Production at WPP Production, said there are no mandates to internalise work and that the driver is volume: clients make three times as much content as three years ago and may need five to twenty times as much within two, on the same budgets. The IPA’s Eliot Liss said it was untrue that agencies pass off internal bids as independent ones.

The order comes from the finance director, not the creative floor

Why now? “They often don’t have clients on retainer any more,” Davies said, “and in the case of some of the groups, they have a precipitous decline in their share value, so they’re doing anything and everything to keep revenue within the door.” WPP, on its figures, has fallen from £18bn to £4bn. It exited the FTSE 100 in December and in February reorganised into four units, one of them WPP Production.

“Professionalism is putting your clients’ interests first and before your own, and this isn’t that,” he said. “It’s putting your own short-term interests first.” A client denied an open market will eventually notice, he argues, and go elsewhere. “But the reality is that some of these groups are so desperate there is no long run.”

The pressure, he said, doesn't come from the people who do the work. “The production and creative departments want to work with the independent sector, because they know that’s where the talent lies. It’s usually the C-suite, simply the Chief Financial Officer, who makes rules that you must do a certain amount of work in-house.” Creatives who push back, he said, are told to comply. The best directors, editors and producers stay in their own companies because they want a variety of work. “They don’t just want to work inside some sort of monolith for one or two clients.”

Half the money never touches an agency, let alone a production company

In-housing is one reason scripts have stopped arriving at independent companies. It is not, Davies said, the largest. “Every year for the last few years the ad spend in the UK has been up significantly, and yet nobody within a production company or indeed an agency has noticed that, because more than half of ad spend now goes directly to Meta, to Google or to Amazon without touching an agency or production company.” In-housing bites hardest in the middle of the market; the media shift does the rest.

Subscribe now

He rejects the IPA’s 2024 explanation that agency studios grew because the traditional production market was under-serving the explosion in digital work. “The independent market has always had the capacity to make all the work that they can get their hands on. I don’t think there’s any evidence to suggest that it was a reaction to a lack of supply. It was a reaction to the agencies’ own problems.”

AI gets a film to 85%. The last 15% can cost more than a shoot

“AI can often make something that’s 85% good as a film extremely quickly, in a way that just wasn’t possible before,” Davies said. “When you look at it, you think, wow, that’s astonishing. But getting it from 85 to 100% can often take more money and be more difficult than if you shot the thing from scratch.”

He saw a set of AI awards recently. “On the face of it, the work was astonishing, but none of it was capable of evoking any human emotion.” Clients still arrive at production companies asking to make “something in AI”, often, he suspects, because a marketing director needs to look current. One producer told him his answer is that he will decide how far AI helps once he knows what he is making. “It’s just another production tool, really.”

Life beyond the TV script

Asked how many production companies the industry can support, Davies would not put a number on it. “There are probably fewer production companies that can exist on TV scripts. But there are plenty of companies within the APA now who never make a television commercial. They’re nimble. They get out there and make things constantly for Instagram, other online content, and that world is expanding. Our capacity as humans to view more audiovisual media seems to grow every year.”

The other opening is that clients no longer route everything through an agency. “We had one model in the past: client to agency to production company. We’re not moving to a second model. We’re moving to a thousand different answers.” Some clients work direct; some keep an agency but insist on a genuine triple bid; some don’t care much as long as the work gets made.

Davies has run the APA since 4 June 2000, a date he now quotes to members who were not born then. A former law-firm partner, he still reviews members’ contracts himself. His public line on the WPP document was that it threatened the independent sector only if clients failed to see through it, and that he thought they would. The document has now gone, without explanation, and WPP Production opens its east London flagship this week, built for the content volumes Brook-Partridge says clients now need. Whether the three bids for that content come from three companies or one is still a decision for the finance and procurement departments that, on Davies’ telling, have been steering the work in-house.

This is a guest post by Steven Wilson Beales, an SEO and Content strategy consultant.

Earlier this month I ran a piece of research to see just how valuable an ‘About’ page was for news publishers when it came to AI visibility - you know, that page hidden down in the website footer that only the most curious ever read.

To investigate this, I tested 201 questions across ChatGPT, Copilot, Gemini and Google AI Mode. The final dataset contained 4,003 answers and 20,016 citations, collected between 2 and 6 September 2026. The questions covered four broad themes: trust, subscription value, specialist authority and publisher identity.

The clearest finding was that the About page is only one part of the picture. In 76% of publisher cases, no About page appeared in the cited evidence at all. Trust questions drew heavily on journalism, standards and accountability material. Value-for-money questions brought subscription and pricing pages into play. Specialist authority was demonstrated largely through journalism itself, while identity questions were much more likely to surface the About page.

This is why I think AI visibility for publishers is better understood as an evidence architecture problem than an About-page problem.

In about half of the cases, AI cited only one page from a publisher. But when two or more pages were cited, they were different types of content in roughly two-thirds of cases. A news article might appear alongside a subscription page, standards document or corporate page.

The question publishers should therefore be asking is not simply “Which page should we optimise?” but “What evidence would an AI need to support the things we want audiences to believe about us, and where does that evidence live?”

Trust needs evidence, not just claims

Trust produced one of the clearest patterns. Editorial articles represented 55.8% of trust citations, while ethics and standards pages accounted for 9.8%. About pages accounted for 11.8%.

Reuters is a good example. When asked which US news brands had the strongest reputation for accuracy and reliability, the evidence included a standards page, an article explaining visual verification and a piece involving editor-in-chief Alessandra Galloni. Its general About page did not appear.

Saying “we are trustworthy” is a claim. Showing how information is verified, how errors are corrected, how complaints are handled and how editorial independence is protected is evidence.

Publishers should therefore think of corrections pages, standards documents, fact-checking policies and explanations of reporting methods as trust assets rather than administrative housekeeping. A static sentence saying that you correct mistakes is very different from maintaining a visible record showing that you actually do it.

Subscription pages are part of the brand story

The evidence changed significantly when models were asked whether a publisher was worth paying for. Subscription and pricing pages accounted for 33.7% of value citations, making them the largest individual page category.

The Guardian illustrates how broad this evidence can become. When asked why someone should pay the publisher every month, ChatGPT cited material covering governance, pricing, mission, financial results, the Scott Trust, and reader revenues. Mediapart produced a similar mixture, including an activity report, its subscription portal, a help-centre pricing article, its About page and editorials about independence.

A subscription page is therefore doing more than trying to convert somebody who has already decided to buy. It may also become one of the main pieces of evidence an AI system uses when deciding whether a publisher represents good value.

That means these pages should explain more than the price. They should clarify what the reader gets, what is distinctive, what different tiers include, what reader funding enables, and why the product is worth paying for.

Help centres deserve attention too. Some of the information shaping how AI describes a publishing product may live inside customer-support software that editorial and SEO teams rarely examine. These pages need to be crawlable, current and consistent with the proposition being communicated elsewhere.

Specialist authority still comes from the actual journalism

There appears to be no shortcut to subject authority. Editorial articles accounted for 40.5% of citations in specialist questions. A further 41.5% was unclassified, but when I checked a sample of those pages, 79% of the ones I could confidently classify were ordinary editorial articles. The real role of journalism is therefore probably higher than the headline figure suggests.

If a publisher wants to be considered authoritative in personal finance, football, travel, politics or product reviews, it ultimately needs to publish strong journalism in those areas.

Methodology can strengthen that evidence. IndyBest, for example, has pages explaining how it tests products and who its experts are. In subjects such as health, science, finance, reviews or investigations, explaining how journalism is produced can provide useful supporting evidence for why it deserves to be trusted.

But there is no evidence here that publishers can simply add a handful of authority pages to thousands of pieces of commodity content and expect AI systems to treat them as specialists. The journalism itself still has to do the work.

The About page still matters, just not for everything

None of this means About pages is unimportant. About and mission pages represented 24.9% of identity citations, and nearly a quarter of identity answers relied on an About page alone.

They remain particularly useful when AI is trying to establish who a publisher is, what it stands for and what makes it different. The mistake is expecting the same page to demonstrate subscription value, journalistic trustworthiness and specialist authority at the same time.

Shorter questions make this even more interesting. Compared with longer questions, models cited less journalism (32% vs 42%) and more pricing pages and homepages. They also stopped after one page nearly six times in ten. Surprisingly, About pages were cited less frequently on these shorter queries.

Axios provides a useful example. For the four-word query “what is axios known for”, ChatGPT cited seven pages across four different types. The largest source of evidence was not the newsroom but the help centre, including pages explaining what Axios is, what it covers and how its product works. Across repeated runs, help-centre pages appeared more often than the About page.

Subscribe now

Think about the evidence behind the claims

The practical lesson is to identify the claims you want AI systems and audiences to make about your brand, then make sure the supporting evidence exists.

If you want to be considered trustworthy, can somebody easily find your editorial standards, corrections, complaints process, fact-checking approach and evidence of independence? If you want to be considered worth paying for, is it obvious what subscribers receive, why the product costs what it does, what reader revenue enables and what makes the proposition distinctive? If you want to be seen as a specialist, is there sustained high-quality journalism, visible expertise and clear methodology where it matters? And if you want to be described accurately, can an AI system easily establish who you are, who owns you, how you are funded and what makes you different?

Then there is another question that may be even more important: are all of those pages telling the same story?

If an About page says independence is fundamental, the subscription page says reader revenues protect that independence, the standards page explains how editorial independence is maintained and the journalism demonstrates it in practice, the publisher has not simply repeated one claim four times. It has created four different forms of evidence supporting the same proposition.

This also means AI visibility is not purely a newsroom SEO job. Important evidence may sit on subscriptions, support, marketing or corporate subdomains owned by entirely different teams.

In summary then, AI does not appear to form one fixed view of a publisher from one definitive page. When the question is trust, it looks towards journalism, standards and accountability. When the question is value, subscription and pricing evidence matters more. When the question is specialist authority, journalism itself is central. And when the question is identity, the About page finally earns its money.

AI visibility for publishers therefore looks less like optimising a couple of corporate pages and more like building a coherent evidence architecture across journalism, commercial pages, corporate information and accountability material.

There is no quick fix. The stronger strategy is to make sure the evidence supporting what you want audiences to believe about your brand exists, is discoverable, and tells a consistent story.

And that’s even before we touch on third-party citations, which is a whole different story.

To read the full report, visit:

Publishing Strategies

Which publisher pages does AI cite when judging a news brand?

If you’re scanning this article and thinking ‘wow, someone’s gone down a real rabbit hole here’, you’d be correct. I have. 😭…

Read more

2 months ago · 2 likes · 2 comments · Steven Wilson-Beales

If you like to reach out and discuss further, please contact Steve via:

https://www.stevenwilsonbeales.com/

Last month, Article 50 of the EU Artificial Intelligence Act came into full force. For UK businesses, at a surface level, the Act does not apply because we are not subject to EU Law.

Indeed, the UK Advertising Standards Authority (ASA) has confirmed there’s no equivalent blanket UK requirement yet, so a UK-only brand isn’t caught by Article 50 at all.

So what’s the problem? If you are running an ad or marketing campaign aimed (partly or wholly) at EU consumers, then the ruling 100% applies to you. In short, you are duty-bound to abide by the Act.

What Article 50 Actually Requires

Article 50 is clear about the rules and guidelines marketers need to follow.

Firstly, if you use AI to generate images, audio, video of real people, places, or events, you now have to disclose it. This might include AI voiceovers, deepfake testimonials, AI footage of real locations and AI versions of true events – anything where AI is used to alter or compose text, imagery or video to make events look real.

Secondly, the disclosure you are required to implement needs to be visible, prominent and transparent. This means that a small disclaimer with a typographic point size of 3 hidden at the bottom right of an ad won’t cut it. Similarly, a verbal disclaimer on a video sped up X2 to get it over with will similarly fall foul of the EU AI Act.

There are simply no shortcuts or workarounds. The EU AI Act demands transparency. Concretely it requires:

• Video/audio: a visible on-screen label (e.g. ‘AI-generated’ or ‘AI voice’) appearing from the first frame or first seconds (not just in a ‘see more’ description or end card).

• Images: a visible watermark or caption on the image itself (not a note in the alt text or a linked policy page).

• Social ads: the label has to appear in the ad unit as the viewer scrolls past it (not require a click-through to a landing page or footer disclaimer).

  

What Marketers Must Do Immediately

Much has already been written about the EU AI Act in the marketing press, but as privacy and data protection professionals, we see a glaring gap. Indeed, this gap is not just limited to the EU AI Act, we come across it frequently in many areas of our work.

And what’s that? It’s not enough to know the compliance obligations needed to be legally watertight. Somebody has to take ownership of the issue and sign off all marcoms as a standard point of procedure.

Without such a procedure - implemented across the business - marketing campaigns, promos and voiceovers can slip through the cracks. And all it takes is just one campaign to be non-compliant and the ramifications could be severe.

How severe? Well, at this point, no organisation has been fined or held to account by the ruling. But it’s coming, and if Uber’s recent €825m fine is any guide, Brussels’ isn’t shy about big penalties once it starts enforcing.

Just as importantly, it will ensure that your own customer-base doesn’t catch you out. The damage to your own trust and reputation could be just as worse as any potential financial fallout.

Workflow, Workflow and…Workflow

To reiterate, it’s vital that someone in your organisation is tasked immediately with signing off any marketing campaigns that could target EU consumers.

This won’t change your legal obligations, nor will it indemnify any organisation who doesn’t abide by the law. But what it will do is ensure that no marketing campaign leaves the building without being signed off for compliance to Article 50 of the EU AI Act.

Creative teams, account staff, client brands can then all relax in the knowledge that there is a professional overseeing each and every campaign, and that nothing will leave the building without that person’s sign off.

With that in place, no UK company should be at risk of being hauled to Brussels to explain why an AI voiceover wasn’t labeled as such, or whether a person was real or not. And no UK company should then be at risk of incurring a fine that could wipe a year’s worth of profits at the stroke of a judge’s pen.

Get your house in order!

Andy Chesterman, Managing Director, Privacy Helper

About: Privacy Helper provides practical, no-nonsense data protection consultancy for UK and EU businesses. Acting as an outsourced Data Protection Officer, the firm helps companies manage GDPR and regulatory change in a way that supports operations rather than slowing them down. It focuses on real-world application rather than legal theory, working with boards and operational teams to identify risk, implement workable processes and give businesses the confidence to use data properly.

Jon Slade wrote the number down because he could not quite believe it. The Financial Times had only recently switched on an in-house tool to spot unauthorised access to FT.com, and in under eight hours it logged 1.23 million access requests from 681,000 unique IP addresses, every one traced back to a single trial account that costs £1 for a month. All were denied.

“That’s not human behaviour,” the FT’s Chief Executive told Press Gazette’s Future of Media Technology Conference in London on 10 September. “That is unauthorised crawling of our website.”

Slade, interviewed on stage by Press Gazette Editor-in-Chief, Dominic Ponsford, had been booked to explain record revenue and profit and the FT

The subscription model was always a bet on control

When Slade joined the FT in 2002, print circulation and advertising supplied about 90% of revenue. Today 80% comes from digital sources, and about half of the total comes from digital subscriptions. Group revenue reached a record £540m in 2024, and operating profit rose 41% to £42.2m. “It’s going to look good again this year,” he said. Some 3.5 million people now pay for FT Group journalism in one format or another.

Search referrals are falling across the industry, and Slade conceded that Google’s AI Overviews “impact us to a degree, but the majority of our traffic to our site, the majority of the monetisation, comes from having a direct relationship with our readers”. That was the design.

The internal scoreboard follows from that: engagement, a composite of recency, frequency and volume, sits at the top, then “recognised users” who have handed over an email address, marketing permission and consent. “If you asked me how many page views I had last month, I couldn’t tell you. If you asked me how many engaged subscribers we had, I wouldn’t tell you, but I know.”

The FT will sell to AI companies, provided it can leave

“We think we can be both the data and the destination in the AI world, and that those things can coexist,” Slade said. The destination half means spending on what a chat window cannot reproduce: data visualisation, video, the experience of the site and apps. The data half is the licensing: deals signed with OpenAI and Google that set terms of use, require the FT to be named as the source, and include an exit.

“If the data shows us that something’s wrong, it’s not working for us, we call it the big red button. We want to press that big red button and take ourselves out of any further live feed provision. We recognise that once the salt is in the soup, so to speak, it’s very difficult to remove that.”

The button has stayed unpressed; the deals are “so far, so good”. The option exists, he noted, only “because we’re doing it in a licence arrangement”. What the FT will not tolerate is “uncontrolled theft of our journalism”. Some of that is second-hand: other sites write up FT stories, fail to keep crawlers out, and what reaches the models is, in Slade’s phrase, a “third photocopy”.

The FT is not suing anyone. Its parent, Nikkei, has taken Perplexity to court in Tokyo alongside Asahi Shimbun. “We’ve chosen not to take that route,” Slade said. “We would rather take the route of presenting what we see as an opportunity for the person that’s accessing our website ... and try and structure the licence arrangement around that.”

Bespoke deals do not scale, even for the FT

Ponsford raised the fear of smaller publishers without the FT’s pull: that journalism becomes a “dark kitchen”, cooking for someone else to sell. “That’s not our plan,” Slade replied. His answer for everyone else is SPUR, the Standards for Publisher Usage Rights coalition the FT launched in February with the BBC, the Guardian, Sky News and the Telegraph. The FT’s own deals are his argument.

Will the buyers keep turning up? “If you’re going to go for a multi-trillion-dollar valuation in an IPO, you need to demonstrate that your customers are getting value from your product,” he said. If the data goes stale, “they’re going to trust it less. I think that’s a problem if you’re an AI developer.”

The under-35 target asks the FT to wait for its money

Press Gazette’s analysis of Ipsos iris data last month gave the FT the highest share of under-35 readers of any UK national newsbrand, at 27%. Slade said that tallies with the FT’s own figures. Behind it sits a cross-functional Future Audiences team and a target of 850,000 recognised readers under 35 by 2030, which he expects to hit early.

The adjustment was financial: the FT is “really, really good” at spending money to acquire subscribers, he said, on roughly a two-year payback. Here it had to “recognise that we’re not going to see that return back in this calendar year. So that was a bit of a cultural shift for us.”

The tactics include FT Explains video for social platforms, a free Alphaville Substack that is “pretty geeky” but growing “much faster than our traditional newsletters”, and FT Schools, free to under-18s whose school registers. Older readers tell Slade they started because a boss, a parent or a teacher told them to. “Those points of influence, of trust, have changed,” he said.

The FT can play both sides of the AI market because it has journalism that developers will pay for and engineers who can withhold it. Few publishers in the room have either. SPUR’s standards are meant to close that gap, and they are being drafted while the crawlers keep working. Slade supplied the caveat himself: once the salt is in the soup, it stays there.

Pinterest has appointed Fintan Gillespie as Managing Director for the UK and Ireland, filling a post vacant since Caroline Orange-Northey left the business in the summer. He starts on 18 September and will be based in London, reporting to the international function run by Greg Owens the VP of International Enterprise Sales.

Gillespie arrives from Snap, where he spent close to a decade, most recently as Global Director of Ad Partnerships. Prior to that position, he ran Snap’s UK sales business and its EMEA gaming vertical. Before Snap, he was an agency lead at Google in the UK. His brief at Pinterest is to focus on the local market, advertiser and agency relationships, and revenue growth in what the company describes as one of its most important international markets.

Pinterest’s second-quarter results showed revenue above $1.1bn, up 18% year on year, and 640 million monthly active users, up 11%, its twelfth consecutive quarter of record users. European revenue grew 12%. More than 80% of monthly users are outside North America, which supports investing in international sales leadership, and the company points to Levi’s, Primark and Samsung among the brands active on the platform in the UK. It also opened a new London office at Oxford Circus over the summer, which the Mayor of London, Sir Sadiq Khan, inaugurated.

“Pinterest has built a uniquely differentiated platform for users and advertisers alike, engineered to move people from inspiration to real-life action. What excites me most about this role is the opportunity to build on the strong foundations already in place in the UK and Ireland and work with the team, our agency partners and advertisers to help more brands connect with people on Pinterest in relevant and meaningful ways.”

Fintan Gillespie, Managing Director for the UK and Ireland

Outside work, Gillespie mentors and invests in startups, and once drove overland from Sydney to Dublin through 44 countries to raise money for mental health charities.

Subscribe now

Share The Media Stack

This session was delivered in Spanish, recorded by an AI recorder, and translated into English by AI. We have made every effort to ensure accuracy.

Chema Alonso’s first conference talk was in Barcelona, on Windows XP, in a suit that didn’t fit because he had packed neither a belt nor socks. On the 7th of September, he was back in the City, at CaixaForum for Media Party Barcelona, as Vice-President and Head of International Development at Cloudflare, the company he joined from Telefónica in August 2025. This time he spoke to journalists about who gets paid when machines, rather than readers, consume what they write.

By Alonso’s account, the network runs in more than 330 cities, fronts around a quarter of internet domains and a similar share of global traffic, and blocks some 250 billion attacks a day. Cloudflare’s own published figure is more than 20% of the web. Either way, few companies see more of the web, and this year that view showed automated traffic overtaking human traffic for the first time.

The traffic that paid for content has stopped arriving.

For two decades the web ran on an exchange: publishers gave crawlers their content, search engines sent back readers, and readers saw ads or bought subscriptions. AI platforms take the content and return agents, which do not click. “AI platforms are taking the data for agents that are not being redirected to websites where they don’t click or see ads,” Alonso said. The pact must be rewritten so that “valuable content continues to be valued and paid for.”

The obstacle is opacity: crawlers can index for search, feed inference, or model training, and many do not say which. Alonso put the share that declares no purpose at 38%, with 17% identifying as search. Cloudflare’s July report cut the data differently: 52% of crawler requests on its network are now for AI training, up from 22% in spring 2025, and more than a third come from mixed-use bots that blend search, agents and training. In a Europe that talks about data sovereignty, he noted, a publisher might reasonably want to know whether its pages are training a model it never agreed to feed.

Publishers have a tool for this and are afraid to use it. Among the 10,000 highest-traffic domains, the share whose robots.txt disallows training crawlers is tiny. “People are afraid of blocking,” Alonso said, because the crawler they most need to block is also the one that keeps them in search.

Spain is 22nd, and Alonso treats that as a diagnosis.

He ran the numbers live on Cloudflare Radar, the company’s public traffic dashboard. Ranked by AI-related traffic, the United States leads, followed by Germany and the Netherlands. Spain sits at 22nd. “We are not even on page two,” he told the room, reading it as a measure of how little AI has penetrated Spanish companies and daily life. His own caveat applied: bots make thousands of requests per interaction, so traffic volume says more about machine activity than about the number of bots.

Radar also hosts Cloudflare’s directory of verified bots and their stated purposes. “If necessary, we call the bots by phone,” Alonso said.

Four announcements, and a default that flips on Tuesday

Cloudflare declared its first Content Independence Day on 1 July 2025, blocking AI training crawlers by default on new domains and launching Pay Per Crawl. The second, on 1 July this year, brought four changes.

The first is a demand for transparency. Crawlers are now classified as Search, Agent or Training, and site owners can allow, block or charge each class separately. Alonso listed what Cloudflare is asking of AI companies in return: honour a training block; offer a way to have content removed from a model; give URL-level visibility of whether a page was used for training, search or inference; publish metrics comparable to Search Console data; and guarantee that blocking training carries no search penalty.

The second is data for publishers: how much bot traffic reaches a site, which URLs are being collected, and a crawl-to-referral ratio showing how often a platform takes content against how often it sends a reader.

The third is a new payment model. Pay-per-crawl paid a publisher when a bot fetched a page. Cloudflare has now declared that insufficient and is moving to pay per use, in which the publisher is paid when content appears in an answer. Two partners are live. Ceramic.ai, the AI search company founded by Anna Patterson, pays for non-paywalled news each time it surfaces in results. You.com pays for paywalled premium content each time it is used in a response.

The fourth is efficiency. Cloudflare’s data shows more than half of AI crawl traffic re-fetches pages that have not changed, at the publisher’s expense. Alonso described a freshness signal that tells a crawler whether a page has changed, delivery of pages as Markdown rather than rendered HTML, and link maps built from real navigation. A pilot with OpenAI, announced in July, is defining APIs to carry site ratings, URL maps, freshness and rate of change; Cloudflare says it covers search only and shares no content for training.

Alonso did not dwell on the deadline. From Tuesday 15 September, Cloudflare’s defaults change: on pages that carry advertising, Training and Agent crawlers are blocked unless the site owner decides otherwise, and a crawler that mixes purposes is judged by its most restrictive behaviour. That covers new domains and free-tier customers, and it is why the industry has spent the summer arguing about Googlebot, which does not separate discovery from AI use.

Subscribe now

A network operator is writing the rules for a market it will sit inside.

Cloudflare’s position is that transparency creates scarcity, and scarcity gives publishers the leverage to strike licensing deals. Its own report counts more than 50 publisher-AI agreements since 2023 and concedes that licensing remains bespoke and will not replace the referral and advertising revenue it is meant to offset.

The company is also a vendor. Every tool Alonso described runs through Cloudflare’s network, and a market in machine-read content with Cloudflare as clearing house is a business, however sound the argument for it. Pay Per Crawl was the model of the future for exactly twelve months. The list of asks to AI platforms is a list of asks, and Google has spent a year declining the central one.

Alonso closed by saying Cloudflare would keep going “so that the inequality that is happening does not occur, where businesses based on content generation are disappearing or falling to historic lows.” On Tuesday, the defaults flip. What the largest crawler on the web does about it is the part of the pact Cloud

Nic Newman opened his Media Party session with a heatwave warning. During the worst of the summer, he had put a handful of questions to Claude. Back came a sourced Q&A, a temperature chart, an energy-mix chart by country and a list of the hottest spots, then the same again in Spanish. “It assembles this information very quickly. Assembles it essentially as a service,” the Reuters Institute’s senior research associate told the room. “From a publisher point of view, of course, this looks more like an existential threat.”

Later in the morning Dmitry Shishkin opened his session with a sentence. “It doesn’t really matter what you do with AI. If your content is crap to start with, then you are not going to succeed.”

Though the programme listed them as separate sessions, the two amounted to a single argument delivered in halves. Newman had the stats: where audiences and referrals have gone, and where publishers say they are placing their bets. Shishkin, the former BBC editor behind the user needs model, had the case that those bets cannot be paid out by newsrooms whose editorial, product and data systems do not talk to each other.

The referral era ended between two Novembers

Newman’s opening chart, aggregated across the 48 countries of the Digital News Report 2026, showed social media overtaking television as the most important source of news for the first time. He was more interested in a line lower down. “Access to news websites and apps are down 12 percentage points since 2021, as people find new and more convenient ways to access the news.” Across countries, 10 per cent now regularly use chatbots such as ChatGPT or Gemini for news.

The supply-side data was worse. Publishers surveyed for the Institute’s trends report at the start of the year expect to lose around half their traffic over the next five years as content is absorbed into AI interfaces. Chartbeat data covering some 2,500 sites shows Google referrals falling by about a third between November 2024 and November 2025, a decline Newman said was not universal and not wholly down to AI Overviews and AI Mode. “The key story is that publishers won’t be able to rely on referral traffic in the way that they have done in the past.”

Publishers have chosen distinctiveness. Shishkin asked whether they can measure it

Asked where they would put more and less effort, publishers gave Newman a consistent answer: more into original investigations, on-the-ground reporting, contextual framing, narrative stories and face-to-face events; less into Q&A, general news and some service journalism, which AI already does well enough.

Shishkin’s session took that strategy at its word and asked what it requires. His opening question to any client is whether they are “unique enough to actually justify people spending time with you”, and then he asks what would happen to their audience’s lives if the organisation disappeared tomorrow.

Every topic a publisher covers, he argued, splits three ways: commodity content everyone has, and which can be automated, expert content only that newsroom can produce, and services built around the topic so that readers can do something with it. “Content by itself is not going to be enough for people to keep coming back.” Most organisations cannot say how their own output divides. “If I were to ask a lot of media, they will not even be able to say what is the balance between our generalist content and expert content, for example. Nobody is tracking that.”

Liquid content needs solid building blocks

Newman’s examples of promising AI work all involved content coming apart and reassembling. Aftonbladet’s front-page chatbot prompts now draw 250,000 questions a week. The BBC, which has text for every Premier League club but could never staff a studio for each, is testing AI-generated daily audio briefings per club in the right regional accent. VG in Norway has built an experimental app for younger readers from its existing content, with no articles in it. “There are no editors, so all there is a system and monitoring after publication.”

The same survey found the industry unmoved by AI’s commercial effect. AI is embedded in research, headlines and workflow at almost every publisher, but heads of AI still decline to call it transformational. The vast majority said no jobs had been cut.

Shishkin’s diagnosis of that gap is structural. Newsrooms run newsletters, social, analytics, video and a CMS, and none of the data connects. “In order to be successful in the AI world, you need to create a shared taxonomy across everything that you do, you need to have one version of truth.”

In practice that means tagging every piece of content three ways. First, a detailed topic, not a single word like politics. Second, a format: interview, feature, write-up or Q&A. (Audio, video and text, in his scheme, are content types, not formats.) Third, the user need it serves. Work with smartocto has shown a personal finance story performing three times better as a listicle than as a write-up, a difference invisible unless format is tracked.

That tagging is what makes Newman’s liquid content intentional. Turning an updating story into an educating one is easy, because the archive supplies the material. Turning it into an inspiring one requires assets gathered on the ground, so the commissioning meeting has to know the permutations before the reporter leaves the building. “It’s not about creating better journalism anymore. It’s also about building systems that deliver the right journalism to the right person in the right form at the right moment.”

The agent will read paragraph five, if anyone tells it where to look

Newman’s next stage is agents working in the background on the reader’s behalf, and publishers expect them to do more damage than chatbots. Shishkin’s version is blunter. “You can be running a really important newsroom where important journalism is being done, and then everything will be completely flattened by AI agents that will come and pick something from you.”

His answer, in development since May, is an agent readiness layer: eight signals covering recency, provenance, attribution, sourcing, clarity and structured claims, so that content is retrievable, verifiable and comparable. His illustration was a ten-paragraph article visited by an agent comparing five sources. “Instead of them reading the whole article, they land on your page and you say, actually, your task means that paragraph number five is the most important paragraph for you to take.” At Monday’s workshop, participants scored a BBC News article and a Guardian article on the same story against the eight signals and agreed the Guardian did better. He offered the worksheet to anyone who asked. “Nobody else will solve it for us.”

Trust has moved to individuals, and Goalhanger has built a company on it

Newman’s second squeeze runs through people. Social media has stopped being social: a Financial Times chart shows only 17% of a Facebook feed coming from people the user knows. “These are content distribution networks designed to engage,” and the engagement is going to video networks and the creators on them. Respondents as a whole rate creators as more entertaining and easier to understand than journalists, and as less knowledgeable and less trustworthy. The quarter who actually use them rate creators better on every measure, trust included. “You see what you get. Whereas with news brands, there’s potentially hidden agendas.”

Publisher responses so far amount to making journalists more like creators, from Wired’s platform personalities to straight-to-camera video on the New York Times site. The model Newman singled out is Goalhanger, no longer in his description a podcast company. “It’s really about shows and talent.” Around the show sit audio, video, text, newsletters, social and increasingly television; around those sit membership, licensing, advertising, merchandise and ticket sales. Its South Bank event the weekend before the conference sold out.

In Shishkin’s terms that is a service built around content. It is also the one case where the taxonomy question barely arises: the relationship belongs to a person, and the format is the show.

On Monday, Media Party, the journalism and technology conference founded in Buenos Aires, opened its first European edition at CaixaForum Barcelona, with close to 600 registrations and a hackathon scheduled for day three. Marc Melillas, the Generalitat’s Director General of Media, used his welcome to note where the money goes: companies commit millions to AI while almost nothing goes into the ethics of how democracies use it. Mariano Blejman, the founder, listed what the format has produced, including GPTZero, presented by Edward Tian at the 2023 Chicago edition; it went on to raise around $15m. Media Party, Blejman said, is evaluating opening a space in Barcelona. The three keynotes that followed all turned on trust, and each speaker meant something different by it.

Verifying the output is the real cost

Teresa Mondría Terol of NPR worked inside the New York Times newsroom on the Epstein files and used that investigation as her anatomy lesson. Her test before any AI project has two parts: which stories become possible when time is no longer the limit, and where checking the output will cost more than the tool saves. “In the end it generates more information than you had at the start, and that doesn’t help us.”

Reconstructing the relationship between Steve Bannon and Jeffrey Epstein in the months before Epstein’s death meant finding their exchanges inside millions of documents, from flight records to iMessages. The two did not always call each other Steve and Jeffrey; initials and nicknames recur. An AI agent working in natural language found the first traces. Machine learning pulled every page on which either addressed the other, and generative AI extracted dates from messages in inconsistent formats. Every message was then verified one by one.

The same team described images and video frames to build a semantic search, so reporters could look for Epstein’s New Mexico ranch in words, and built agents that let journalists new to the story query the documents with citations back to the originals. For the second anniversary of 7 October, AI extracted everyone from or living in Gaza whom the paper had spoken to over two years; verification meant reading each source document in full and re-running the instructions to catch omissions.

The methods are not confined to investigations. The New York Times runs hundreds of hours of manosphere podcast audio through models to produce five-minute daily digests, so the reporters on that shift can follow the discourse without listening to all of it; the Wall Street Journal, she said, does the same. A local media group she cited transcribes and summarises recordings of public meetings no reporter attended. An open-source plugin now let’s newsrooms run document-by-document classification inside Google Sheets; she uses it herself, and at NPR is building internal search engines tailored to the production processes of individual podcasts. Her advice for newsrooms starting out: use it early in an investigation, when the job is locating the metal in the haystack, and in internal processes where mistakes surface sooner.

3Cat is chasing trust to wherever it settles next

Judith de Argila, Director of Digital Audience Development at 3Cat, Catalonia’s public broadcaster, described a news transformation begun from a position of strength. 3Cat leads television news in Catalonia. The result is 3Catinfo, a news brand built on integrating television, radio and digital production.

She named three problems.

Visibility: algorithms set the agenda, and the live questions include whether 3Cat’s content surfaces in ChatGPT and whether its app sits in Samsung’s marketplace.

Disconnection: journalism’s taste for nuance and data reads to many people as elitist. Trust: the hierarchy in which governments and media set the agenda has gone.

The responses: content made natively for social platforms; WhatsApp, Telegram and Instagram channels outside algorithmic feeds; distribution deals sought with Spotify and Samsung; an app rebuilt around scroll and vertical video; a verification unit whose head appears on bulletins explaining how stories were checked, with capsules pushed through WhatsApp to counter falsehoods on the same channel. The media, she added, are lobbying at European level for AI companies to prioritise their sources in answers. The 3Catinfo Instagram profile now reaches around three million people with news content alone.

The broader claim was that trust is being deposited elsewhere, with independent journalists, creators and individuals, and the next question is what happens when people ask ChatGPT, Gemini or Claude instead of a journalist.

PRISA bet on patience over speed

Olalla Novoa Ojea, PRISA Media’s Director of Artificial Intelligence, described a year spent on governance. When the work began, the group already had more than 30 AI projects catalogued, some in production, and a choice. “We could carry on doing pilot after pilot,” she said, “or pause momentarily and build the foundations that would let us run faster afterwards.” A group-level responsible-use policy, approved in December 2024 to follow the EU AI Act, went to every inbox; she doubted many read to the end.

An AI Oversight Committee meets every two weeks, with legal, compliance, privacy, technology, AI, sustainability and HR at the table, and reviews use cases from newsrooms and non-editorial teams alike. An executive committee led by the CEO meets every two months to set strategy and take escalations. Every proposal answers three questions: what problem, how, and why, the last covering alignment with editorial principles and audiences. Tools and use cases are approved as a pair; change either and the review restarts. Nothing reaches the committee without a manager’s budget. “No money, no project.” She had expected the fortnightly cycle to become a bottleneck. “Quite the opposite: it runs like clockwork.”

The instruments are deliberately plain. A use-case sheet, adapted from a technology template to a journalistic setting, makes teams write down what, how and why before they ask. An intranet AI hub, still in beta, holds the policies, training on demand, a prompt library and a community of practice where staff post the problems they hit. Monthly office hours and two ambassador networks carry questions back to the AI team.

A year in more than 1,000 staff trained, though that is 60% of those invited; 21 use cases approved; eight projects live or in development. Shadow IT persists, and PRISA chose training over blocking. Agent governance is the 2027 problem, with a first agent sheet drafted, alongside vibe coding by staff who have never programmed, which has to be squared with the group’s secure software protocol. Her close named the asset all this protects: 50 years of El País, more than a century of Cadena SER, more than 75 years of Caracol. Competitive advantage, she said, will not go to whoever deploys the newest model. “Trust is built with governance.”

PRISA’s oversight committee meets every fortnight. The audience de Argila described is moving its trust to creators and chatbots on a shorter cycle than that. The gap between those two clocks is where the rest of the programme sits.