OpenAI Supports Publishers Everywhere except the Balance Sheet
OpenAI came to WAN-IFRA's World News Media Congress with a polished defence of its publisher record. On what publishers get back, it had less to say.
The session had been billed as one of the most anticipated at the congress, and the shadow of the New York Times lawsuit, whose publisher, A.G. Sulzberger, had addressed the conference the previous day, hung over it throughout, even though both OpenAI speakers declined to discuss it directly.
Varun Shetty, Vice President of Media Partnerships at OpenAI, and Tom Rubin, the company’s Chief of Intellectual Property and Content, came to the congress with a coherent message: OpenAI is not a hostile actor towards publishers. It has invested in the relationship, is building responsibly, and wants credit for showing up in conversations that it implies other AI platforms are avoiding. Whether that message landed depends on how much weight you give to goodwill versus commercial terms.
The case for the defence
Shetty opened with a biography. Both he and Rubin have backgrounds in the news industry. Shetty spent time at the New York Times, and Rubin worked as a journalist before becoming a media lawyer and serving 12 years on the board of the Reporters Committee for Freedom of the Press. The subtext was clear: these are not Silicon Valley disruptors who regard journalism as an obstacle. They are, they argued, genuine believers in the mission.
The institutional commitments they cited were real. Through the WAN-IFRA partnership, OpenAI has supported more than 160 newsrooms through an accelerator programme. A new cohort of publishers has just entered a six-month focused development programme through the Lenfest Institute, which OpenAI is co-funding. These are not trivial investments, and Shetty was right that comparable commitments from other AI labs are harder to point to.
On training data, the issue on which publishers most disagree with OpenAI, Shetty was concise and unapologetic. OpenAI considers the use of publicly available content in AI training to be consistent with copyright law, as US courts have recognised. He acknowledged that publishers disagree. He declined to let that disagreement end the conversation.
What publishers are asking
The questions from moderator Ezra Eeman, AI in Media Lead at WAN-IFRA, pressed on the points that matter commercially. The value exchange question came up repeatedly: if not traffic and not direct financial deals, then what exactly? Shetty’s answer remained consistent: audience exposure, quality traffic that converts at higher rates, and the opportunity to build on OpenAI’s technology. He acknowledged that the answer sounds vague at the top level, and on the evidence offered it is.
On traffic specifically, Shetty offered something that landed between candid and convenient: anecdotal evidence from publisher partners suggests that the traffic OpenAI sends is higher-intent and more likely to result in subscriptions, even if the volume is lower than expected. He was explicit that more statistical work is needed to substantiate this. That publishers are being asked to trust anecdote while adjusting to structurally lower referral traffic is not a point the session dwelt on.
The search product itself came up repeatedly. Shetty pointed out that ChatGPT Search launched just over a year ago and that the current experience, while imperfect, is an early iteration. He described it as a product that highlights trusted journalism, cites sources, and provides click-through opportunities, while acknowledging that news search accounts for only a small part of how users interact with ChatGPT overall. The implication, present throughout, was that publishers should take a long view, at a moment when many of them cannot afford one.
The partnership model and who it reaches
One of the more pointed exchanges concerned who OpenAI’s existing deals reach. Most publishers in the room do not have a licensing agreement with the company. Shetty was straightforward about the logic: when launching a search product, you start in markets where you have the most users and secure relationships with a small number of trusted publishers to get it off the ground. You cannot start everywhere.
For publishers outside that initial cohort, the answer for now is primarily AI implementation work: helping newsrooms use the technology internally to improve editorial operations and audience products. Shetty placed considerable emphasis on this and cited several examples: publishers using AI to simplify and adapt content for different readers’ needs, to make archival material more accessible to journalists, and to structure complex data from public documents into reader-friendly formats. The pitch, essentially, is that publishers who are not yet in the search product can get real value from transforming their operations in the meantime.
Rubin added that OpenAI has active dialogue with a long list of publishers who are not formal partners, and that the door is open. When asked whether the partnerships team would grow to handle more of that demand, Shetty confirmed it would.
What was guarded
The session was not a cross-examination, as Eeman noted from the outset. Several questions went unanswered in any useful way. How much news-related activity is there on ChatGPT? Shetty said the volume is relatively small and that the company is working to release data in a form it can stand behind. Would OpenAI support emerging machine-readable standards for content permissions, beyond the existing robots.txt protocol? Something to discuss further. Publisher analytics dashboards? On the product team’s list.
Regarding advertising, Shetty confirmed that ads appear in the free ChatGPT product and are clearly distinguished from model responses. There is currently no arrangement for publisher content to appear alongside ads in a revenue-sharing model. When Eeman pressed on whether that constituted a fair value exchange, Shetty’s answer was brief: not at this point.
The Pulse footnote
A journalist in the room caught Shetty using the past tense to describe a product called Pulse, an AI-driven personalised digest that would surface content based on a user’s stated interests. Had it been discontinued? Not exactly, Shetty said. Pulse has always been compute-intensive and gated to a subset of users, and the company is currently making significant decisions about compute allocation. It was a minor exchange, but it carried a larger point: the gap between what OpenAI signals publicly and what it can deliver is a live issue, and publishers asking about it are right to do so.
The frame that didn’t shift
At the close, Eeman offered OpenAI the chance to make a concrete commitment for the year ahead. The responses were aspirational: Shetty wants to have evolved the search experience and be in a deeper conversation with publishers about technology adoption; Rubin wants to see meaningful innovation in reader experience. These are intentions, not commitments with a shape.
The session confirmed something the publishing industry has been navigating for three years. OpenAI is, by the standards of its category, a genuinely engaged interlocutor. It shows up, it listens, and it has put real money into newsroom support programmes. It also controls the architecture of a product on which publishers increasingly depend for audience discovery, and it has decided, so far, that the value exchange does not include revenue sharing or publisher analytics as standard. The relationship is real, but its terms still belong, for now, almost entirely to one side.
Another great piece. Thanks.