In February 2026, OpenAI quietly launched advertising inside ChatGPT. By July, ads were appearing in roughly 49% of US replies. The platform is projecting $2.5 billion in ad revenue for the year. OpenAI showed up at Cannes Lions and declared itself an advertising business. It now has a self-serve Ads Manager with no minimum spend, CPA bidding, and audience and creative tools that are evolving faster than most agencies can track.
That last point matters more than the revenue number. Digiday reported that most agencies are still learning about ChatGPT's ad product from third parties — not from OpenAI directly. That is not a comms failure. That is a signal about how this rollout was designed: fast, opaque, and asking forgiveness rather than permission.
Welcome to the third gatekeeper.
Two Was Already Too Many
The digital advertising industry has spent fifteen years negotiating its dependence on two platforms that control the majority of global ad spend, set their own measurement standards, define their own brand safety frameworks, and answer to nobody's supply chain transparency requirements except their own. Publishers and agencies alike have built entire businesses around the edges of that duopoly — finding margin in the gaps, building direct relationships, diversifying into programmatic, and hoping the CPMs hold.
Now there is a third. And this one is structurally different in ways that should concern anyone who depends on the open web for revenue.
ChatGPT's advertising model is not programmatic. It does not plug into the pipes that fifteen years of adtech infrastructure were built to serve. There are no DSPs bidding into it, no SSPs connecting supply, no auction dynamics publishers can optimise against. It is a closed, conversation-based environment where ads surface contextually inside a dialogue — and the audience never leaves the platform to visit a publisher's site.
That is the threat in one sentence: the audience stays inside the LLM, and so does the ad revenue.
The Audience Fragmentation No One Is Modelling Yet
Publishers are already managing fragmented attention across search, social, video, newsletters, and podcasts. Add to that list: AI-native query environments that are increasingly the first stop for information that used to drive organic search traffic to publisher pages.
If someone asks ChatGPT a question and gets a good answer — with a contextually relevant ad embedded in that answer — there is no click to a publisher's article. There is no page view. There is no programmatic impression. There is no revenue. The publisher's content may have trained the model that generated the answer, but the commercial value of that interaction flows entirely to OpenAI.
The scale of this is still emerging, but the direction is not ambiguous. As LLM interfaces become habitual for more users, the audiences that remain on the open web will shrink. The publishers who survive that shift will be the ones who extracted maximum value from their retained audience — not the ones who assumed programmatic CPMs would hold indefinitely.
The Measurement Problem No One Has Solved
Here is the part that should keep agency trading desks awake. Every measurement framework, brand safety standard, viewability metric, and incrementality model that the industry built since 2010 was designed for the open web. It assumes a page, a placement, a cookie (or its successor), and some kind of audit trail.
ChatGPT's ad environment has none of that. Ads appear inside conversational responses. There is no standard for what a viewable impression means inside a dialogue. Brand safety verification as currently defined does not translate to conversational AI contexts. Attribution for a CPA bid placed inside a chat interface is a genuinely unsolved problem. The agencies who are still learning about this product from third parties are not behind because they are slow — they are behind because OpenAI moved before any of this infrastructure existed.
That creates an opening for whoever defines the standards first. It also creates enormous risk for advertisers who move budget into this environment before those standards exist.
The GDPR Question Nobody Is Asking Loudly Enough
There is a dimension to this story that European publishers and buyers should be pressing hard on, and largely are not: the data practices underpinning ChatGPT's advertising targeting.
OpenAI's ad product uses audience and contextual signals to match ads to conversations. What data is being used? How is it collected? Under what legal basis? Where is it stored? How does it interact with the EU AI Act's requirements around transparency and high-risk AI systems used in commercial contexts?
These are not hypothetical questions for a future regulatory cycle. They are live compliance obligations right now for any European publisher, agency, or advertiser that interacts with this platform. The rollout speed and opacity that Digiday documented is not just a commercial frustration; it is a due diligence problem under GDPR. And given OpenAI's track record with EU regulators on training data, the idea that its advertising data practices will sail through without scrutiny is optimistic at best.
European adtech players who have built privacy-first infrastructure — no third-party cookies, first-party data under customer control, full audit trails — are not just better positioned commercially. They are better positioned legally.
What Publishers Should Actually Do
The strategic response to a third gatekeeper is not to chase it. Publishers are not going to outbid OpenAI for position inside ChatGPT. The response is to make the audience and the inventory that remains on owned channels as valuable as possible — and to stop relying on infrastructure owned by the same intermediaries who are quietly routing budget away from them.
That means owning your ad server, not renting one from a platform that has commercial interests in your audience data. It means building first-party data relationships with your readers that no LLM can replicate or monetise without your consent. It means using contextual targeting that works without third-party cookies — because the open web's cookieless future and the LLM advertising future are both converging on the same requirement: know your audience from signals you own, not signals you borrow.
It means shortening the supply chain between your inventory and the buyers who want it — because every intermediary in that chain is a margin leak you cannot afford when the addressable audience is getting smaller.
Infrastructure that publishers own and control — their own ad server, their own data, their own direct buyer relationships — is not just a nice philosophical position. It is the only credible hedge against a market structure where three closed platforms increasingly decide where the money goes.
The Gatekeeper Pattern Is Repeating. The Response Should Not.
The industry watched two platforms become indispensable, then extracted. It built workarounds, called them strategies, and largely accepted the margin compression that came with dependence. The pattern is now repeating with a third platform that moves faster, operates more opaquely, and sits even further outside the existing compliance and measurement infrastructure.
The publishers and buyers who come out of this period intact will be the ones who treated the arrival of ChatGPT advertising not as a new channel to test, but as a forcing function to get their own house in order. Own your data. Own your stack. Own your audience relationships.
Because the alternative — waiting to see how the third gatekeeper's terms evolve — is a strategy the industry has already tried twice, and it did not end well either time.