Two transactions happen when someone asks ChatGPT a commercial question, and they never meet. The first is a read: the model draws on publisher pages, licensed archives and crawled reporting to compose an answer. The second is a sale: OpenAI auctions a sponsored placement beneath that answer to whichever advertiser wins on bid and relevance. OpenAI is party to both. The publisher is party only to the first, and is paid for it either through a negotiated licensing fee or, far more often, not at all. That gap is the whole publisher story of ChatGPT advertising, and it has widened considerably since the pilot opened in February 2026.
What ChatGPT Ads is now
OpenAI began testing ads in ChatGPT on 9 February 2026 for logged-in adult users in the United States on the Free and Go tiers. Plus, Pro, Business, Enterprise and Education tiers carry no ads, and OpenAI has said Free-tier users can opt out of ads in exchange for fewer daily messages.
The market footprint has expanded three times since. OpenAI's announcement post, "Testing ads in ChatGPT", records a March update extending the pilot to Canada, Australia and New Zealand, a May note flagging five further markets, and an update dated 11 August 2026 confirming that ChatGPT Ads has launched in the United Kingdom, Mexico, Brazil, Japan and South Korea, with more markets promised this year. That is nine live markets, including three of the largest advertising economies outside the United States.
The buying infrastructure has moved faster than the map. In the space of a few months OpenAI launched an Ads Manager and then opened it widely in the United States, scrapped its minimum commitment fee, turned on cost-per-click buying, shipped a conversion pixel, added automation letting retailers generate ads from product catalogues, and updated the visual treatment of the unit. Digiday reported on 28 May 2026 that cost-per-action buying had been switched on for select advertisers, following a press briefing in which OpenAI's head of monetisation Asad Awan said bidding on outcomes and conversions was "in the plan" and "should be done soon". Enders Analysis senior research analyst Claire Holubowskyj told Digiday that CPA "expands its offering while diversifying its pool of advertisers, and aligns its product more closely with that of Meta and Google".
By late July, OpenAI had reached the stage every ad platform reaches eventually. Digiday reviewed promotional offers giving new advertisers a $50 credit after spending $50 within fourteen days of opening an account, with a $100 version also in circulation. In August, Digiday found six live job postings for a dedicated small and medium business ads unit, including a Dublin vendor manager to oversee outsourced SMB sales. The people being hired come from the ads teams at Meta, Google, TikTok, Amazon and Reddit. This is not a company experimenting with advertising. It is a company building the org chart of a mature ad business.
How the auction works
OpenAI's help centre article "Ads in ChatGPT: The Basics" documents the mechanics directly, and publishers should read it as a competitor's rate card rather than an advertiser FAQ.
Ads appear below the response. Each unit carries an advertiser name, favicon, headline, description, landing page and image. Advertisers choose a Reach objective and pay per thousand impressions, or a Clicks objective and pay per valid click, setting a maximum bid at ad-group level. OpenAI recommends starting at a maximum bid of $3 to $5 per click. Selection runs through what OpenAI calls a relevance-weighted, second-price auction, so the winning advertiser pays only enough to beat the next bid, adjusted for how well the ad matches the conversation.
Targeting is not keyword buying. Advertisers supply context hints at ad-group level describing the conversations, topics or keywords where their product may be relevant, and OpenAI is explicit that these are not exact-match keywords and do not guarantee delivery. The system also weighs the landing page, the ad copy itself and, where ads personalisation is switched on, signals from the user's wider ChatGPT history. Reporting in the Ads Manager beta covers impressions, clicks, spend, click-through rate, average CPC, average CPM and conversions.
For a publisher, the significant detail is that every input to that auction is either advertiser-supplied or platform-owned. There is no field in it for the source of the answer.
How often ads appear, and how relevant they are
SE Ranking studied more than 50,000 commercial prompts across twenty niches, and Search Engine Land reported the findings on 10 August 2026. Ads appeared on 25.94 per cent of commercial prompts, which puts ChatGPT close to the 29.45 per cent SE Ranking previously found for Google's AI Mode. Roughly a quarter of commercial conversations now carry a sponsored placement.
Two further findings matter more for publishers than the headline rate. First, 14.35 per cent of the ads served were semantically unrelated to the prompt that triggered them, with mismatch rates ranging from 2.6 per cent in Pets to more than 50 per cent in Relationships and in News and Politics. Second, and more pointedly, in 96.37 per cent of placements the advertiser was not cited anywhere in the answer copy.
That last figure is the clearest statement available of how the answer layer is structured. The organic answer is assembled from one set of sources. The commercial placement is sold to a different party entirely. The two are adjacent on screen and disconnected in economics, which is exactly the arrangement OpenAI's answer-independence principle promises and exactly the arrangement that leaves publishers outside the transaction.
Why publishers are not in the revenue line
Nothing in OpenAI's published advertising material describes a mechanism by which publishers share in ChatGPT ad revenue. The announcement post, the ads principles and the advertiser help documentation set out user protections, buying models, measurement and brand safety. None of them establishes a payment relationship between the ad auction and the sources of the answer.
Content licensing does not close this. A licensing deal is a negotiated fee for training and retrieval rights, agreed once and available in practice only to publishers with the scale and legal weight to negotiate one. It is not a revenue share, it does not scale with ad load, and it does not track the commercial value of the specific answers a publisher's reporting informs. If ChatGPT ad revenue grows tenfold, a fixed licensing fee grows by nothing.
The economics compound in the platform's direction by design. eMarketer projects ChatGPT revenue per query rising from about $0.002 in 2026 to $0.041 by 2030, and annual ARPU rising from $3.50 to $60 over the same period, driven by higher-value commercial interactions and heavier ad load. Every point of that curve is value created at the moment of the answer and captured above the layer that supplies it.
The sharper commercial threat is to product research. As ChatGPT adds catalogue automation, product carousels and conversion tracking, it becomes a place where discovery, comparison and purchase happen in one exchange. That is precisely the sequence that recommendation and review franchises monetise. If the research happens inside the assistant, the publisher loses the click, the impression and the affiliate event in a single move.
What the forecasts actually show
Publishers should hold two facts about ChatGPT advertising at once: the ambition is enormous, and the delivery is unproven.
The Information reported in April 2026 that OpenAI projects around $2 billion in advertising revenue by the end of 2026, rising to $102 billion by 2030. Axios has reported an internal target of $2.5 billion for 2026 growing roughly fortyfold to $100 billion by 2030. Enders Analysis charts about $25 billion in ChatGPT consumer advertising within two years, $53 billion in 2029 and $102 billion in 2030.
Against that, eMarketer has forecast that OpenAI will fall roughly 90 per cent short of the 2030 target. Digiday's own reporting through the pilot describes under-delivery, limited inventory and early-stage infrastructure constraints, with advertisers holding back until third-party measurement arrives. Apptopia data cited by Digiday showed daily time spent per daily active user falling 18.3 per cent between 2 March and 11 May 2026, from 25.2 minutes to 20.6 minutes, with a 14.5 per cent decline even among the top ten per cent of users by engagement. No causal link to ads was established, but engagement is not yet moving in the direction a scaling ad business needs.
The honest reading is that the answer layer is being monetised aggressively and imperfectly, at a pace set by a company that burned $2.5 billion in 2025 and is expected to burn considerably more this year. Publishers should not plan around the $102 billion figure. They should plan around the fact that the surface is now permanently commercial, whatever it earns.
What publishers can do
Blocking is a real lever but a blunt one. Refusing the crawl forfeits the citation, the brand presence and any prospect of being paid for the read, in exchange for denying a platform one input among millions. It is worth doing where a publisher's content is genuinely differentiated and its leverage genuinely high. It is not a revenue strategy.
Three things are more durable.
Structure content for extraction and attribution. Similarweb's 2026 Generative AI Brand Visibility Index found that the brands surfacing consistently in AI answers are separated by how their information is structured, and that visibility in AI environments is becoming structurally distinct from search ranking. Being retrievable is now a precondition for being anywhere near the commercial surface.
Pursue licensing where scale makes it available, while pricing it honestly. It is a floor, not a share, and it should be negotiated with the expectation that ad revenue on the other side of the deal will grow faster than the fee does.
Monetise the read itself, at the layer the publisher still owns. When an AI system retrieves a publisher's page to build an answer, that retrieval happens at the publisher's edge, on infrastructure the publisher controls. This is where blankspace operates: detecting Live Search Agent and LLM traffic at the CDN edge and injecting contextual brand mentions into the retrieved content, so that the AI read produces value for the site being read rather than only for the platform selling against the answer.
That approach carries a genuine open question, and it is fairer to state it than to skip it. Publisher-side advertising to AI agents is contested territory. Perplexity blocked Time's agent-served ads in August 2026 and called the practice deceptive, and no vendor operating at this layer can guarantee that a model will honour what it is served or refrain from penalising it. Anyone selling certainty here is overselling. What is not in doubt is the asymmetry the ChatGPT auction demonstrates: the platform has built a priced, measured, nine-market mechanism for monetising the answer, and the open web that supplies the answer has not been given one.
Frequently asked questions
Does ChatGPT show ads to everyone?
No. Ads appear only for logged-in adult users on the Free and Go tiers, and only in markets where the product has launched. As of OpenAI's update of 11 August 2026, that means the United States, Canada, Australia, New Zealand, the United Kingdom, Mexico, Brazil, Japan and South Korea. Plus, Pro, Business, Enterprise and Education tiers carry no advertising, and Free-tier users can opt out of ads in exchange for fewer daily messages.
How are ChatGPT ads bought and priced?
Through a self-serve Ads Manager with no minimum commitment. Advertisers pick a Reach objective and pay per thousand impressions, or a Clicks objective and pay per valid click, setting a maximum bid at ad-group level; OpenAI recommends starting between $3 and $5 per click. Cost-per-action buying has been switched on for select advertisers. Selection runs through a relevance-weighted, second-price auction, so the highest bid does not automatically win.
Do publishers get paid when their content shapes a ChatGPT answer that carries an ad?
Not through the advertising. OpenAI has published no mechanism connecting ad revenue to the sources of an answer. A small number of large publishers hold content licensing deals, which are negotiated fees for training and retrieval rights rather than a share of what the resulting placements earn. The great majority of publishers whose content is read receive neither a payment nor an impression from that read.
How often do ads actually appear, and are they relevant?
SE Ranking's study of more than 50,000 commercial prompts, reported by Search Engine Land on 10 August 2026, found ads on 25.94 per cent of them, against 29.45 per cent for Google's AI Mode. Relevance is uneven: 14.35 per cent of placements were semantically unrelated to the prompt, from 2.6 per cent mismatch in Pets to more than half in Relationships and in News and Politics. In 96.37 per cent of cases the advertiser was not cited in the answer itself.
If the platform keeps the ad revenue, how can a publisher earn from AI answers at all?
By monetising the retrieval rather than chasing the answer surface. Structuring content for extraction keeps a publisher present in AI responses, licensing captures value where scale allows it, and edge-layer monetisation of the AI read, which is the approach blankspace takes, lets a publisher earn from infrastructure it still controls rather than from a click that increasingly never happens. The category is young and contested, and publishers should instrument and test rather than assume.