14 SEP 2026 — OpenAI has stopped accepting ChatGPT advertising for image- and audio-generation products that compete with its own. Adobe, which took part in the initial advertising pilot and used it to promote its Firefly image generator, is among the advertisers affected. Video-generation tools can still buy space.
The change was not announced. It was communicated directly to advertising partners and surfaced through reporting on 9 and 10 September. Doug Wyatt, Adobe's senior director for Americas media, summarised it in a sentence that reads like a forecast: "Image and voice/audio are closed off. Video's still open…for now."
What changed, exactly
According to the reporting, OpenAI told partners that campaigns for standalone image and voice generation products would no longer be approved. The line is drawn around product category rather than company: Adobe can still advertise other products, and a firm selling video generation is unaffected.
The timing explains the categories. OpenAI has just shipped ChatGPT Images 2.5, and its voice product arrived in July. The two categories now closed to outside advertisers are the two in which ChatGPT itself most recently became a competitor. Video, which remains open, is the category where OpenAI has not yet placed a comparable product inside the same surface.
OpenAI has not published a policy or commented on the record in the reporting we have seen.
Normal for a publisher, not for this
The defence of the policy is straightforward and mostly correct. Media businesses routinely decline advertising from direct competitors. A newspaper does not usually run campaigns for a rival paper; a streaming service rarely sells slots to another streaming service. The advertising analyst Alan Chapell made that point in coverage of the change, and nobody seriously disputes it.
ChatGPT differs in scale and in timing. OpenAI has said ChatGPT serves more than a billion monthly users, most of them on free tiers, and in April it laid out an advertising revenue path of $2.5 billion this year, $11 billion in 2027, $25 billion in 2028, $53 billion in 2029 and $100 billion in 2030. A business with those targets is not a publisher declining one awkward booking. It is building an advertising platform whose most valuable inventory sits next to a user asking for help with exactly the task the blocked products perform.
A newspaper analogy leaves that out. Its reader can at least see that the rival's advert is missing. A ChatGPT user asking how to make an illustration sees only the answer, and cannot tell whether the absence of an alternative is editorial, algorithmic or commercial.
The self-preferencing question
Regulators already have a vocabulary for this. Europe's Digital Markets Act prohibits designated gatekeepers from favouring their own services over rivals', and much of the recent enforcement history — Google Shopping, the Amazon marketplace cases — turns on a platform using control of a surface to steer demand towards itself.
ChatGPT is not a designated gatekeeper, and refusing an advert is a narrower act than ranking your own product first. The legal exposure today is limited, and it would be wrong to suggest otherwise. But the shape is familiar. A platform with a large audience enters adjacent markets and adjusts, as it goes, who else may reach that audience. That is the pattern regulators have spent a decade learning to recognise, and advertising is where it is easiest to see because the rules are written down and the rejections arrive by email.
What it tells advertisers
Advertising eligibility on ChatGPT is not stable. The categories that are open today are open because OpenAI does not yet compete in them. A roadmap announcement from OpenAI is now, in effect, an advertising-policy announcement for everyone else in the category.
That is a risk marketers price in with any platform that also sells products, but it is sharper here than with a search engine. Search engines have spent years separating advertising from their own vertical products under regulatory pressure, and the separations are documented. ChatGPT's advertising business is months old, its policies are being communicated partner by partner, and the product surface it monetises is expanding into new categories every few weeks.
There is also a revenue trade inside the decision. Makers of AI tools are natural buyers of an audience that is, by definition, already using an AI assistant, and the image and voice categories are full of well-funded firms looking for exactly that customer. Closing them narrows the pool of advertisers in the categories where demand for this inventory should be strongest. OpenAI has evidently judged that protecting its own products is worth more than that spend, which is a coherent choice for a company that earns most of its money from subscriptions today, and a revealing one for a company that says advertising will be worth $100 billion by the end of the decade.
Adobe is a useful bellwether precisely because it is not small. If a company of that size learned about the change through its account team, smaller advertisers building on the pilot should assume they will learn about the next change the same way.
What would change this
Look first for a written policy. An advertising platform that aspires to $100 billion a year will eventually need published category rules, and whether they carve out OpenAI's own products explicitly will say a great deal.
Then watch video. If OpenAI places a video-generation product inside ChatGPT and video advertising closes shortly afterwards, the policy stops looking like a set of judgements and starts looking like a rule: compete with us and you cannot advertise to our users.
Regulatory attention may take longer. Nothing in the reporting suggests a complaint has been filed. The first one will probably come from an advertiser rather than a regulator, because advertisers are the people receiving the emails.
