MEIKAIChatGPT ads in 2026: What marketers need to know
ChatGPT Ads has moved from a single-market experiment to a buyable channel with an auction, campaign reporting and conversion measurement. That makes it testable, which is a different thing from making it understood.
Updated 27 July 2026. Everything below comes from OpenAI's own documentation, linked at the end. Availability, eligibility and controls are still changing, so check the current documentation before committing budget.
What you can actually buy today
OpenAI began testing ads in ChatGPT in February 2026, initially for logged-in adults on Free and Go plans in the United States. In May 2026 it announced expansion to Canada, Australia and New Zealand, with further markets planned. Ads are withheld from users OpenAI identifies as under 18 and excluded from sensitive topics including health, mental health and politics.
The buying system currently offers:
- CPM campaigns for reach and awareness
- CPC campaigns optimised toward clicks
- a relevance-weighted second-price auction, so the highest bid does not automatically win
- reporting on impressions, clicks, spend, CTR, CPC, CPM and conversions
- conversion measurement via supported website events
Ads are labelled and rendered separately from the answer. OpenAI states that it does not sell personal data to advertisers, that advertisers do not receive users' conversations, and that ads do not influence ChatGPT's organic responses.
Planning against conversations, not keywords
Paid search treats the query as the demand signal. Here, relevance can draw on the current prompt and on context established earlier in the conversation, which breaks the one-line-of-intent assumption a keyword list encodes.
What replaces it is closer to a brief than a list: the need the person is trying to resolve, the constraints they have stated (price, market, use case, timing), the stage of consideration the conversation has reached, and creative and landing pages that answer that same need, not a keyword adjacent to it.
A reviewed prompt model is a practical way to organise that before spending. It separates informational prompts from commercial consideration, preserves the variations that matter, and records why each cluster exists. Our prompt-modeling methodology covers how to build one that holds up over time.
Why organic visibility cannot be your control group
The obvious way to evaluate a ChatGPT Ads test is to measure AI visibility before the campaign, run it, and measure again. Our own data says that will mislead you.
Between March and July 2026 we tracked a fixed cohort of 126 brands against a set of cited pages that their publisher removed, briefly restored in June, then removed again. On ChatGPT, citations per 1,000 responses fell from 166.1 to 20.2 across that window, a retention of 12.2%, while Perplexity, measured the same way over the same period, sat at 109.5% of its baseline. The underlying analysis explains what drove it.
Organic visibility did not drift on its own here. A known change to the source pages drove it. What matters for a media team is that the same change produced an eightfold fall on ChatGPT and no fall at all on Perplexity. Your organic baseline moves whenever your source environment moves. A partner page is restructured, a review site is redesigned, a retailer delists a line. Those things happen during campaigns, they hit each platform differently, and a before-and-after comparison will book the result as media effect. If you want a causal read you need a holdout, a geographic split or a matched comparison period, the same discipline you would apply to any other channel.
Keep paid and organic apart
An ad impression and an organic citation are different outcomes with different meanings. An impression means you bought a placement. A citation means the model reached for your content while composing an answer. Combining them into one visibility score destroys the only interesting distinction between them, and it lets a paid budget flatter an organic problem.
Report them separately, compare them deliberately, and never describe a paid placement as evidence that the assistant recommends you.
How to run a first test
- Define the prompt clusters before the creative. Decide which needs and decision stages you are buying into, and why those and not others.
- Set the conversion event and attribution window up front, along with the budget and the conditions under which you stop.
- Build in a comparison. A holdout region or a matched period is worth more than a larger test without one.
- Agree brand-safety exclusions and who reviews them before launch, not after the first surprise.
- Re-read the documentation before each flight. Markets, formats, policies and measurement capabilities are all still moving.