ChatGPT Ads Cannot Buy You a Recommendation: What OpenAI Says

TL;DR

ChatGPT ads are expanding into more markets, and they still cannot buy you the recommendation. OpenAI is unusually blunt about this: ads run on separate systems from the chat model, advertisers have no ability to shape, rank or alter ChatGPT’s responses, and seeing an ad does not mean OpenAI recommends that business.

There are two further catches almost nobody has priced in. Ads only appear for people on the Free and Go plans, so anyone on Plus, Pro, Business, Enterprise or Edu never sees one. And whole categories of conversation are closed to advertising altogether. If your buyers pay for their AI, or if they come to you through the kind of conversation OpenAI keeps ad-free, then being in the answer is the only route that exists. The answer is earned, not bought.

On 4 September, Search Engine Land reported that OpenAI is taking ChatGPT Ads into select countries across Europe, India, the Middle East and North Africa, along with a serious stack of new advertiser tooling.

The reaction from most marketing teams was the one you would expect. Good. A new channel. We can buy our way in.

You cannot. Not into the part that decides anything.

There are now two separate surfaces inside the same chat window, running on completely different rules. One is sold at auction. The other is earned slowly. Most of the advice written this month quietly treats them as one thing, and that is how a business ends up paying for a placement while staying invisible in the sentence that actually makes the sale.

Here is what changed, what it does not change, and what to do about it.

What actually changed in September

The expansion is the headline, but the tooling is the tell.

OpenAI added custom audiences above five million members and support for Google Advertising ID. The measurement pixel now takes hashed customer information including phone numbers, names, regions and postal codes. The Conversions API picked up Android GAID support. Product-feed advertisers can see performance at the carousel-card level, with impressions and clicks reported separately from billable ad impressions. There is even an Ads Manager plugin that lets advertisers build and analyse campaigns from natural language instructions inside ChatGPT or Codex.

That is not a pilot’s feature list. That is an advertising platform being built properly, with the plumbing that performance marketers expect before they will move real budget.

So take the channel seriously. Just be precise about what it is you are buying, because the same announcement makes the boundary clearer than ever.

What ChatGPT ads are, and where they sit

ChatGPT ads are paid placements that appear below the end of a response, clearly labelled as sponsored and visually separated from the answer itself. That is OpenAI’s own description, not an interpretation of it.

The test began in the US on 9 February 2026. By May, Digiday reported that testing had opened to the UK, Brazil, Japan, South Korea and Mexico, joining the US, Canada, Australia and New Zealand, with average monthly ad spend running at around 109 million dollars at that point. September’s expansion pushes it further across Europe and into India and MENA.

The categories matter for working out whether this is your problem yet. OpenAI’s ad policies, updated 31 August 2026, list the launch verticals as lifestyle and household goods, local services, travel and experiences, and digital products or education. Financial services, healthcare and medicine, and legal services are being approved case by case, manually, for approved advertisers only.

If you run a hotel, a restaurant, a local service business or a training company, your category is going live. If you run a clinic or a law firm, read the next few sections carefully, because the rules bite differently for you.

How big is the surface, honestly

Worth sizing before anyone reallocates a budget.

In the same week, Search Engine Land published Semrush Traffic Analytics data comparing July 2026 with July 2025 in the US market. ChatGPT was up 48.38 per cent year on year, sitting ninth among all US websites on 1.09 billion monthly visits. Bing was down 50.43 per cent over the same period.

Two honest caveats before you quote that at your board. It is US data, not UK. And website visits are a rough proxy for a product most people use in an app.

But the direction is not ambiguous, and the comparison with Bing is the part worth sitting with. A product that did not carry advertising eighteen months ago is now one of the largest destinations on the American internet, and it is opening an ad auction. The surface is real. What ChatGPT ads actually sell is narrower than it looks.

Why a paid slot is not a recommendation

This is the part worth reading twice, because it comes from OpenAI rather than from an agency selling you something.

Asked directly whether ads influence the answers ChatGPT gives, OpenAI’s help centre says: “No. Ads do not influence ChatGPT’s answers. Ads run on separate systems from our chat model, and advertisers have no ability to shape, rank, or alter ChatGPT’s responses. Ads are separate and clearly labeled.”

And then, in case anybody was still hoping: “Ads are paid placements, and seeing an ad doesn’t mean OpenAI endorses or recommends the advertiser or its products or services.”

Read that as a buyer rather than as a marketer. The ad is a box under the answer. The recommendation is inside the answer, in the sentence where the assistant says which three places are worth booking and why. Those are not the same real estate, and one of them is not for sale.

OpenAI has written the separation into its rules for advertisers, too. The ad policies carry an interface imitation clause: ads must be “clearly distinguishable from the ChatGPT product experience”, and ads that imitate the appearance, functionality or voice of ChatGPT can be modified or removed. The one thing a marketer might want most, an ad that reads like the assistant’s own advice, is specifically prohibited.

Ranking inside the ad auction works the way ad auctions always have, on a combination of relevance and advertiser bids. Ranking inside the answer does not work that way at all. No bid enters that calculation. Nothing you can buy from OpenAI changes which businesses the model names when somebody asks it for a shortlist.

Your best customers will never see the ad

Here is the detail that changes the maths, sitting in plain sight in OpenAI’s own FAQ.

ChatGPT ads may appear for users on the Free and Go plans. Plus, Pro, Business, Enterprise and Edu accounts will not have ChatGPT ads. OpenAI also does not show ads to accounts identified as belonging to people under 18.

Now think about who actually pays for ChatGPT. The consultant who lives in it all day. The operations manager whose company bought Business seats. The frequent traveller who upgraded because they plan every trip in it. The buyer researching a five figure purchase on a Tuesday afternoon.

Those people are on paid plans. For that entire audience, your ad does not exist. It is not that your ad underperforms. There is no placement to buy.

This is the single most important line in your media plan this year, and it is a product decision rather than a marketing one. If your customers skew towards people who take AI seriously enough to pay for it, the only way to appear in front of them is to be in the answer.

The conversations where ads are not allowed at all

The second exclusion is broader than most coverage has noticed, and for some businesses it closes the channel completely.

OpenAI’s ad policies keep ChatGPT ads out of what it calls sensitive user contexts: “conversations involving personal, high-stakes, or emotionally vulnerable situations where ads could undermine user trust or create a poor experience”. The list of ad-free contexts explicitly includes mental and personal health conversations, emotionally reliant contexts, and sensitive user journeys.

Sit with what that means if you run a clinic.

Someone works up the courage to ask an AI assistant about a procedure they have been thinking about for two years. That is a personal health conversation. No ad will appear in it. Not yours, not a competitor’s. The only thing that conversation can surface is whatever the model already believes about the providers in that field, drawn from the sources it trusts.

The same logic applies anywhere the buying decision is emotional, private or high-stakes. Health. Money worries. Legal trouble. Anything a person would rather ask a machine than a friend.

Those are frequently the highest-value conversations in the entire funnel, and advertising has been deliberately designed out of them. What is left is whether you are in the answer.

What your ad audience can switch off

Even inside the permitted audience, the targeting is less absolute than a media pack suggests.

Users get an ad controls panel. They can turn off ad personalisation at any time, dismiss ads and send feedback, see why a particular ad was shown, and clear the data used for ads. Turn personalisation off and you still see ads based on the current chat thread, but OpenAI stops using other threads, ad history or topics to choose them.

Personalised ads are also not initially available in the European Economic Area or Switzerland. For a UK business selling into Europe, that is a live constraint on how well the channel can be targeted, not a footnote.

And advertisers never see the conversation. OpenAI does not share chats with advertisers and does not sell that data. Advertisers receive aggregated, non-identifying performance information, things like total views and clicks.

That is good for users, and it should shape your expectations. You are buying contextual reach with limited signal, in a walled measurement environment, against a slice of the audience.

What this means for your budget

Two surfaces, two disciplines, two lines in the spreadsheet. ChatGPT ads sit on one side of that line and your visibility work sits on the other.

ChatGPT ads buy reach among free-tier users, in permitted categories, in permitted conversations, in markets where the test has landed, measured the way paid media has always been measured. That is a legitimate thing to spend money on, and the new tooling makes it easier to run properly.

Visibility inside the answer is a different job entirely. It comes from being the kind of source an engine can find, parse, trust and quote: clear answers to real questions on your own site, structure a machine can read without guessing, and evidence about you spread across the places engines actually go looking.

We set out the mechanics in why your business is not showing up in AI search, and the difference between the two ways of working in GEO vs SEO explained. If the vocabulary is still blurry, what answer engine optimisation actually is is the place to start.

The mistake to avoid is booking the ad spend and then reporting on it as though it says something about your standing in the answers. It does not. OpenAI has told you plainly that it does not.

How to tell whether AI actually recommends you

You cannot manage this from a dashboard of impressions, because the thing you care about is a sentence, not a number.

The discipline is simple, and most people skip it.

Ask the questions your customers actually ask, in the engines they actually use. Save the answers verbatim, with the date and the engine named. Record whether you were mentioned, in what position, and which sources the engine leaned on to get there. Record the gaps honestly too, including the runs where you did not appear at all.

Then do it again, because answers differ run to run and from one person to the next. A single check tells you what one engine said in one session on one day. A run of checks over weeks tells you whether you are genuinely in the consideration set, and a saved answer from six weeks ago tells you whether anything you changed since has moved it.

That record is also the only honest way to see whether your ad campaign and your visibility work are doing different jobs, which is precisely what you need to know before you renew either one. If you are not sure where to start, is AI killing your business covers the behaviour change underneath all of this.

FAQ

Next step

ChatGPT ads are a real channel, and worth testing if your category is live and your buyers sit on the free tier. Just do not mistake the receipt for the ad for proof that AI recommends you. OpenAI has been clear that these are two different systems, and only one of them is for sale.

The businesses that will do well here are the ones that can tell the difference on paper: what the ads bought, what the answers actually say, and which of the two is moving.

If you want to see what the engines are saying about you now, and how that gets tracked over time rather than guessed at, start for free with Aurora.

Ayse Durmush

AI Visibility Expert and Founder of The GEO Agency, helping travel, hospitality and property companies win visibility in AI search. 28 years of digital marketing, brand strategy and enterprise digital transformation experience with brands including Virgin Media, O2, Samsung, LEGO, RBS/NatWest and ITN. Author of the ChatGPT Blueprint, Anthropic Claude Partner Network member and Claude Certified Architect.