An Ad Beneath the Answer Is Not in the Answer

There are now two ways for a company to appear in ChatGPT, and the industry will spend the next year using one phrase for both. The first is being named in the answer: the paragraph the model writes when a buyer asks who makes a particular kind of valve, or which suppliers in northern Italy can machine a given alloy. The second is being shown in a sponsored unit underneath it. Both count, loosely, as appearing in ChatGPT. They are not the same event, they are not produced by the same system, and a report that adds them together is adding apples to invoices.

Until this summer the distinction was academic for most European exporters. It is not academic any more.

What changed at the end of August

OpenAI began testing ads in the United States in February. On 18 August it announced that ChatGPT Ads would expand to 31 European countries, naming Germany, France, Spain and Italy among them, and an update to the same announcement, dated 31 August, says that self-service access through Ads Manager is now available across all 31 markets. In practice an advertiser in Lyon or Stuttgart no longer needs OpenAI's sales team or an agency partner to buy a placement. They can set one up on a Tuesday afternoon.

Who sees the ads matters more for this argument than who buys them, and OpenAI's help centre is specific. Ads may appear for users on the Free and Go plans; Plus, Pro, Business, Enterprise and Edu accounts do not have them. They are not shown to accounts identified as belonging to under-18s, nor in Temporary Chats. Where they do appear, they sit below the end of a response, labelled as sponsored and visually separated from it.

One further line on that page is easy to skim past. Personalised ads are not initially available in the European Economic Area or Switzerland. A European on a free account who sees an ad is, for now, seeing one chosen from the context of the conversation in front of them rather than from their history.

Two systems on one screen

The mechanism, as OpenAI describes it, is a clean separation. The answer is written by the chat model. The ad is selected by a separate ads system, which weighs the context of the current conversation, the advertiser's landing page, copy and targeting, and ranks eligible ads on a combination of relevance and bids. OpenAI states that ads do not influence ChatGPT's answers and that advertisers cannot shape, rank or alter them. By default, the help centre adds, ChatGPT cannot even see the ads shown to the user.

Nobody outside OpenAI can audit that separation, and this article does not pretend to. Taken at face value, though, it has a useful consequence for anyone trying to work out what buyers encounter. There are two outputs on the screen, each with its own logic:

• The organic answer is whatever the model says in response to the question. Whether your company is in it depends on what the model has absorbed and, where it searches, what it retrieves. Nobody pays for a line in it.
• The sponsored unit is the outcome of an auction. Whether your company is in it depends on whether you bid, what you bid, how you targeted, and whether the conversation was judged relevant. It can show a company the answer above it ignores entirely.

A buyer on a free plan may therefore read a paragraph recommending three German manufacturers and, directly beneath it, a sponsored card for a fourth company from Spain. For that buyer, on that afternoon, four names were on the screen. For the purposes of measurement, three were earned and one was bought, and that difference is rather the point.

What a buyer sees and what a measurement sees

This is where it becomes awkward for anyone who measures, ourselves included.

Tools that measure AI visibility at any scale, PSentry among them, query the models through their APIs rather than by driving a consumer account through a browser. The OpenAI API is a different product from the ChatGPT app: its own models, its own settings, its own terms, and not a Free or Go account in any sense. OpenAI's advertising material concerns ChatGPT, the consumer product, and its developer documentation on ads covers the Advertiser API, which is for managing campaigns rather than receiving them. As of mid-September that documentation says nothing either way about ads inside API responses, so neither will we. What can be said is that a measurement through the API is a reading of the model's answer. It was never a screenshot of the free app.

That has always been a limit, and it deserves stating plainly. The app wraps its answers in things the API does not, and ads are now one more of them. PSentry measures the organic answer, across ChatGPT, Claude, Gemini and Perplexity, in each language you sell in. It does not see, count or report ChatGPT's sponsored units, and it would be misleading to let a score imply that it does.

Why the two figures must not be blended

The temptation will arrive from both directions.

A marketing team that starts buying ChatGPT placements in France will, quite reasonably, want one line on a slide saying how visible the company now is in ChatGPT. The easiest way to produce it is to add organic mentions to ad impressions. The result is a number that rises the day the budget is approved and falls the day it is cut, and that says nothing whatever about whether the model has any idea who you are.

The opposite error is quieter. A company named reliably in the organic answer may conclude it has nothing to think about, without noticing that on free accounts a competitor's sponsored card may now sit directly beneath the paragraph that recommends it. The organic measure is unchanged. What the buyer sees is not.

Kept apart, the two figures answer different questions, and each is useful on its own terms:

• The organic measure tells you whether the model, left to itself, considers you a credible answer to the question in that language. It moves slowly, because it follows what the web says about you, and it cannot be bought.
• Ad reporting, which comes from the advertiser side, tells you how often a paid placement was shown and clicked. It moves the moment you change a bid, and it stops when the spending stops.

There is also an asymmetry an exporter ought to notice. Ads reach Free and Go accounts only. A procurement engineer whose employer pays for a Business or Enterprise seat sees no ads at all, only the answer. In industrial B2B, where some of the serious research happens on accounts the employer pays for, the organic answer may be the only thing a technical buyer ever reads. How many of your own buyers fit that description depends on your market, and nobody has a trustworthy figure for it. Your sales team is better placed to guess than any vendor.

Where this leaves the budget conversation

Whether to buy ChatGPT ads is a commercial decision, and not one a measurement tool is entitled to make on your behalf. The reasoning is ordinary enough. An ad buys presence in front of a subset of users for as long as you pay, and says nothing about the answer. Organic presence is not for sale, and says nothing about the placement underneath. A company can sensibly do either, both or neither.

What does not hold up is treating one as evidence about the other. If the answer does not name you in German, a sponsored card beneath it does not change that, and a report that folds the two together will make it look as though it has. Two columns, labelled honestly, will do. And when someone asks how visible the company is in ChatGPT, it is worth establishing which of the two they mean before anyone answers.