Your Competitor Can Now Pay to Sit Under ChatGPT's Answer About You

TL;DR: ChatGPT now sells one sponsored card under its answers, and OpenAI says the ad never changes the answer above it. In SE Ranking's sample of 50,006 US commercial prompts, ads showed up on about a quarter of them, and the advertiser was also cited in the answer only 3.63% of the time. Two prizes, one screen. The card is rented by the click, while a place in the answer is earned and then keeps working. For most small businesses the answer is worth more, reaches people the ads never reach, and costs nothing per click. Start by checking that ChatGPT can read your site at all with the free AI visibility check.

Picture a customer asking ChatGPT which bakery in town does gluten-free birthday cakes. ChatGPT names yours, with a line about the lemon one. Under that answer sits a card marked Sponsored, from the bakery across the road, offering same-day pickup if you order before noon. That's allowed. It works in both directions, too, and once you see why, most of the decisions a small business owner faces about ChatGPT ads get a lot easier to make.

It is a competitor handing out flyers at your door, and the landlord is taking a fee for the spot.

Two systems on one screen

OpenAI built the ads to sit beside the answer and not inside it. Its advertising principles promise "answer independence": ads do not influence what ChatGPT says, and every ad is labeled as sponsored and visually separated from the answer. They also promise conversation privacy, which means advertisers never see anyone's chats or memories.

Behind the screen, different machinery picks each one. The answer comes from ChatGPT searching the web and choosing which pages to trust. The card comes from an auction, where advertisers describe the conversations they want in plain-language context hints and bid, and OpenAI scores relevance. Nothing in the auction asks who the answer named. So the card under an answer that praises you can belong to anyone who bid on that kind of conversation, including the business the answer just ranked below yours.

The first data: advertisers are rarely in the answer

The best public numbers so far come from SE Ranking, which collected 50,006 commercial prompts across 20 US niches on 23 July 2026 and published the results on 10 August. An ad appeared on 25.94% of the prompts, from 1,159 different advertisers, never more than one card per answer.

The overlap is the striking part. The advertiser's own site was among the sources cited in the answer in just 3.63% of ad placements, and the exact page being advertised showed up in 0.09% of them, which is 12 cases in the whole sample. The businesses buying the cards and the businesses ChatGPT recommends are, for now, almost entirely different groups. Read plainly, many early advertisers are paying for conversations they can't win on merit.

Two more details are worth knowing. About one ad in seven had little to do with the prompt it appeared under. And a single comparison site, BestMoney, accounted for at least 13.56% of all the ads SE Ranking saw. Ad density also varied by field: healthcare prompts showed ads 28.69% of the time, while education (14.18%) and travel (14.96%) saw about half that.

This is a single snapshot of US prompts. OpenAI changed its ad policies several times over the summer and keeps adding markets, so treat these as the shape of the thing and not a forecast.

People who never see an ad but always see the answer

Ads only go to logged-in adults on the Free and Go plans. Everyone on Plus, Pro, Business, Enterprise and Edu gets answers with no cards at all, and so do accounts identified as under 18. OpenAI also keeps ads away from conversations about health, mental health and politics.

That makes the answer the only way to reach several valuable groups. The consultant who pays for Plus because they live in ChatGPT all day will never see a card. Neither will the purchasing team at a company on ChatGPT Business, and neither will anyone asking a health question, even though that is exactly when a local clinic wants to be found.

Category rules close more doors. Outside the United States, OpenAI's ad policies do not allow health, financial or legal services to advertise at all. A physiotherapist in Dublin or a mortgage adviser in Madrid can't buy the card at any price, but either can still be the business ChatGPT names.

Sponsored Agents: the ad side turns into a conversation

On 16 September OpenAI started testing Sponsored Agents with select US advertisers. A person who clicks an ad can open a separate, clearly labeled conversation with the business's own agent, ask about sizes or delivery or whatever they need, and then follow a link to the site. OpenAI keeps that conversation apart from ChatGPT's independent answers.

The same announcement added an optional setting that rewrites an ad's headline to fit the conversation and translates the copy into the reader's language, along with an Ads Manager plugin that runs campaigns from inside ChatGPT. The paid side is getting more capable quickly. None of it touches the answer, which is still decided by what ChatGPT can find and trust about you on the open web.

How to get named in the answer

Getting into the answer is the same work that makes a business easy to find anywhere, with a few details that matter more for AI assistants. We go deeper in What is GEO?, and this is the short version for owners:

Start by letting OpenAI's search crawler in. OpenAI's crawler documentation says sites that block OAI-SearchBot will not appear in ChatGPT search answers. Blocks usually come from a CDN or firewall setting nobody remembers switching on, so check what actually gets through to the site, which can be very different from what robots.txt says. Our AI visibility check does that in a minute, and the robots.txt article goes through the rules line by line.

It helps to know that OpenAI runs a separate bot for each job. OAI-SearchBot finds pages for ChatGPT's search answers, ChatGPT-User fetches a page when someone asks ChatGPT to open it, GPTBot collects training data, and OAI-AdsBot checks the landing pages of ads. OpenAI says the settings are independent, so a business that blocks GPTBot to keep its content out of training can still allow the search bot and stay eligible for answers.

Prices, the areas you serve, opening hours and what you don't do belong in plain text on the page, not in a PDF menu or a photo of a price list. An assistant can't quote a number it can't read.

Under each heading, lead with the answer a customer wants and put the caveats after it. A paragraph that makes sense when quoted on its own is the kind that gets quoted.

Finally, ChatGPT usually reads more than your own site before it names a business, so reviews and mentions on other people's pages, local directories included, all count toward whether you look like a safe recommendation.

When paying for the card makes sense

The card earns its place in a few situations. A new business or product with no reputation online yet has nothing for the answer to find, and the ad is the only way into the conversation for now. A seasonal offer with a hard deadline, like a Christmas order cut-off, can't wait months for the answer to catch up. And a small, time-boxed test can tell you which pitch makes people act, which is worth knowing even if you stop afterwards.

Outside those cases, fix the answer first. The card costs money every time someone taps it, and the answer keeps working while you sleep. If you do decide to test ads, our guide to ChatGPT ads for small businesses covers who can buy them, what clicks cost and how to set up a test that tells you something.

Check where you stand once a month

Write down the ten questions customers ask before they buy from you, the way they would say them out loud. Once a month, ask ChatGPT each one from a free account, because a paid account never shows the cards and you want to see both layers. For each question, note whether your business is named, whether your site is cited as a source, which business is named instead, and whose ad appears underneath.

Then look at your analytics for visits from chatgpt.com. Links clicked inside ChatGPT often arrive tagged utm_source=chatgpt.com, which makes them easy to filter. After three months you will know whether you have an answer problem, an ad problem or neither.

If you would rather have someone else find out why ChatGPT names the bakery across the road and not yours, our full report has specialist AI agents go through your site in depth and hands you the fixes in order, starting with the ones that decide whether you get named at all.