You can buy an ad in ChatGPT. You can't buy the answer.

When I pulled the questions Google shows under "how to get my business recommended by ChatGPT," the first one was this: "Can you pay ChatGPT to recommend your business?" It's a fair question to be asking this month. OpenAI has spent 2026 building an ad business inside ChatGPT, and in early October it announced another new ad format. From the outside, it looks like the recommendations went up for sale.

They didn't, and the difference matters more than it sounds.

There are now two ways to show up in a ChatGPT conversation. One is an ad, which you can buy this afternoon. The other is the answer itself, which you can't buy, and which is where your buyers are actually making decisions. Most of the confusion I see comes from treating them as the same thing.

Can you pay ChatGPT to recommend your business?

No. You can pay for an ad that appears below ChatGPT's answer, labeled as sponsored, but you can't pay to be named inside the answer. OpenAI's help page says advertisers have no ability to shape, rank, or alter ChatGPT's responses. The recommendation has to be earned from what the model reads about you.

That's OpenAI's own line, and it has repeated it at every step. Before the first ad ran, it published a set of ad principles with "answer independence" near the top: ads don't influence answers, and they're always separate and clearly labeled. The same rule covers shopping. ChatGPT's product results, per its shopping help page, "are selected independently by ChatGPT and are not ads."

You can be skeptical of a company grading its own homework, and I'd encourage some of that. But the way the system is built supports the claim. Ads run on separate systems from the chat model, they sit below the response, and they carry a label. If an advertiser could buy a mention inside the answer, the whole product would stop being worth using, and OpenAI knows that better than anyone.

What the ad actually buys

The ad business is real and growing fast, so it's worth knowing what you'd be paying for. OpenAI began testing ads in the US on February 9 for logged-in adults on the Free and Go plans, then expanded to Canada, Australia and New Zealand, and by August to the UK, Mexico, Brazil, Japan and South Korea. In May it opened a self-serve Ads Manager in beta with cost-per-click bidding, so US advertisers can apply for access and run campaigns directly, though the rollout is still limited.

The newest piece arrived in early October. OpenAI announced a visual ad format that will appear while ChatGPT generates images, tested in the US with a first group of advertisers later in October. In the same announcement it said ChatGPT reaches 1.2 billion people a week and added conversion integrations with Hightouch, Tealium and LiveRamp, the plumbing an advertiser needs to see whether the clicks turned into anything.

How often do people see these ads? Similarweb, which sells a tool for tracking them, estimates that 26% of ChatGPT responses on the Free and Go plans already carry a sponsored ad. That's a vendor estimate from a panel, so treat it as a rough size, not a count.

Here's the detail most coverage skips. Ads are matched to the topic of the conversation, your past chats, and past ad interactions, and Plus, Pro, Business, Enterprise and Education accounts don't see ads at all. If you sell to companies, sit with that.

A buyer researching software on a company ChatGPT Business seat can't be shown your ad, however much you bid. The ad reaches people on the free and cheap plans. The answer reaches everyone.

So the ad is a normal ad channel with an unusual setting. It might be a good one for some businesses, and I'll say more about that below. It just isn't the thing people mean when they ask to be recommended.

How do I get ChatGPT to recognize my business?

ChatGPT recognizes a business from what other sites say about it: licensing records, review profiles, directories, comparison articles, and your own pages. It searches the web, reads those sources, and names the businesses that keep turning up in them. The more consistent and plentiful the evidence, the more likely you are to be named.

The best recent look at this for local businesses was published on October 5. 8ight.AI, a company that sells AI-visibility services, asked ChatGPT and Claude who the best tradesperson was across 39 trades in 49 places, then checked which websites each one pointed to.

Across 1,412 ChatGPT answers, according to PPC Land's write-up of the study, 38.7% cited the Better Business Bureau and 25.1% cited a government or licensing site. Yelp was cited in none of them.

Reviews mattered, but only the count. Businesses with fewer than 25 Google reviews were named by ChatGPT 3.4% of the time. Businesses with 250 to 999 reviews were named 20.9% of the time.

Star rating made no difference: the median was 4.9 for named and unnamed businesses alike. A 4.9 is ordinary. Three hundred reviews is not.

The sources also change from place to place. A UK firm called Whito ran 36 local questions in three towns in July and found Reddit was ChatGPT's most-cited source, in 10 of 18 answers. In the 8ight.AI data, Reddit appeared in 6 of 1,630. Different country, different method, different month.

The honest summary is that ChatGPT reads where the evidence about your kind of business happens to live, and that varies.

For B2B software, the evidence lives in a different set of places, and the stakes are higher. G2 surveyed 1,076 software buyers in March and found that 51% now start their research with an AI chatbot more often than with Google. 69% said a chatbot led them to a different vendor than they had planned, and 33% bought from a vendor they had never heard of before. A third of these buyers paid a company that was new to them when they started looking.

The part money can buy, sort of

This is where the clean "you can't pay" answer gets messier, and I'd rather show you the mess than pretend it isn't there.

You can't pay ChatGPT. You can pay the pages ChatGPT reads.

GetCited, a Berlin agency, traced the sources behind 3,334 "best X" questions and found that 25.4% of the cited pages had a disclosed commercial interest, mostly affiliate reviews that earn a commission on a sale. 86% of answers used at least one such page, and ChatGPT did so in 91% of them. Paid rankings on major news sites, the Los Angeles Times among them, showed up as sources too.

Look at who ran that study. According to PPC Land's examination of it, GetCited's business is writing and placing comparison articles on publisher sites so that AI assistants recommend its clients, with campaigns starting at $3,000 a month. The sponsored label stays on the publisher's page. It doesn't travel into the AI's answer.

So a version of "paying ChatGPT to recommend you" does exist. It's paying for placement in the sources, and it's an old trade: advertorials, sponsored listicles, pay-to-play directories. I won't tell you it never works. I'll tell you that it's renting someone else's credibility, it's invisible to the buyer only until they click through and see the label, and the evidence that it reliably moves AI answers is thin.

How thin? In July, Olivier Martinez published a critical survey of the research on generative engine optimization, the industry's name for this work. It found that no reviewed technique showed a stable, longitudinal, cross-platform effect on whether a page gets discovered in the first place. Content already in front of the model can change how it gets cited, but getting in front of the model is the part nobody has proven they can engineer.

How do I get my business ranked on ChatGPT?

You can't, because there's no ranking to get. ChatGPT gives a different list almost every time it's asked. What you can measure, and improve, is how often you're named across many versions of the same question. That share is stable enough to track, and a position is not.

SparkToro and Gumshoe.ai ran 2,961 prompts through ChatGPT, Claude and Google's AI and found the same list of brands came back less than 1% of the time, and the same list in the same order less than 0.1% of the time.

Gumshoe sells AI tracking, and SparkToro's Rand Fishkin says he went in expecting to show that tracking was pointless. What survived the test was how often a brand appears, not where it ranks. His own summary of the work: rankings are "probably nonsense," while visibility across many prompts and runs is "surprisingly useful."

The 8ight.AI data says the same thing at the local level. When ChatGPT got the same question twice, a median of 39 minutes apart, only 29% of the businesses named appeared in both answers. Even the most-reviewed business in a town was left out about three times in four.

That should change how you judge anyone selling this work, including me. If someone shows you a screenshot of you at "number one in ChatGPT," they're showing you one roll of the dice. The IAB's new measurement guidelines note that only 16% of brands track AI visibility systematically at all. Most of what gets reported is anecdote.

How can I get AI to recommend my business?

Make the evidence about your business plentiful, consistent and easy to find in the places AI tools read: licensing and accreditation records, review profiles with real volume, directories, honest comparisons, and pages on your own site that answer the questions buyers actually ask. Then measure your share of mentions over time.

Here's where I'd start, in order of cost.

First, fix the records that already exist. Your business name, address, phone number and description should match everywhere: your site, your Google Business Profile, the BBB, industry directories. If you're a licensed trade here in Virginia, your record on the state's DPOR licensing lookup is exactly the kind of page ChatGPT cited in a quarter of local answers, and an old trade name or lapsed licence there is now a marketing problem.

Second, ask for reviews consistently, not in a burst. The 8ight.AI numbers suggest volume does the work, and you only get volume by asking every happy customer, every month.

Third, write the pages a model can quote. For each question a buyer asks before hiring you, one plain answer near the top of a page, in a sentence that would still make sense lifted out on its own. For a software company that means pricing logic, integrations, who it's for and who it isn't for. Vague pages give the model nothing to repeat.

Fourth, get described by other people. Industry press, podcasts, honest review sites, a comparison article that includes you because you belong in it. If you pay for any of it, know that you're buying a source, not an answer, and check what the label says.

Fifth, measure the right thing. Pick ten questions your buyers ask, run each one several times a month in ChatGPT and at least one other assistant, and record how often you're named. Add a free-text "How did you hear about us?" field to your contact form. It catches the visits your analytics files under direct traffic.

And if you want to test the ads themselves, test them like any other channel. Small budget, conversion tracking switched on, judged on leads your team would actually call back. Just don't expect the ad to do the recommendation's job. They're bought differently, shown to different people, and believed differently.

Where I fit

The line on my answer engine optimization page is "Customers are searching for exactly what you do. They're finding someone else." In 2026 that search is often a conversation, and the someone else is whoever the assistant found the most consistent evidence for. Part of the work is earning a place in those answers. The other part is a report that shows whether that place turned into revenue, not just mentions.

The work is unglamorous on purpose: records cleaned up, reviews asked for, pages that answer real questions, coverage earned in places models read, and a share-of-mentions number tracked over months instead of a screenshot. It's the same work I'm doing on this site.

What I can't promise you is a position in ChatGPT, because there isn't one, and nobody has proven they can engineer one. If someone offers you a guaranteed spot, ask them to run the same question ten times while you watch.