Picture the moment a ChatGPT ad shows up. A buyer has asked which brand to choose, and the assistant has answered: two or three names, the reasons for each, maybe a warning about one of them. Then, under the answer, comes a single sponsored card.
By the time the buyer sees the ad, the assistant has already told them what to think. So the ad competes with the answer above it, and the answer is the part the buyer trusts. That’s how we think about ChatGPT ads:
The ad sits next to the answer. The buyer decides inside it.
I saw it first with a client, and the research published since points the same way.
What we saw with a client
A client of ours tested ChatGPT ads. They spent a few hundred and saw more traffic from ChatGPT, both organic and from the ad, but almost no extra conversions compared to what they got organically before the campaign.
My hypothesis was that the ad is mostly a trigger. The buyer sees a brand in the sponsored card and asks the assistant they already trust: “what about this brand?” Then the assistant’s answer, about that brand vs the alternatives, decides what happens next.
The free recommendation beats the paid card
A new study of ChatGPT ads now puts numbers on it. In a working paper Christian Schulze shared on 2 November, he and Maximilian Kaiser followed the traffic and sales of 1,898 online shops for 60 weeks, while ads rolled out country by country and the EU stayed ad free. Paid ChatGPT visits came last of the major channels on every measure they tracked: behind Google’s paid and organic search, behind direct traffic, and behind the visits ChatGPT sends when it recommends a shop for free.
A paid ChatGPT visit brought in about a third of what a paid search visit did.
Revenue per visit, as a share of a paid Google Search visit. US traffic, February to August 2026.
Paid Google Search
100%Paid ChatGPT
36%Kaiser and Schulze, "Can Advertising Fund Free AI?", working paper, October 2026. The share is our arithmetic from the paper's 0.89 and 2.45 USD per visit.
Most of the commentary will be about prices, targeting and how early the platform is. All three are real, but none of them explains why ChatGPT’s free recommendation beats its own paid card.
The paper can’t see the conversation either. Like any analytics data, it sees the last click, not what was said before it. But two of its findings fit the answer coming first. Most advertisers who bought both ChatGPT ads and Google search ads got no sale at all from ChatGPT:
Advertisers that bought both ChatGPT ads and Google search ads, US traffic, February to August 2026.
No sale from their ChatGPT ads
84%At least one sale
16%Kaiser and Schulze, October 2026. 87 advertisers.
And the authors’ own reading is that the clicks are worth little because the sponsored card sits apart from the answer, where ChatGPT’s value is.
The ad and the answer rarely agree
So what does the answer above an ad usually say? SE Ranking captured ChatGPT’s answers to 50,006 shopping questions in the US in July. About one in four came with an ad, always a single sponsored card under the answer.
ChatGPT ads on US shopping questions, 23 July 2026: is the advertiser among the sources of the answer above its own ad?
Not cited
96%Cited
4%SE Ranking, August 2026. 12,974 ads on 50,006 questions.
The brand was named in the answer text less than one time in twenty, and about one ad in seven had nothing to do with the question.
Think about what the buyer has just read: an answer that recommends two or three brands and explains why, and then a card for a brand the answer didn’t mention. Or worse, the ad sits beside a reply that recommends a competitor, or argues against the advertiser’s own product!
No ad can win an argument the answer has already lost.
Why the answer wins
Because the buyer trusts the answer more than the ad. In a study of ads placed inside chatbot answers, people tended to treat the chatbot as an authority, and when the ads weren’t disclosed they mostly didn’t notice them.
A Princeton experiment from this year shows how much that matters. Over 2,000 people picked a book to keep, either from a search page or through a chat with an AI assistant. A fifth of the books were secretly sponsored.
Where a product sat barely changed the choice. What the assistant said about it did.
Share of people who picked a sponsored book, by how it was promoted. January and February 2026.
By chance
20%Upranked in search
22%Placed first in the chat
27%The assistant argued for it
61%Argued for it, labelled “Sponsored”
56%Salvi, Cuevas and Horta Ribeiro, Princeton, "Commercial Persuasion in AI-Mediated Conversations", 2026. 2,012 people.
The strongest lever was what the assistant said about the other books: caveats, warnings and short, flat descriptions. That was a lab task with cheap books, and today’s ChatGPT ads sit outside the answer. But turn it around for a brand.
When the assistant hedges about your product, that hedge travels with the buyer to your landing page.
The same goes when it describes a competitor with more enthusiasm than it describes you.
What the ad report can’t show
ChatGPT’s ad reporting shows impressions, clicks and, once the pixel is set up, conversions. It doesn’t show the conversation the ad sat in. Advertisers can’t see the prompts, so they can’t see what the assistant said about them, which competitors it named, or whether it argued against them in the same answer.
Search marketers learned a version of this more than a decade ago. In eBay’s experiments, switching off the ads on searches for its own name lost almost no sales, because the buyers who used to click the ads clicked the free result instead. The ad report had been counting sales the ads didn’t cause.
The ad report counts the click. What the assistant said about you stays outside it.
Some advertisers will do well on ChatGPT, and the reason may be the answer too. One agency running these campaigns reports that brands already mentioned in ChatGPT’s answers do better with ads than brands that aren’t. In the paper, the best known public success story is matched by only one of 102 advertisers. Whether an ad converts better when the answer agrees with it is the obvious next question, and no public data answers it yet.
Before you buy the slot
None of this means ChatGPT ads can’t work. It means the answer comes first. Before buying the slot, find out what the assistants say about you in the conversations your ad would join, against the competitors they name.
If the answer already recommends you, an ad may amplify it. If it recommends a competitor, or argues against your product, the ad pays to send a buyer who’s already doubtful. And judge the ad by the answer it sat under as well as by its clicks.
We research what the AI assistants tell buyers about a brand and its competitors, from the first question to the decision. If you want it for yours, you can request a free assessment.
Sources: Kaiser and Schulze, "Can Advertising Fund Free AI? Evidence from ChatGPT's International Advertising Rollout", working paper, October 2026, shared by Christian Schulze on LinkedIn, 2 November 2026. SE Ranking, Yulia Deda, "ChatGPT shows ads on 1 in 4 commercial queries", 10 August 2026. Salvi, Cuevas and Horta Ribeiro, "Commercial Persuasion in AI-Mediated Conversations", arXiv, April 2026. Tang, Sun, Curran, Schaub and Shin, "Ads That Talk Back", Proceedings of the ACM on IMWUT, 2025. Blake, Nosko and Tadelis, "Consumer Heterogeneity and Paid Search Effectiveness", Econometrica, 2015. Q1Media, "ChatGPT Ads: What Changed, What Didn't", 24 July 2026. The revenue share in the first figure is our arithmetic from the paper's published values.