Search Engine Journal reported on 2 November 2026 on a working paper by Maximilian Kaiser of the Vienna University of Economics and Business and Christian Schulze of the Frankfurt School of Finance and Management. They used Google Analytics ecommerce data from Grips Intelligence covering 102 US advertiser sites with full sales tracking, up to 22 August 2026. ChatGPT ads had launched in the US on 9 February 2026, shown as sponsored cards kept separate from the answer itself.
The headline numbers, US traffic:
| Measure | ChatGPT ad visits | Google paid search visits |
|---|---|---|
| Revenue per visit | $0.89 | $2.45 |
| Conversion rate | 0.37% | 1.17% |
- 73 of the 102 advertisers made no sales at all from ChatGPT ads, across 18,726 visits.
- 84 of the 87 advertisers running both channels earned less per ChatGPT visit than per paid search visit.
- 14 advertisers made all 570 ChatGPT purchases, and one of them made 66% of those.
Source: Kaiser and Schulze working paper, October 2026, as reported by Search Engine Journal, 2 November 2026.
The excitement was about the slot
When ChatGPT opened up ads, the pitch wrote itself. A shopper types out exactly what they want, in full sentences, and your ad sits right there. That sounds like paid search with better intent and less competition.
So I ask the usual question. What actually changed, and what didn’t?
What changed is that there is a new place to put an ad. What didn’t change is why the shopper came. They came for a recommendation. The answer gave them one, with brand names and reasons. The sponsored card sits beside that answer, and the data says most shoppers walk past it.
The authors are honest about the limits. One advertiser skews the totals. The study could not see any halo effect on organic traffic. Ad crawlers may inflate the visit counts. Delayed purchases are not counted. I would not call the channel dead on one paper. But 73 out of 102 advertisers with zero sales is a shape, not a rounding error, and the shape fits how people use these tools.
The shopper clicks the name, not the banner
Take a skincare shop. A shopper asks ChatGPT which vitamin C serum suits sensitive skin under fifty dollars. The answer names three brands and says why each one fits. Beside it sits a sponsored card from a fourth brand.
The shopper has just been told what to buy. The three named brands got the recommendation for free. The fourth brand paid to be the thing the shopper skips on the way to the recommendation. That is what a revenue per visit of $0.89 against $2.45 looks like from the shop’s side.
The same applies to a clinic, a tuition centre or a B2B software team. The buyer asked the engine a question, and the engine answered it. Being named in that answer is the outcome. Being cited as a source is a leading indicator. Buying the slot next to it is neither.
The AI does not pick the biggest brand or the one with the biggest ad budget. It picks the one it understands most clearly and can quote most precisely. That is something you can work on, and no ad slot substitutes for it.
I have paid for the slot before
At AeroChat we tried ads and at one point paid $53 for every install. That was never going to work for a bootstrapped company. Then I noticed Google’s AI Overview showed several options side by side, the ones it understood. So we stopped asking how to outrank them and started asking how to get the AI to see us as one of those options.
We wrote down exactly what AeroChat is, who it is for and how it is different, with no marketing language. Then we built comparison pages around the questions that actually produced an AI answer. Within three weeks AeroChat was in the same answer box as Tidio, Gorgias and Intercom. That was Google AI Overviews in early 2026, and AI answers change week to week, so I date every result.
I don’t know from the article what the 14 advertisers who did sell through ChatGPT ads did differently. My guess is that it matters less than whether the answer beside their card already named them.
What a business should do
- Ask ChatGPT, Google’s AI Overview and Gemini the questions your buyers actually ask. Write down who gets named, and the date. Do it again next month.
- Work out which of two things is broken. Ranking and being named are separate problems, and you fix them differently.
- Write down what you are in facts a machine can quote. Numbers survive in AI answers. Adjectives get dropped.
- Build pages that answer the specific questions that produce an AI answer in your category. Your own pages get you cited fast. Other people’s pages keep you there.
- If you still want to test ChatGPT ads, run a small budget, measure revenue per visit against your paid search, and treat it as a channel test. Keep the main effort on becoming the brand the answer names.
Source: Search Engine Journal: Google Search Ad Visits Out-Earned ChatGPT Ads In U.S.