SE Ranking published a ChatGPT ads study on August 10, built from 50,006 commercial prompts across 20 niches in the US and a one-day data snapshot taken July 23. Ads appeared next to 25.94% of those prompts, roughly one in four, close to the 29.45% rate SE Ranking measured the same way in Google AI Mode. The finding that stands out: the advertiser’s own domain is cited as a source in the answer sitting next to its ad only 3.63% of the time.
How Often Does ChatGPT Show Ads?
ChatGPT displayed an ad next to 25.94% of the 50,006 commercial prompts SE Ranking tested on July 23, a figure the study rounds to about one in four. Frequency varied by niche: Pets ran highest among the niches SE Ranking broke out, at 42.37%, while News & Politics, Technology, and Healthcare clustered close together at 28.76%, 28.76%, and 28.69%. Google AI Mode, tested with the same method, showed ads on 29.45% of commercial prompts, a slightly higher rate than ChatGPT’s.
Ad Frequency and Relevance by Niche
SE Ranking also scored whether each ad matched the prompt’s actual intent. Across the full sample, 14.35% of ads, about one in seven, came back off-target. The average hides a wide spread, shown below alongside ad frequency for the same niches.
| Niche | Ad frequency | Off-target rate |
|---|---|---|
| Pets | 42.37% | 2.6% |
| News & Politics | 28.76% | 54.2% |
| Technology | 28.76% | — |
| Healthcare | 28.69% | — |
| Relationships | — | 51.1% |
| Travel | — | 4% |
Pets runs against type: the niche with the highest ad frequency in the published figures also carries the lowest off-target rate among them, at 2.6%. News & Politics and Relationships run the opposite way, with more than half their ads scored unrelated to the prompt. SE Ranking scored relevance by embedding each prompt, ad, and answer, then measuring cosine similarity, with the off-target threshold set from distance ranges typical of random, unrelated pairs and checked against a manual sample.
The Ad Layer and the Answer Layer Don’t Overlap
Buying an ad slot in ChatGPT doesn’t buy a spot in the answer next to it. SE Ranking checked whether the domain behind each ad was cited by the answer it ran alongside: 3.63% of the time it was. The exact URL being advertised, not just the domain, showed up in just 0.09% of cases. The brand name appeared somewhere in the answer text in 4.44% of cases, a separate measurement from the 3.63%: a brand name in the answer text, not a source citation. Google AI Mode cited the advertiser’s domain in 11.53% of cases, roughly three times ChatGPT’s rate, though still a small minority of ad impressions there too. Tooling has been moving the same way. Earlier this month Microsoft Clarity started labeling whether the grounding query behind a citation carried the brand name, reporting Share of Authority for each group instead of one blended score: what an answer cites, and what prompted it, are counted as separate things.
A Concentrated Advertiser Market
SE Ranking counted 1,159 unique advertisers across the sample, and the inventory isn’t spread evenly among them. BestMoney alone accounted for 13.56% of every ad shown across all 20 niches combined. Within single categories the concentration runs further: Hungryroot took 79.92% of Food & Beverage ad impressions, Booking.com took 62.57% of Travel, and Robinhood took 52.24% of Finance. In this sample, a handful of names capture an outsized share of the visible inventory rather than the spread of a broad, even auction.
What a One-Day Snapshot Can’t Show
SE Ranking is explicit about the limits of its own numbers. The July 23 data is a snapshot, not a steady-state reading, and ads rotate: running the same prompt again can return a different set of advertisers. The niche rankings above describe patterns in this particular sample, not a complete inventory of who bids on what. The study doesn’t disclose how OpenAI targets these ads, and it says nothing about pricing, revenue, or how the numbers move over time. Two vendor studies on AI ghost citations landed on figures that don’t agree, roughly 40% against nearly 62% on different sample sizes and engine mixes, a reminder that one vendor’s snapshot is a data point, not a settled benchmark.
For anyone buying ads in ChatGPT, the ad slot and the answer slot are sold, and should be measured, separately. A domain paying for placement next to an answer is cited inside it 3.63% of the time, and the exact advertised URL lands in the response in roughly 1 case out of 1,100. Whether that ratio holds outside a single July day, or outside the US, is what SE Ranking’s own data cannot yet say.
Sources: SE Ranking, “ChatGPT Ads Study”; Search Engine Land coverage.