ENTRY 013 · CHATGPT · By Answer Engineered Research
96.37%
ChatGPT Ads Reached Five More Countries. The Ad Rarely Buys the Citation.
OpenAI launched ChatGPT Ads in five more countries. A 50,006-prompt study finds the advertiser absent from the answer's own sources 96.37% of the time.
What actually shipped on 11 August
OpenAI has left every dated update stacked on one URL, which is unusually good practice and makes the timeline checkable without trusting anyone’s summary of it.
| Date | What the page says |
|---|---|
| 9 February 2026 | ”Today, we’re beginning to test ads in ChatGPT in the U.S. The test will be for logged-in adult users on the Free and Go subscription tiers.” |
| 26 March 2026 | ”The early results are encouraging. We’re seeing no impact on consumer trust metrics, low dismissal rates of ads, and ongoing improvements in the relevance of ads as we learn from feedback.” |
| 7 May 2026 | ”In the coming weeks, we plan to expand the ads pilot in ChatGPT in the United Kingdom, Mexico, Brazil, Japan, and South Korea.” |
| 11 August 2026 | ”ChatGPT Ads has now launched in the United Kingdom, Mexico, Brazil, Japan, and South Korea. We’re continuing to expand to more markets this year.” |
The May list and the August list are the same five countries. This is a genuine pilot-to-launch graduation, not a separate rollout being counted twice.
Two mechanics from the same page are worth holding onto. On targeting: “During the test, we decide which ad to show by matching ads submitted by advertisers with the topic of your conversation, your past chats, and past interactions with ads.” And on independence: “Ads do not influence the answers ChatGPT gives you.”
Six months of updates, not one published number
Read the March entry again. “No impact on consumer trust metrics.” “Low dismissal rates.” “Encouraging.”
There is no percentage attached to any of those words. Not on that page, not in any of the five dated update blocks, not on the advertising principles page OpenAI published on 16 January 2026. No baseline, no sample, no definition of what a trust metric is or what counts as low.
We are not saying those claims are false. We are saying they are unfalsifiable by anyone outside OpenAI, and that six months is long enough for a company to publish one figure if the figure were flattering.
That principles page also contains the most falsifiable commitment OpenAI has made here, from Fidji Simo: “Long-term value: We do not optimize for time spent in ChatGPT. We prioritize user trust and user experience over revenue.” Worth writing down now, while the ads business is small, so it can be checked later when it isn’t.
The number someone else did publish
SE Ranking ran 50,006 commercial prompts across 20 niches in the US market and recorded, for each one, the generated answer, its cited sources, and any sponsored placement.
The headline is that ads showed on 25.94% of commercial prompts. Roughly one in four.
The finding that matters more for anyone reading this blog is what the ad does not do.
| Measure | Result |
|---|---|
| Advertiser appears among the answer’s cited sources | 3.63% |
| Advertiser absent from the cited sources | 96.37% |
| The exact advertised URL appears in a citation | 0.09% |
| Advertiser brand mentioned in the body of the response | 4.44% |
Buying the ad places a sponsored unit next to the answer. It does not, in 96 cases out of 100, put you inside the answer’s source list. The two surfaces are sold together and behave separately.
That is consistent with what OpenAI itself says — ads do not influence the answer — but consistency cuts both ways. If ads genuinely do not influence answers, then ads cannot be an answer-visibility tactic. You cannot claim the independence as a trust feature and sell the placement as a citation strategy.
Where the targeting breaks
SE Ranking also ran a semantic-relevance check on the ad against the prompt it appeared on. About 14.35% of ads, roughly one in seven, had no topical connection to what was asked. Its named examples: a dating-app prompt served a clothing-retailer ad, a newspaper-subscription prompt served an electricity-provider ad.
The mismatch rate is not evenly spread. It was 2.6% in Pets, and above 50% in both Relationships and News & Politics.
Ad frequency skews toward exactly the categories where a sponsored unit sitting beside a generated answer is most uncomfortable: ads appeared on 28.69% of healthcare prompts and 28.76% of News and Politics prompts.
Who is already spending
The advertiser base is thin and concentrated, which is what a six-month-old ad product looks like.
One comparison site, BestMoney, accounted for at least 13.56% of all ChatGPT ads in the dataset. Within individual niches the concentration is sharper still: Hungryroot took 79.92% of Food and Beverage placements, Booking.com 62.57% of Travel, and Robinhood 52.24% of Finance.
SE Ranking also ran its own campaigns for about two weeks, reporting more than 97,000 impressions and 1,263 clicks, with what it describes as “very few sign-ups”. It does not publish the sign-up count, so that last part is an adjective too.
What this study is, and is not
SE Ranking sells SEO and AI-visibility software. This is vendor-published research, not peer-reviewed work, and no independent party has replicated it.
Two things partly offset that, and neither cancels it. The methodology is disclosed — sample size, market, niche count, and its own advertiser-side test. And the central finding cuts against the pitch a vendor in this space would most want to make, which is that AI ads and AI visibility are the same product.
The limits that matter when you quote this:
- US only, single market.
- Commercial-intent prompts only. Nothing here describes informational or navigational queries.
- Free and Go tier users only. SE Ranking says so itself: ads currently show only on those tiers, so every figure describes the ad-supported slice of ChatGPT’s users, not the whole base.
- The study measures citations, not answer text. It does not test whether ads bias what ChatGPT says. Nobody outside OpenAI can currently test that.
- “Cited as a source” is SE Ranking’s own extraction method, and the exact matching logic is not fully specified.
What to do with this
If you are being sold ChatGPT ads as a route to being cited by ChatGPT, ask for the citation rate. The only public number is 3.63%, and it was not published by anyone selling ad inventory.
If you are buying ChatGPT ads as media, buy them as media. One in four commercial prompts is real reach on a surface with almost no competition in most niches yet, and the concentration figures suggest the auction is not crowded. That is a legitimate case. It is a different case.
And if you are tracking the economics rather than the tactics: this is the sharpest version of the zero-click problem yet. In search, the publisher at least had a link that might get clicked. Here, a monetised answer is assembled from sources that are not paid, next to an ad slot that is, and 96.37% of the time the party paying is not even among the sources being drawn on.
What we do not know yet
Whether the ad load rises now that five more markets are live. Whether the citation-overlap rate moves as more advertisers arrive. Whether OpenAI ever publishes a trust or dismissal figure to go with six months of adjectives. Whether any independent party replicates the 50,006-prompt study on a non-US market, or on the paid tiers where ads do not currently run.
We will update this post when a second measurement exists. Right now there is one.