ChatGPT ads have disappeared from signed-out answers

In July, ads were a routine part of the ChatGPT answers we collect. A sponsored card from Monday.com under a question about CRM pricing, Guitar Center under one about beginner guitars, SoFi under one about buying a house. We built a whole feature around recording them.

Since 25 August we have not recorded a single one.

Not fewer ads, or ads on different prompts: none, across 1,878 answers. The answers themselves are fine, complete and cited. The ads are simply no longer in them. This post is what we found, how we checked it was not our own pipeline, and what it means if you track ChatGPT ads with BotScope or anything else.

TL;DR

• The last ad we recorded was at 12:01 UTC on 24 August. Our next ChatGPT collection, on 1 September, had none, and nothing since has.
• Like for like, the drop is total. The same 44 prompts in the same market went from 23 ads on 12 August to none on 1 September. Five prompts that each served an ad on 28 July served none on 7 October.
• It is not a broken scraper. On 7 October we ran ten prompts through three separate collection methods in the same hour: 30 complete answers, zero ads. Our data provider investigated independently and reached the same conclusion: the ads are absent from the response itself, not missed by a parser.
• We are not alone. A public index sampling about 100,000 ChatGPT answers a day has read 0% every day since 28 August, after a two-week slide.
• OpenAI has said nothing. Its help page still says signed-out users see ads. Either that has quietly stopped, or ChatGPT now holds ads back from automated traffic, which is how every AI-visibility tool collects. Nothing we have seen says ads stopped for real, logged-in users.
• If your tool’s ChatGPT ad coverage fell to zero in late August, that is a measurement change, not a market change. A zero there is not a finding.
• BotScope has paused ad recording. History up to 24 August stays. Answers since then show as not measured, not as zero. We check every morning and will post here, and email you, when ads come back.

What we saw

The obvious way to show this would be a weekly chart of ad coverage falling off a cliff. We are not going to show you that one, because it would overstate the case. The projects scanning ChatGPT after August are, mostly, not the ones that attracted ads before it, so a blended rate would drop even if ChatGPT had changed nothing.

So here is the comparison that holds the prompts and the market fixed.

The first row is the cleanest evidence we have. One project’s 44 prompts ran in the US on 12 August and again on 1 September, unchanged. More than half the answers carried an ad the first time. None did the second. Every one of the 44 answers came back with its full text and citations, so this is not a partial scan or an empty response being read as “no ads”. The answer arrived; the ad did not.

Checking it was not us

We have written before about a failure that looks exactly like valid data, so our first assumption was our own pipeline. A field renamed, a parser that stopped matching, a vendor change that dropped one column. All of those produce the same symptom: answers that look normal, with nothing in the ad slot.

The way to rule that out is to ask the same questions through methods built by different companies, each with its own parser. On 7 October we took the ten US prompts from our daily check, including the five that all served ads in July, and ran them in the same hour through our production pipeline and two independent scraping APIs.

Thirty answers, all with real content, and not one ad. Three methods, parsing ChatGPT in different ways, agreeing exactly. If one of them had broken, the other two would have disagreed with it.

We also asked our data provider to investigate. Their answer matched ours: the ads are not being returned in the ChatGPT response at all, so there is nothing for a parser to capture. They added that ads are generally more likely to appear for logged-in users than in unauthenticated sessions, which is how their scraper, and ours, collects.

We are not the only ones seeing it

Writesonic publishes a daily AI Ads Index built from roughly 100,000 ChatGPT answers a day. Its ChatGPT ad rate peaked at 30.45% on 9 July. It was 15.3% on 17 August, 8.2% on 24 August, 3.3% on 25 August and 2.0% on 27 August. Since 28 August it has read zero every day. On 6 October: “0 of 111,555 answers · 0 advertisers”.

LLM Pulse’s public ChatGPT ads library holds more than 62,000 ads. The newest were first observed on 27 August.

Neither says whether it collects signed in or signed out, but both line up with our own record to within days. Writesonic’s curve is worth a second look, too. It is not a switch flipping overnight but a two-week taper, which is what a staged rollout tends to look like.

What OpenAI says

Nothing, about this. Its help page on ads in ChatGPT has changed several times this year. In February it listed “When you’re logged out” among the places ads do not appear. In April, AdExchanger reported ads appearing for signed-out users anyway. By June the exclusion had gone, and the page as it stood on 4 October says: “Ads shown in logged-out experiences are appropriate for all audiences, while mature-content ads are limited to eligible logged-in users.”

So officially, signed-out users still see ads. In practice, nobody publishing data from signed-out collection has seen one since late August.

And ads have clearly not stopped overall. On 31 August, three days after the public trackers went to zero, OpenAI announced that ChatGPT Ads had reached “$1 billion in annualized revenue run rate” and was “used by tens of thousands of advertisers”. That money is being spent on impressions somebody is seeing. It is just not anybody collecting answers automatically.

Two explanations fit, and from the outside we cannot tell them apart:

1. OpenAI has stopped or paused ads in signed-out sessions and has not updated the help page.
2. ChatGPT now holds ads back from traffic it judges to be automated. That is ordinary practice in ad tech, where advertisers do not want to pay for impressions served to bots, and every AI-visibility tool collects its answers automatically.

For anyone measuring ChatGPT ads, the consequence is the same either way.

Why a zero is the dangerous number

An ad tracker that sees no ad has two very different stories it could tell you. Nobody is advertising on this question, or we could not see whether anybody was. From the outside the two look identical: a complete answer with nothing under it.

Tell the first story when the second one is true and you get a dashboard that quietly reports your competitors have stopped advertising. That is a confident, specific, wrong conclusion, and nothing in the data flags it.

That is why every ChatGPT answer collected since 25 August is marked not measured in BotScope rather than zero, so your Ads tab makes no claim about that period either way. We have applied the same rule to our other “nothing found” numbers, like citations and shopping results: we do not count something as missing unless we can show we were able to see it.

If you track ChatGPT ads

Check your own numbers. If ad coverage on your prompts dropped to zero, or close to it, in the last week of August, assume a change in what your tool can see before you assume a change in the market.

Ask how your tool collects. Signed out or signed in? Most AI-visibility tools, BotScope included, collect signed out. It is the only way to see what a new, anonymous person is shown, and it is what makes answers comparable from one scan to the next.

Be wary of logged-in measurement as the fix. Signed-in sessions may still get ads, but they come with their own problem. OpenAI says that with ads personalisation on, “select signals from your broader ChatGPT experience” help decide which ad you see. One tracker account’s ads tell you what that account is shown, not what your customers are. That is not a market view either.

If you buy ChatGPT ads, nothing here says your campaign stopped delivering. It says signed-out answers no longer show us one way or the other.

What BotScope is doing

Ad recording is paused. We still look for ads in every ChatGPT answer we collect, but since 25 August there have been none to record, and those answers count as not measured rather than no ads. Everything up to 24 August stays in your Ads tab, reports and API exactly as it was.

We check every morning. A fixed set of prompts that reliably served ads in July runs through our pipeline daily. The day ads appear in those answers again, recording resumes on its own, and we will update this post and email every customer whose projects track ChatGPT.

Everything else is unaffected. Visibility, citations, sentiment, rankings and product tracking all work on the answer itself, which is arriving complete. Only the ad slot is empty.

FAQs

Does this change my visibility scores? No. Ads have never been part of visibility scoring. An ad is a purchase, not a mention.

Will my ad history disappear? No. Everything recorded up to 24 August stays exactly as it was.

Do I need to do anything? No. If ads come back, recording resumes without any change on your side.

Why not just collect signed in? Because a signed-in account sees ads chosen for that account, which makes them a poor proxy for what your customers see, and the results would not be comparable with anything collected before. If that changes, we will say so.

---

Ads in AI answers are new, and the rules about who sees them are still moving. We would rather show you an honest gap than a confident zero. When the ads come back, so will the data, and you will hear it from us first.