ChatAdsWatch

It is independent of OpenAI, and says so on every page

The first thing to know is who is behind the data. ChatAdsWatch is not an OpenAI product. Its footer states that it is not affiliated with or endorsed by OpenAI, and the About page extends that to Google, Semrush and the advertisers shown. So this is a third party observing ChatGPT from the outside, not a window into OpenAI’s ad system. Support runs through an email address on a different domain (conversaic.io), and the About page points that out openly rather than hiding it.

The counts describe a captured sample, not every ChatGPT ad

The headline numbers are large. At review time the public snapshot listed 45,315 observed ads from 11,433 advertisers, 34,360 associated buyer prompts and about 194.8K captured appearances, in a snapshot dated September 17, 2026. Read them carefully, because the site itself does. It says the catalogue is “a dated snapshot, not a live census of every ChatGPT ad”, and that an observed appearance is one record in the collected sample, not an impression, a click, reach or spend.

The Methodology page is short and worth reading in full. It separates three kinds of information:

• Observed: an ad found beside a captured prompt in the imported source sample, with upstream appearance counts kept where they were supplied.
• Inferred: industry, topic and buyer intent, classified from the creative and its prompt. Low-confidence records stay “Other” or “Unknown”.
• Estimated: modelled figures such as CPC, which the terms describe as comparisons, not verified ChatGPT ad prices.

One gap: the page talks about an “imported source sample” and “upstream” counts but does not explain how the ads are actually collected, from which accounts, regions or prompt sets. Geography and targeting are marked unavailable. We would like that spelled out, since it decides how representative the sample is.

The inferred labels also need checking. In the public sample we saw, a Rocket Mortgage home-buying ad and a shipping-container ad were both tagged with the topic “Cloud & DevOps”. The methodology admits classification can be wrong and invites corrections by email, which is the right attitude, but treat topic labels as hints.

The homepage dashboard is a demo, and it is labelled as one

The homepage shows a market overview for an outdoor-apparel brand, with AI traffic, competitor share and an estimated CPC range. Those figures are not real measurements. The panel carries a DEMO tag and a note that it is an illustrative report demonstrating the interface. The real public material is the industry directory (19 industries, led by SaaS and productivity, finance and insurance, and home and garden), a short list of advertisers that pass a minimum-evidence threshold, and a few truncated ad cards. Entering your domain gives a limited market preview without a card.

There is also a free Chrome extension, AdContext, for collecting ads, product recommendations and brand mentions from your own ChatGPT sessions. It stores records locally in the browser, exports CSV or JSON, and is not synced to the website.

Pricing: free preview, Pro at $49 a month

The free public preview covers the catalogue and industry aggregates, a small sample of truncated ads, the methodology, and a limited market preview for your domain. ChatAdsWatch Pro was listed at $49 per month plus applicable tax at review time, shown against a struck-through $99. It adds full ad copy with the associated buyer prompts, complete library and competition access, saved market reports and 1,000 analysis credits per paid month.

Credits are only spent on new provider-backed analyses, which show a quote before you confirm. Browsing and filtering do not use them, failed analyses release them, and unused credits expire at the end of each month. It renews monthly, can be cancelled at any time, and checkout runs through Stripe. Sign-in is with a Google account.

Arabic support: none in the interface, and no sign of Arabic markets

The site is English only, with no language switcher and no right-to-left layout. The domain preview form sends the country as US, and every advertiser and industry in the public preview is an English-language, mostly US brand. We could not see the full Pro library, so we cannot rule out Arabic ads in it, but nothing on the public pages suggests Gulf or other Arabic-language markets are covered, and the methodology says geography is unavailable. The AdContext extension offers English and Chinese interfaces only.

Who it suits, and who should wait

It suits paid-media, growth and competitive-intelligence teams selling into the US who want early evidence of who advertises inside ChatGPT, what the ads say and which prompts trigger them. The public preview is a fair way to judge whether your industry has enough data before paying.

It is a weaker fit if you need spend, impressions or performance numbers, because the product deliberately does not claim them. It is also a weak fit for Arabic-speaking marketers targeting Arabic audiences: there is no Arabic interface and no visible coverage of those markets. And since the catalogue is a dated beta snapshot, check its freshness date before relying on it.