How we capture ChatGPT ads
ChatGPT Ad Library is a public record of the sponsored ads that appear inside ChatGPT. This page explains exactly how the data is collected, what each number means, and where the limits are.
September 2026 data: the full-corpus evidence record was re-extracted and verified on September 6, 2026, covering ads observed from July through September 2026. The library includes at least 11,103 advertisers across 415,289 placements in 970 niches.
How the data is captured
We probe ChatGPT with realistic consumer prompts drawn from ~100 prompts per niche — the kinds of questions real people ask when researching a purchase. When ChatGPT returns a sponsored card, we record three things: the creative (image + copy), the advertiser, and the exact prompt observed with it. Collection can continue beyond the public browse snapshot, but the public figures on this site remain tied to a dated evidence record. The current public snapshot was observed through July 16, 2026.
Provenance
Every capture stores the raw response it came from, with a timestamp. That means each datapoint in the library is backed by evidence — the ad, where it appeared, and when — rather than an unverifiable claim. This is the trust model the ad-intelligence category has largely lacked.
What the numbers mean
Counts describe our captured sample, not OpenAI's internal data. A "placement" is one observation of an ad on a prompt; an "advertiser" is a distinct brand we've seen running ads; a "creative" is a unique ad. Share-of-voice figures are computed over the placements we captured. We don't measure ad spend — advertiser presence and prompt coverage are evidence of observed presence, not a measure or proxy for spend, bids, conversions, or return on investment.
Limitations
Our sample is a subset of everything running inside ChatGPT — coverage is deepest in the niches we probe most. The ad surface changes frequently (new formats, verticals, and geographies), so figures move over time. We publish these caveats openly because an honest sample is more useful than an overstated one.
Explore the public ad library or the personalized ads intelligence built on top of it.