How we capture ChatGPT ads

The ChatGPT Ad Library is a public record of the sponsored cards that appear inside ChatGPT in France and the United Kingdom. This page explains how the data is collected, what each number means, and where the limits are.

Every morning, the same questions

We keep a fixed set of realistic, brand-neutral prompts: the questions people actually type when they research a purchase, about 50 per sector and market, written in French for France and in English for the UK, across 18 sectors. Every morning we ask ChatGPT each of them from the market it belongs to, with the consumer experience that carries sponsored cards. When an answer comes back with a sponsored card, we record the creative (picture, advertiser, headline and copy), the landing page it points to, and the exact prompt it was observed with.

What the numbers mean

A placement is one observation of one ad on one prompt on one day. A creative is one distinct ad: the same advertiser, landing page and copy seen on ten prompts over a week is one creative and ten placements. An advertiser is a distinct brand, identified by the domain its ads land on. “Seen N×” counts placements. The ad rate is the share of prompts whose answer carried at least one sponsored card.

Counts describe our daily sample, not the platform. We do not measure spend, bids, impressions or results: presence on our prompts is evidence that an advertiser is running, not a measure of how much.

Provenance

Every placement keeps the answer it came from and the time it was captured, so each datapoint on the site is backed by an observation rather than a claim. A creative's picture is kept as it was served on the first day we saw it.

Limitations

Our prompts are a subset of everything people ask, so coverage is deepest in the sectors we probe and the library says nothing about sectors we do not. Sponsored placements change by market and by day: an advertiser absent today may appear tomorrow, and the set of markets where ChatGPT shows ads evolves. We publish these caveats because an honest sample is more useful than an overstated one.

About

The library is built and run by GetMint, the AI visibility platform. Something missing, a brand you expected to see, or a question about the data: write to us from the GetMint site. Start browsing in the library.