Methodology

We can't match the largest trackers on volume. Instead we keep a smaller, checkable public record: a fixed prompt panel, run in the same frozen wording across markets, with a screenshot behind every entry.

The panel is the product

Our core panel is B2B and martech: 30 fixed prompts across martech & CRM, work management & SaaS, finance software, HR & recruiting, dev tools and professional services. It is the basis of every weekly report. Seasonal consumer panels run alongside it as monthly specials: Q4 2026 covers consumer electronics, retail & gifting, travel, personal finance and beauty & fashion; Q1 2027 switches to fitness & wellness (gear, apps, gyms, nutrition, sleep; no medical questions).

Markets: same prompts, same week

Every panel runs in the same week in five markets: the United States (lead market), the Netherlands, Germany, France and the United Kingdom. Each market has its own dedicated free-tier ChatGPT account, registered and used in that market. Prompt wording is frozen per market and language: each prompt has a fixed key (for example b2b-core-01-nl), and a reworded prompt gets a new key instead of replacing the old one. Every entry records its market, language and an internal account label. We don't rotate locations or use tricks to appear in a market we have no account in. A market we can't capture properly in a given week is reported as missing, not estimated.

Capture daily, publish weekly and monthly

Panels are captured daily at a human pace, in fresh chats, one session at a time, with no rate-limit evasion. Data accumulates daily in our log, but we publish only a weekly report every Monday and a monthly report on the 1st, after editorial review.

What we record

When a sponsored card appears we save a full-page and a cropped screenshot, and log the timestamp with its time zone, market, language, frozen prompt key, advertiser, headline, body text, landing URL, format and account. Landing URLs are read by hover and never clicked, so we don't create fake clicks.

Verification and corrections

verified a screenshot is on file and the fields match it. unverified logged, but not yet cross-checked. needs review something is ambiguous or disputed. Only verified entries count in reports. Corrections stay visible: a wrong entry is marked retracted with a note, never silently deleted.

Official reporting

Under its designation under the EU Digital Services Act, OpenAI must publish an EU ad repository by January 2027. Where official reporting like this exists, we compare our observed panel sightings with it and note where they differ. It is not yet settled whether that repository will cover ads shown inside chat answers, so we don't assume it does. Our data is what we observed on our own panel; it is not data from the AI provider.

Limits

The panel is small and fixed by design. Ad selection varies by conversation, language and location, and in some markets by personalisation, so what we see is not universal. Comparisons between markets are indicative. The ad surface changes often.

Privacy

We crop personal information from screenshots before publishing.

Corrections

If you are an advertiser and an entry is wrong, contact [email protected] and we will correct or annotate it.