How we measure ChatGPT
We send each question and your country to ChatGPT through a data provider, and collect the answer as it appears in the ChatGPT interface. No separate language setting goes with it, so the language of the question sets the language we ask in. The backup route covers fewer countries.
Does ChatGPT show sources?
Yes, in our test. All 12 ChatGPT answers in our test of 4 November 2026 listed sources, from 12 to 23 per answer. That test is small, and a listed source does not prove that ChatGPT used it in the text.
What can go wrong with ChatGPT
ChatGPT can answer the same question differently each time, which is why the paid scan asks each question five times. We do not know whether a collected answer came from a signed-in account or which plan it used, and the model name does not always come back. A scan also cannot show whether ChatGPT would search the web for a real user. OpenAI says sites that opt out of OAI-SearchBot will not be shown in ChatGPT search answers, though they can still appear as navigational links.
If ChatGPT gives no answer, we retry and use a backup route. Failed answers stay out of the count, and a system with too few answers is labelled Not measured.
Which scan includes ChatGPT
ChatGPT is in both the free scan and the paid scan. The free scan asks each question once, and the paid scan asks each question five times.
How to read the result
You see a count: named in X of Y answers. Each cell opens the full answer from ChatGPT, with your name marked and any sources that came with it.
What we do not promise
Nobody controls what an AI system says. We show what it says now and what you can change. Read our method