How does ChatGPT search the web? For free users in instant mode, barely at all. The answer is grounded in page titles and cached 200-character snippets from OpenAI’s own index, with no page opened 93% of the time, an SEO firm reports. Paid thinking mode adds scraped Google results and real page fetches.
- A study tracked 27,000 pages through ChatGPT's search pipeline: 58,000 URLs retrieved, 7,600 promoted to sources, 5,000 cited, but only 760 actually read by the model.
- A 200-character snapshot appears to be the primary representation of your site in ChatGPT's pipeline, and cached versions can lock that snapshot in place indefinitely.
- For publishers, the funnel is brutal: roughly 1 in 76 retrieved URLs gets read, which reframes the question of what "being cited by AI" actually means.
