Skip to main content
ChatGPT serves many pages from its own cache rather than fetching them live. Peec’s investigation (full study, Sep 2026) confirmed it via ChatGPT’s lockdown mode — a setting that cannot browse live pages but can still return offline or cached results. Asked “show me the latest version of homepage” for major SEO publishers (Search Engine Journal, Search Engine Land, Search Engine Roundtable), lockdown-mode ChatGPT returned the pages anyway — served from cache, not fetched live. Why a cache has to exist: live retrieval is too slow at ChatGPT’s scale. Per the HTTP Archive Almanac, 58% of pages take longer than 0.8 seconds to load. Caching also lets OpenAI front-load the expensive work — computing embeddings in advance and filtering out low-quality pages before they reach the pipeline. The crawl behavior matches. Peec researcher Metehan Yesilyurt published a test website with one billion pages to measure how much ChatGPT would crawl: by the beginning of September 2026 it had crawled 6 million of those pages and was still going at 35,000 requests per hour. Crawling that aggressively only makes sense if the pages are headed into a cache or an index — see ChatGPT has its own search index. Practical check from the study: enable lockdown mode in ChatGPT settings and ask about your most important pages — it reveals whether ChatGPT has already cached them. Source: ChatGPT built its own search index, Peec (Tomek Rudzki), Sep 2026.