Does ChatGPT search read my live site or a saved copy?

ChatGPT search does not usually read your live site while it writes an answer. As of October 2026, OpenAI's documentation describes separate ways your pages reach ChatGPT, and most of them depend on a copy fetched before the user asked.

• OAI-SearchBot crawls the web for ChatGPT's search features. OpenAI says it "is used to surface websites in search results in ChatGPT's search features." What it fetched earlier is what search can draw on.
• Partner search providers supply results too. OpenAI's help article says ChatGPT search "sometimes partners with other search providers" and rewrites the user's question "into one or more targeted queries" sent to them. Those providers answer from their own indexes.
• ChatGPT-User fetches pages when a user's action triggers it. That is the closest thing to a live read, but OpenAI says it is "not used to determine whether content may appear in Search."

So the practical answer: ChatGPT sees the version of your page that a crawler or provider stored, unless a user action prompts a fresh fetch. If you changed a price on Monday, ChatGPT may quote the old one until the stored copy catches up. For a deeper read of each bot, see OpenAI's crawler documentation, digested.

Each path has its own control, and OpenAI states that "each setting is independent of the others." That matters, because a single blanket AI block in robots.txt can switch off more than you meant.

Blocking OAI-SearchBot has a documented consequence: sites opted out "will not be shown in ChatGPT search answers, though can still appear as navigational links." Here's what navigational links mean when you block OAI-SearchBot. GPTBot is not a search lever in either direction. Allowing it does not get you into search, and blocking it does not remove you. The OAI-SearchBot definition covers the search crawler in more detail.

Several separate things can keep an edit out of ChatGPT's answer. Most are outside your control, so the useful move is to rule out the ones that are inside it.

1. The stored copy is older than your edit. OpenAI does not publish how often OAI-SearchBot recrawls a page. Your logs show when it last visited. Here's more on whether AI crawlers visit as often as Google.
2. A partner provider's index is stale. If the answer came through a provider, its crawl schedule decides freshness, not OpenAI's.
3. The answer used no search at all. Search in ChatGPT is model-discretionary. Without it, the model answers from training data with a fixed cutoff. OpenAI's model docs list a February 16, 2026 cutoff for the GPT-5.6 family and August 31, 2025 for the Instant path (chat-latest). Anything newer cannot appear in an ungrounded answer.
4. The new content only appears after JavaScript runs. Vercel found that ChatGPT and Claude crawlers fetch JavaScript files but "don't execute them. They can't read client-side rendered content."
5. Your CDN blocks the crawler. OpenAI says eligibility needs both an OAI-SearchBot allow rule and a host or CDN that admits traffic from its published searchbot IP addresses.

You can fix numbers 4 and 5. You cannot speed up numbers 1 to 3 on demand.

Testing "does ChatGPT see my update" is harder than it looks.

• Query rewriting. ChatGPT sends its own rewritten queries to providers, so the page it retrieves may not be the one you edited. A different URL on your site, or a third-party page, may be the source.
• Memory and instructions. OpenAI says that if memory is enabled, ChatGPT may use saved memories when rewriting a search query. Test logged out or with memory off.
• Retrieved is not cited. OpenAI's API docs say the number of sources consulted "is often greater than the number of citations." Your updated page can be read and still not be named.
• Being in Bing is not enough. In Otterly's 2026 experiment, two sites stayed in Bing's index and ChatGPT's crawlers kept visiting, yet ChatGPT cited them effectively zero times. One test, but it shows indexing is necessary, not sufficient.

Work from access outward. Each step removes one possible block.

1. robots.txt per bot. Confirm OAI-SearchBot is allowed, separately from GPTBot. The free ChatGPT search block check reads this for you.
2. Firewall and CDN. A robots.txt allow means nothing if the edge returns 403. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.
3. Server logs. Look for OAI-SearchBot and ChatGPT-User requests on the edited URL, with dates and status codes. Use this free way to see which AI bots crawl your site.
4. Raw HTML. Make sure the changed fact is in the HTML before JavaScript runs. The JavaScript rendering check diffs the two.
5. Other indexes. Check Bing Webmaster Tools separately from Search Console, following index status engine by engine.

If all five pass, the remaining lag sits with OpenAI's crawl schedule and model choices. Nobody outside OpenAI controls those.