How long until ChatGPT updates what it says about my brand?

The speed depends on which mechanism ChatGPT uses#

ChatGPT answers your brand question in one of two ways, and each updates on a completely different clock.

• Live web search. When a question needs current, specific, or local detail, ChatGPT searches the web, reads a handful of pages, and cites them. This path reflects whatever your pages say right now. In Prefer’s September 2026 test, 47 of 50 ChatGPT answers were grounded in live web search, so this is the common case for brand and category questions.
• Training data. When ChatGPT answers from what it learned during training, that fact only changes when OpenAI trains and ships a new model. New models arrive months apart, so a detail the model “remembers” can be well behind your live site.

So a fresh page can start showing up in searched answers within days to a few weeks of being re-crawled, while a stale fact baked into the model can persist until the next model. The two are not the same, and confusing them is why the timeline feels unpredictable.

What you actually control#

You cannot set OpenAI’s re-crawl schedule, but you can remove every reason for it to be slow. OpenAI does not publish how often OAI-SearchBot re-crawls, so treat re-crawl timing as out of your hands and focus on the parts that are not.

• Keep the crawlers allowed. If OAI-SearchBot is blocked in your robots.txt, no update can reach ChatGPT’s searched answers. Check this first. Blocking GPTBot only affects training.
• State the fact plainly on the page. Put the correct, current claim in clear text near the top of the relevant page, not buried in a PDF or an image, so a live read can lift it.
• Give it a stable URL. A fact on a page that keeps moving is a fact ChatGPT keeps losing. One durable URL per key claim re-crawls and re-cites more reliably.
• Refresh the sources that describe you. Because searched answers repeat what third-party pages say, an outdated review, listicle, or directory entry will keep feeding the old version even after your own site is fixed.

How to tell which clock you are on#

You can usually spot which mechanism produced an answer, which tells you how to fix it. If ChatGPT’s reply links to or names current web sources, it searched, and the fix is a content-and-crawlability job that moves at crawl speed. If it answers confidently with no sources and repeats something out of date, it is likely drawing on training data, and no on-page edit will change it until the next model or until a searched answer overrides it. In our September 2026 run, grounded answers cited about 3.8 sources each, so the presence of links is a reliable tell that the searched path is in play. When the stale claim lives in the model’s memory, the practical move is to flood the searched path with correct, well-sourced pages so the live answer wins, rather than waiting on a retrain you cannot schedule.

The honest summary is that there is no guaranteed number of days. Searched answers move at the speed of the next crawl, memory-based answers move at the speed of the next model, and your job is to make the searched path fast and correct. To see whether the crawlers behind ChatGPT can even reach your pages today, run your robots.txt through the free Crawler Access Checker, then run a free AI visibility audit to see what ChatGPT says about you right now.

Sources

1. In Prefer's 2026-09-02 ChatGPT run, 47 of 50 answers were grounded in live web search rather than answered from memory, and grounded answers cited about 3.8 web sources each, so most brand answers reflect current pages rather than old training data. (Prefer AEO loop run, 2026-09-02, dated research note)
2. OpenAI's Help Center explains that ChatGPT searches the web and cites the sources it reads, and that a site becomes eligible to appear by allowing the OAI-SearchBot crawler. OpenAI Help Center: Searching the web with ChatGPT
3. Access is the floor, not the finish: allowing OAI-SearchBot lets ChatGPT search read your pages, but a crawler still has to choose to visit before an update can show up. Prefer AI Crawler Access Checker