The two clocks: training and retrieval
Every ChatGPT model has a knowledge cutoff: the date its training data ends. Facts absorbed in training change only when the underlying model is replaced, which happens on the scale of months, not days. You cannot accelerate that clock.
Retrieval is the fast clock. When a question needs current information, ChatGPT searches the live web and reads pages in real time. Anything it can retrieve, it can update within days of your pages and sources changing. The same split applies to Gemini, Claude and Perplexity, each with its own cutoff and its own retrieval behavior.
What this means for your brand
If an assistant answers questions about your category from memory, old facts persist until the next model version. But most commercial questions, prices, availability, comparisons and recommendations, trigger retrieval. That means the practical update speed for your brand is set by how quickly your sources reflect reality: your site, the directories that list you, the review platforms and the articles that mention you.
The failure mode we see most in audits: a company updates its pricing page but the three comparison articles ChatGPT actually cites still show last year’s numbers. The assistant is current, its sources are not. Diagnosing that gap is covered in why AI has wrong information about your business, and the repair steps in how to correct wrong information in ChatGPT.
How to get on the fast clock
Make every important brand fact retrievable: stated in plain text, on a crawlable page, in a place an assistant would look. Keep third party sources synchronized when facts change, starting with whatever the assistants currently cite. And re-test monthly across ChatGPT, Gemini, Claude and Perplexity, because each moves at its own pace. If you want to know which clock your brand facts are on today, a free AI visibility audit shows you the current answers, their sources and their age, walked through on a 30 minute call.
