Skip to content
Glossary

What is a knowledge cutoff?

A knowledge cutoff is the date after which a model's training stopped, so it has no built-in awareness of anything published after that point unless a separate tool fetches it live.

Every model version has a knowledge cutoff: the point at which the text used to build it stops. Anything published after that date simply isn't part of what the model absorbed, from the model's own memory alone, it has no way to know it exists.

This explains a specific, common failure: an AI can state an old price, describe a discontinued product, or name someone who's left a company, confidently and without any warning, because its built-in knowledge is frozen at that cutoff. When an AI answer does reflect something current, this week's news, today's price, that's a sign a separate tool fetched it live at the moment of the question (see web-search-tool), not that the model's own knowledge had updated. It also follows that a model's cutoff date has nothing to do with whether a page written after that date can still be cited today: live citation runs through search at question time, not through training (see training-vs-search-crawler).

Related

Want to know where you actually stand on this? Run a free visibility check or try the free tools.