AI search · Glossary
What is a knowledge cutoff?
A knowledge cutoff is the date up to which an AI model's training data runs. Without web search, the model can't know about anything that happened or changed after that date.
Written by Noor KhanWhy Knowledge cutoff matters
It explains many outdated AI answers. OpenAI says its models are trained on data up to a certain point and don't include later events unless tools such as search are used, so a price or service change after the cutoff only reaches answers through search.
How Knowledge cutoff works
- A model is trained on text collected up to its cutoff, then released and used for months afterwards.
- Anthropic, for example, publishes a reliable knowledge cutoff for each Claude model: the date through which its knowledge is most reliable.
- Assistants with web search can add newer information, but only when they search.
How to apply it
- Expect answers from memory to lag behind changes to your business.
- Keep current facts easy to find on your site and listings, so search-based answers pick them up.
- Test with search on and off to see which layer an outdated answer comes from.
Example
A restaurant changed its opening hours after a model's cutoff. The assistant gives the new hours when it searches, and the old hours when it answers from memory.
Common mistake
Waiting for a new model to fix an outdated fact instead of correcting the pages that search reads now.
put this into practice
Not sure how Knowledge cutoff applies to your site? Book a strategy call, request a free gap snapshot, or ask the assistant.