Knowledge cutoff · Training data cutoff date
The knowledge cutoff is the date beyond which a language model has no more training data. For anything later, it must search the web (grounding) or risk answering with outdated or invented information.
Standalone definition · 33 words
Why does it exist?
A model is trained on a corpus frozen at a date, and its parametric memory does not update by itself. LLMrefs devotes analyses to ChatGPT’s and Gemini’s cutoff dates, which condition the use of search.
How does it relate to citations?
For recent topics, an engine triggers a search and cites sources. For stable topics, it may answer from memory with no citation at all. A brand that appeared after the cutoff does not exist in the model’s memory: only live search can make it appear.
What should you do?
Check what each assistant knows about the brand without web search, publish clear, crawlable pages for live retrieval, and maintain the third-party sources that will feed future training runs.
Frequently asked questions
Is the knowledge cutoff public?
Vendors often state one. It is sometimes approximate and differs from the model’s release date.
Does a model know it is out of date?
Not always: without a search tool, it may answer as if its cutoff were the present.
Related terms
Further reading
- Profound vs Otterly vs Ahrefs Brand Radar: query fan-out comparison
- ← Back to the GEO glossary (114 terms)