Term: Knowledge cutoff
~5 min read
Estimated time: ~5 min read — for the in-app brief plus opening the primary source.
What this is
A knowledge cutoff is the last date the model’s built-in training reliably covers. It is not live news unless the product adds retrieval or browse.
Everyday example
You ask Grok or ChatGPT what a regulator published last Tuesday. Without browsing or your own files, the model may guess from older training — that date gap is the knowledge cutoff.
A knowledge cutoff is the last date the model’s built-in training reliably covers — it is not live news unless you add retrieval or browse.
- Asking “what did the regulator publish last week?” without a live source is a lottery.
- Products may add browsing or RAG to close the gap — ask whether yours does.
- Cutoff is why two tools disagree on a recent event.
- Your internal facts need your systems, not the model’s memory.
Next action: For time-sensitive work, require a dated source in the answer — or do not use the chatbot alone.
What changes in how you lead
How decision rights, process, and ownership should change.
- Legal, Risk, and comms do not treat model memory as current law or market fact.
- Procurement asks how the product stays current.
Compare related ideas
Knowledge cutoff vs RAG
Cutoff is what the model was trained up to. RAG fetches your current documents at ask-time so the answer can include what the model never memorized.
Open RAGDeep dive
Cutoff explains why two tools disagree on a recent event.
Browsing and RAG close the gap — ask whether your product actually does that.
Your internal facts belong in your systems, not in the model’s memory.
Legal, Risk, and comms require a dated source on anything they will act on.