Glossary

What is Knowledge Cutoff?

The date after which a model's training data ends. Anything you launched after it doesn't exist in the model's memory — only retrieval can surface it.

Definition

A knowledge cutoff is the end date of a model's training corpus. Ask an assistant (without browsing) about anything after that date and it either admits ignorance or hallucinates. Every model has one; they typically trail release by months.

Consequence: your latest product, rebrand, or pricing exists in AI answers only via retrieval — until the next training cycle absorbs it.

Why it matters

Cutoffs split AI visibility into two clocks. The slow clock: training-data presence, which updates per model release and rewards years of consistent, corroborated coverage. The fast clock: retrieval, which can pick up today's page today. Brands that only think in the slow clock miss the fix available now — publish crawlable, answer-shaped, corroborated pages and win the grounded answers immediately.

Frequently asked

My company rebranded and AI still uses the old name. Why?

Training memory. Until models retrain, the old identity persists in ungrounded answers. Mitigate via retrieval: authoritative crawlable pages connecting old and new names, updated third-party profiles, and consistent schema.

How do I benefit from the next training cycle?

Be widely and consistently described before the snapshot: reviews, press, documentation, community mentions. Models 'know' brands the corpus talked about.

Related terms