Model Knowledge Cutoff
The knowledge cutoff is the date beyond which an LLM has no training data — historically a hard limit, now mitigated by retrieval-augmented generation in modern AI search.
In context
When ChatGPT or Claude answers a query without retrieval, they rely on training data with a cutoff (typically 6–18 months old). When they answer with retrieval (the default in ChatGPT Search, Perplexity, Claude Search, etc.), they can cite live web content. The implication for SEO: if your content publishes after the model's cutoff, retrieval is the only way it gets cited. Always optimise for the retrieval path — that's the live citation surface — and treat training-data inclusion as a secondary, harder-to-influence path.
A Toronto SEO article published in May 2026 cannot appear in any model trained before that date through model recall — but it can and does appear via ChatGPT Search retrieval immediately after indexation.
Have a question this didn't answer?
Book a free 60-minute strategy call. We'll explain how this concept applies to your specific situation.