What is knowledge cutoff?
Also called: training cutoff, training data cutoff
Definition
A knowledge cutoff is the point in time after which a language model has no information from its training data. Anything that happened, launched or changed after that date is unknown to the model unless it retrieves it through search.
Explained
Each model version has its own cutoff, usually months before its release. A brand founded or renamed after the cutoff will not appear in answers that come from the model's memory, and older facts, such as previous pricing, can linger.
Two levers help. In the short term, grounded answers (with search) can pick up current pages, so being retrievable and well covered on trusted sites matters. In the longer term, being widely and consistently described on the web increases the chance the next model's training data includes you accurately.
Example
A startup that launched last spring is never named when ChatGPT answers without search, but appears in some answers when search is used, because recent review articles mention it.