Knowledge in an AI assistant does not age uniformly. A fundamental concept may remain stable for years, while a price or policy changes tomorrow. Expiry tags make freshness a property that retrieval systems and people can check.
Classify the pace of change
Separate stable, periodically changing and event-driven knowledge. Legal policies, tutorials and historical records should not share one expiry rule.
Add an owner and review date
Every critical asset needs an owner and a review-by date. A label without someone responsible for acting is merely metadata.
Specify review triggers
A new product version, legal change, price list or incident may require immediate review before the scheduled expiry.
Define behaviour after expiry
The system may exclude, deprioritise or warn about material depending on the risk. It should not present stale knowledge with unchanged confidence.
Monitor the expiry backlog
Track assets without owners, overdue reviews and answers using expired material. Adjust capacity before the backlog becomes widespread.
This framework is an original DigitalNow editorial methodology.