When to use it: Any time you’re deciding whether an AI output needs verifying. This is the mental model under the pressure test.
When to avoid it: N/A. It’s a reference. Keep it close.
Check these every time. It invents when reaching for something it doesn’t actually have: specific numbers and stats; sources, citations, links, study names; dates and counts; quotes and who said them; recent events, prices, anything that changed lately; legal, medical, or financial specifics.
You can usually relax here. It’s steady when working with what you gave it: rephrasing or tightening your own text; outlining and structuring; brainstorming options for you to pick from; summarizing a document you pasted in; changing tone, or explaining a concept you can sanity-check yourself.
Expected result: You check the claims that actually matter and stop re-reading the rest. Less time proofreading, same safety.
Where it breaks: The line moves with the stakes. “Relax” assumes the output stays internal. The moment it goes to a client, or a number carries real weight, treat it as “check,” even for a summary.
What to check before using it: Whether the output is leaving your hands. If yes, default to checking.