How to know if you can trust an AI’s answer to your question

A University of Virginia librarian and dean proposes sorting AI-generated answers into four types: factual (checkable against a source), interpretive (built on evidence but no single right answer), constructive (written rather than discovered, like a eulogy) and strategic (advice weighing personal risk, like whether to take daily aspirin). She notes Google's AI search answered a screen-time question by citing a number, then complicating it, and answered an aspirin question by citing risk factors and offering to tailor guidance if given her medical history.

Her advice: verify factual claims against the cited source rather than trusting the AI summary itself, and for interpretive or strategic questions, ask the system what evidence or personal details could change its answer. She stresses AI presents all four answer types in the same confident tone, making it easy to mistake a constructed or strategic answer for a simple fact.

The full dispatch is available from the source below.