How to Evaluate Different Types of AI-Generated Answers

Testing Google’s new AI chat feature, the author received varied answers on screen-time limits for teenagers and the risks of daily aspirin. He argues that accuracy alone is insufficient, introducing a typology of four answer styles: factual claims that can be verified, interpretive responses that depend on evidence selection, constructive outputs like drafts or plans that lack a single correct answer, and strategic advice that mixes data with judgment.

Each type requires a different verification approach—checking citations for facts, probing omitted evidence for interpretations, assessing relevance and tone for constructions, and confirming personal suitability for strategic guidance. The librarian, who leads AI-competency initiatives, suggests follow-up questions to uncover hidden assumptions and recommends consulting experts when stakes are high. Recognizing the answer type, he says, is the first step toward responsible AI use.