An AI response can sound equally confident whether it is reporting a fact, interpreting evidence, creating something new or recommending what you should do. Those answers require different levels of scrutiny.

AI answers are increasingly appearing before users ever reach a traditional search result. That makes one question more important than simply asking whether an answer sounds correct: What kind of answer is the system actually giving you? Leo S. Lo, dean of libraries at the University of Virginia, proposes four broad categories — factual, interpretive, constructive and strategic — because each requires a different way of checking whether the response is useful enough to trust.

A factual answer makes a claim that can usually be checked against evidence, such as a historical date or scientific fact. The practical response is straightforward: follow the cited source and verify the claim. An interpretive answer is harder because the facts may be accurate while the conclusion depends on which evidence the system emphasized. A question such as whether remote work improves productivity or how much screen time is appropriate for teenagers may have several defensible answers. In those cases, users should ask what evidence was excluded and whether another interpretation fits the same facts.

Constructive answers are different because they are created rather than discovered. A cover letter, lesson plan or rewritten paragraph does not have a single correct answer. Accuracy matters less than whether the result fits the audience, purpose and voice. A polished response can still be unusable if it sounds generic, misses the intended tone or fails to reflect the person who will actually use it.

The highest scrutiny belongs to strategic answers — questions about what someone should do. Medical choices, major purchases and other decisions combine facts with risk, goals and personal circumstances. Lo uses daily aspirin as an example: an AI can accurately summarize medical guidance while still lacking the information needed to determine whether that guidance applies to one specific person.

The useful habit is therefore not simply “double-check AI.” It is to match the verification method to the kind of intellectual work the AI performed. Facts should be sourced. Interpretations should be challenged. Constructive work should be judged against purpose. Strategic advice should be tested against context, consequences and expertise. As AI-generated answers become a routine layer of search, recognizing the type of answer may be one of the most practical forms of AI literacy.