How to Ask an AI to Compare Two Things Fairly
Why the default response to a comparison question is two separate summaries, and the specific instruction that forces an actual comparison instead.
Published April 29, 2026
"Compare X and Y" sounds like a comparison request, but it's underspecified enough that most models default to something else: a paragraph describing X, a paragraph describing Y, and a brief closing line that doesn't really compare them so much as restate that both have strengths.
Why models default to parallel summaries
Describing each option independently is the safest, most defensible response to an open-ended comparison request, it can't really be wrong. Actually comparing requires taking a position on specific criteria, which is a slightly riskier kind of answer, and without explicit instruction to do exactly that, most models take the safer path.
What forces a real comparison
Naming specific criteria to compare on, and explicitly instructing the model to state which option wins on each one, with a reason, rather than describing each option in isolation. That combination, named criteria plus a direct instruction to compare rather than describe, is what actually changes the shape of the answer.
Why context matters as much as the criteria
"Which is better" doesn't have a context-free answer for most real comparisons, a database that's the right call for a small team's simple app is often the wrong call at a different scale. Including the actual use case turns a generic comparison into one that ends with a specific, useful recommendation instead of an "it depends" that leaves the real thinking undone.