This is how to make AI compare options against your real criteria

This is how to make AI compare options against your real criteria.

A pros-and-cons list gives every feature a vote. Your life does not. A chair that looks best may lose because it cannot fit under the desk. A cheaper chair may lose because its return window is too short to discover whether eight-hour days hurt. The useful comparison begins when the criteria stop being equal.

Turn preferences into a brief

Separate dealbreakers from preferences. A dealbreaker removes an option. A preference changes how attractive it is. Then tell AI which facts are public, which are your judgments, and which are missing.

The criteria prompt:

Compare three fictional desk chairs for my home office. Do not pick a winner.

My dealbreakers: total price must stay under $550, it must fit beneath a 29-inch desk, and I need a return option after trying it at home.

My priorities, in order: comfortable for two four-hour work blocks, adjustable arms, breathable material, quiet casters on hard flooring, then appearance.

Chair North: $420, 27 to 31 inches high, mesh back, adjustable arms, 30-day returns, caster material unclear.
Chair Field: $510, 28 inches high, upholstered, fixed arms, 60-day returns, soft casters.
Chair Harbor: $360, 27 inches high, mesh back, adjustable arms, returns unclear, soft casters.

Build a table with dealbreaker status, evidence for each criterion, missing facts, and the tradeoff I accept if I choose each option. Keep preference judgments separate from product facts.

Swap in public facts and your own harmless priorities. The dealbreaker and preference split is the part that matters.

Level up the ask:

Rewrite my comparison prompt so each criterion is observable, missing facts remain visible, and no option wins merely because it has more listed features. Tell me what you changed and why, then run the improved comparison.

Read the blank cells first

The missing return policy for Chair Harbor may matter more than five positive features. The fixed arms on Chair Field are not automatically bad, but they conflict with a stated priority. A useful table makes those tensions visible instead of converting everything into an unexplained score.

If you use weights, keep them transparent. Ask how the result changes if comfort matters twice as much as appearance, or if the budget drops by $75. A robust comparison should show what wins under different values, not manufacture one objective winner.

Finish outside the table

Use the table to create a short fact-finding list. Confirm measurements, return terms, exact model details, and stock with the responsible source. Then make the decision yourself. Criteria help AI organize your values; they do not give the system values of its own.

The technique scales from chairs to travel routes and service quotes: real criteria first, evidence second, gaps visible, tradeoff named.

• Share only: Public option details plus harmless preferences, a rounded budget, and practical constraints; exclude account data, private messages, addresses, and sensitive records.
• Verify: Every product fact, price, dimension, warranty, policy, and availability claim with the manufacturer, seller, or other responsible first-party source before choosing.

This is how to use AI.