Lesson 5 of 6
Check a response
A polished response can still miss a constraint or make up a fact.
The idea
Before running a prompt, write a short checklist based on the task. Separate format checks from factual checks. You can count rows or validate fields mechanically; factual claims need appropriate evidence. Asking the same model to critique itself may reveal problems, but it is not independent verification. Record failures so the next prompt revision addresses a concrete issue.
Compare two instructions
Before
A request with important details left open.
Is this answer good?
After
An instruction with an explicit task and constraints.
Review the proposed schedule against these requirements: 7 days, no more than 45 minutes per day, all three supplied topics covered, and one self-test each day. For each requirement return pass, fail, or unclear and quote evidence from the schedule. Do not judge scientific accuracy without sources.
Try it yourself
Use the scenario below, then compare your approach with a worked example.
Compare and reflect. No live AI generation.
Key takeaway
Define what success looks like, then check the response against it.