TutorialPart 3 of 3, A first week of deliberate AI use
Check an answer before you use it
Three quick checks (source, arithmetic and scope) that catch the most common problems in an AI answer before anything depends on it.
SampleWritten to demonstrate the format. It does not describe the owner’s experience, results or recommendations.
A fluent answer is not the same as a correct one. Before an answer goes into an email, a document or a decision, spend two minutes on three checks. They will not catch everything, but they catch the problems that cost the most.
Check 1: every claim has a source you can see
Highlight each factual claim. For each one, find where it came from: the material you pasted, a source the tool cited, or nowhere.
- From your material: find the sentence. Confirm it says the same thing.
- From a cited source: open the source. Confirm the source exists and supports the claim.
- From nowhere: treat it as unverified. Remove it or check it independently.
No highlighted claim is left without one of the three labels above.
Check 2: redo the arithmetic
Models can state totals, percentages and dates that look plausible and are wrong. Recalculate every number that matters, outside the chat.
# Recalculate, do not trust the stated total.
items = [1250, 980, 432]
print(sum(items)) # compare with the total in the answerCheck 3: the answer stays in scope
Compare the answer with the "done when" list from your brief. Remove anything the brief did not ask for, especially advice, recommendations or confident conclusions that go beyond the material.
Claims are traced, numbers are recalculated, and the answer covers what the brief asked for and nothing it did not. Note the outcome in your log.
When to skip these checks
For throwaway work, such as brainstorming names or rephrasing your own sentence, the checks cost more than they save. Use them whenever someone else will rely on the result.