Chapter 2 · Lesson 4

Fix This Prompt

Practice turning weak prompts into usable instructions.

5 min · Reviewed July 2026

Scenario

Your AI answer is not wrong. It is just not useful enough.

Do not restart. Repair.

When the first answer misses, keep the conversation going. Add missing context, point out what is wrong, and ask for a revision.

Name the failure

Say "too long", "too formal", "missing examples", "not actionable", or "wrong audience". Clear feedback is a prompt too.

Build the mental model

Useful AI work is usually iterative. The first answer gives you something concrete to inspect; your feedback turns that inspection into a better second instruction.

Diagnose before you rewrite

Name the observable problem: unsupported claims, wrong audience, weak structure, missing evidence, too much detail, or no clear next action. “Make it better” gives AI no test for improvement.

Protect what already works

A revision prompt should say both what to change and what to keep. This prevents a useful fact, tone, example, or constraint from disappearing while another part is repaired.

Worked example

AI drafted a useful proposal, but it is too long and the recommendation is buried.

Weak approach

Try again and make this better.

Improved approach

Keep the three cost estimates and the implementation risks. Put the recommendation in the first paragraph, reduce the draft to 250 words, and replace general benefits with two concrete outcomes for a 12-person studio. Do not add new facts.

Why it works

The follow-up identifies the failure, preserves valuable material, defines the target length and audience, and blocks invented evidence. The existing draft becomes working material rather than wasted effort.

Why iteration is more than repeated prompting

Iteration works when each round adds information. You review the answer against visible criteria, identify the largest gap, and give targeted feedback. Repeating the same request merely samples another answer from the same unclear brief.

For important work, finish with a separate verification pass. Ask which claims rely on assumptions, compare the output with the source material, and make the final decision yourself. Better prose is not automatically better evidence.

Apply it to your work

Write feedback that can be acted on

Take a recent AI answer you did not like. State one thing to keep, one failure to fix, and one criterion the revision must meet.

Starter templateKeep […]. Change […] because […]. The revision must […]. Do not […].

prompt choice

The answer is too generic. What is the best follow-up?

Compare the prompts closely. Look for useful context, constraints, and a clear output.

Takeaway

The second prompt is where useful work often starts.