Lesson 7. Iterations and prompt improvement
Most of prompt engineering isn't writing the first prompt — it's properly analyzing the response and building the next, stronger iteration.
Topic breakdown
A common mistake is looking at the model's first response and deciding that AI works well or poorly. In practice, the first output is often just a draft and serves as material for the next step.
Iteration isn't resending the same prompt. It's analyzing weak points in the result, giving precise feedback, and requesting improvements only on the specific issues.
This approach is especially useful in copywriting, research, strategic notes, presentations, and any content where quality grows through multiple passes.
What you'll learn
- find weak spots in a response
- give targeted feedback to the model
- improve quality over 2-3 iterations
- separate review and rewrite into different stages
Lesson plan
Why the first answer isn't enough
The first result often sets the direction, but may have problems with accuracy, audience fit, and format. Iteration fixes exactly those.
Review and rewrite as two stages
First analyze the response, then rewrite. If you mix the stages, the model often starts fixing the wrong things.
What feedback is useful
Useful feedback names a specific place and a specific problem: where it's too generic, where the CTA is weak, where the audience doesn't come through.
A strong iteration strategy
Fix one or two important problems per pass. This way improvement happens under control and the prompt doesn't become chaotic.
Weak vs strong prompt
Rewrite this better.
Evaluate the text. Problem 1: audience isn't clear. Problem 2: CTA is too soft. Problem 3: the benefit is stated too generically. Fix only these three points but preserve the short style and strong headline.
The strong prompt shows the model exactly what to improve and what not to lose in the new version. This makes the iteration controllable.
Ready prompt template
Copy and adaptEvaluate the following response against 4 criteria: clarity, practical usefulness, audience fit, CTA strength. Score each on a 10-point scale. Then identify 3 specific problems. After that, create a new version fixing only those problems while preserving the strengths of the original text.
Why it works
Without a separate review step, iteration becomes random: the model writes another generic response and doesn't necessarily fix the right things.
Clear criteria turn feedback into a measurable system: it's easier to understand what exactly needs improvement.
You don't need to change everything at once. Usually fixing 2-3 specific problems works better.
It's important not only to point out errors but also to preserve what was already good.
Practice
- Take one of your previous prompts.
- Evaluate the AI response against 4 criteria.
- Formulate 3 specific problems: too generic, weak CTA, wrong tone, etc.
- Build a rewrite prompt and compare the new result with the old one.
Mini-project
Mini-project: review-rewrite cycle
Improve one prompt through three steps: initial response, review by criteria, and rewritten version. Goal — get a noticeably stronger result for the same task.
Tasks
- Build the initial prompt and response.
- Evaluate the response against 4 criteria.
- Write 3 targeted notes.
- Make a new rewrite pass and compare the result.
Deliverables
- 1 initial prompt and response
- 1 review block
- 1 improved version and brief difference analysis
Checklist
Common mistakes
- writing simply 'improve this' without explanation
- changing the task goal on each iteration
- giving overly generic feedback
- not specifying which strong parts should be preserved
Lesson FAQ
Do I need to iterate every response?
No. For simple one-off tasks this isn't always needed. But for business results, templates, and important content, iterations give a noticeable quality gain.
How many iterations are usually enough?
Often 2-3 passes are sufficient. If there's no noticeable improvement anymore, it's better to rebuild the prompt from scratch rather than endlessly editing.