What it is
Iterative prompting is the practice of improving a result over several rounds: you send a prompt, evaluate what comes back, and respond with targeted feedback or a revised prompt. Rather than expecting a perfect answer on the first try, you steer toward it.
It applies both within a conversation ("make the second paragraph shorter") and across attempts (rewriting your original prompt based on what went wrong).
When to use it
- Creative or nuanced work where your taste matters (writing, naming, design ideas)
- Complex tasks where you can't specify everything up front
- The first answer is close but not right
- You're building a reusable prompt and want to make it reliable
How it works
Why it works
It's hard to know exactly what you want until you see a draft. Each round gives you something concrete to react to, and specific feedback is much easier to write than a perfect brief. The model can also use the earlier drafts in the conversation as context.
Give feedback the model can act on
The quality of each round depends on the precision of your feedback:
That's not quite right, try again.
Good start. Three changes: 1. Cut the intro paragraph; start with the offer. 2. The tone is too salesy. Write like a friendly neighbor, not an ad. 3. Keep the bullet list, but shorten each bullet to under 10 words. Leave everything else as it is.
"Try again" gives the model nothing to work with. Targeted feedback says what to keep, what to change, and how.
A typical loop
- Draft: send your best first prompt.
- Diagnose: what's wrong? Content, tone, length, format, accuracy?
- Direct: give specific, numbered changes, and say what to keep.
- Repeat until it's good enough, usually two to four rounds.
- Capture: fold what you learned back into the original prompt for next time.
Before revising, list what you think I didn't like about the last version. Then revise it.
That check helps confirm the model understood your feedback before rewriting.
When to start over
Long conversations can drift: earlier instructions get diluted, and the model may stick to an early wrong direction. If things are getting worse, open a fresh conversation with an improved prompt that includes everything you've learned.
Tip: Ask for options to speed up iteration: "Give me 3 very different versions," then pick one and refine it.
Note: AI output varies between runs, so a regenerate can produce a different result. Iteration with clear feedback is more dependable than hoping for a lucky draw.
Step by step
- 1
Send your best first-draft prompt.
- 2
Identify specifically what's wrong: content, tone, length, format, or accuracy.
- 3
Reply with numbered, concrete changes, and say what to keep.
- 4
Repeat for a few rounds, changing one or two things at a time when possible.
- 5
If the conversation drifts, start fresh with an improved prompt.
- 6
Save the final improved prompt so you get the good result first time next time.
Weak vs strong
No. Make it better.
This is close. Keep the structure and the opening line. Changes: (1) the second paragraph is too technical for parents, so replace the jargon with everyday words; (2) add one concrete example of a homework routine; (3) cut the total length by about a third.
Practical use cases
Polishing writing
Take a cover letter or blog post from rough draft to final through a few rounds of focused edits.
Naming and brainstorming
Generate options, pick favorites, and ask for more variations in the direction you like.
Fixing code
Run the suggested code, paste back the exact error or wrong output, and ask for a targeted fix.
Building reusable prompts
Test a prompt on several inputs, note failures, and refine the wording until it works consistently.
Practice exercise
Three-round polish. Ask an AI for a short bio for a professional profile (use real or made-up details).
- Round 1: send the request.
- Round 2: give three numbered, specific changes, and say what to keep.
- Round 3: fix one remaining issue.
Finally, write a single new prompt that would have produced the final version in one go.