Iterating toward better output

What you’ll be able to do

Turn “this isn’t quite right” into a follow-up Claude can act on
Refine a draft in deliberate rounds and keep track of what each round changed
Know when to stop prompting and finish the job yourself

What you need to know

Iteration is normal, not failure

Even a well-written prompt rarely produces the final version on the first try. The first response is a draft that shows you how Claude understood the request. Iteration is the follow-up conversation that moves that draft toward what you need: tightening, reshaping, adding a missing point, adjusting tone.

This lesson is about improving output that is basically on track. If results are consistently off target, no matter how you follow up, the problem is likely upstream (missing context, the wrong source file, an unclear goal), and that's a diagnosis job covered in lesson 7.1.

Make feedback specific and actionable

"Make it better" forces Claude to guess what "better" means. Useful feedback names the part, the problem and the direction:

Point to the part — "the second paragraph," "the pricing row of the table," "the subject line."
Say what's wrong — "too formal for a team chat," "buries the deadline," "repeats the intro."
Say what you want instead — "lead with the deadline," "cut to 100 words," "use the phrasing customers use in the survey."
Say what to keep — "keep the three-bullet structure and the closing line exactly as they are." Without this, a rewrite can lose the parts you liked.

Showing beats describing. If you want a different tone, paste a sentence written the way you want it. If one item in a list is exactly right, say "make the others more like item 3."

Before

Not quite. Try again and make it more engaging.

After

Keep the structure. Open with the customer quote instead of the statistics, shorten each section to two sentences, and end with one clear call to action: register by Friday.

One meaningful change at a time

If you ask for a shorter, warmer, more technical version with a new structure all at once, and the result is worse, you don't know which change caused it. Making one or two related changes per round keeps cause and effect visible. It also makes it easy to go back: "the previous version's opening was better, restore that."

Regenerating the same prompt can show you a different take, which is fine for brainstorming. But hitting regenerate repeatedly on a draft that misses the point is gambling, not iterating. If the same miss keeps appearing, change what you're asking.

Refine in the chat, fix the prompt, or edit it yourself

Follow up in the chat when the draft is close and you're adjusting details. Claude can see the earlier turns.
Edit the original prompt and resend when you realize the first request was missing something fundamental, such as the audience. Correcting the starting point is often cleaner than stacking corrections on a draft built on the wrong premise.
Edit it yourself when what's left is a few words or a fact only you know. Three more rounds to change one phrase costs more than typing it.

Artifacts are useful when you'll iterate on a longer piece, such as a policy draft, a one-page brief or a table, because the current version stays visible as a standalone document alongside the chat while you request changes.

Capture what worked

When a refinement fixes something you'll need again ("always put the deadline in the first line," "use a table, not paragraphs, for comparisons"), fold it back into your reusable prompt, template or Project instructions. Otherwise you'll repeat the same three follow-ups every week. Iteration on one draft improves that draft; capturing the lesson improves every future draft.

Know when to stop

Decide in advance what "good enough" means: accurate, fits the audience, right length, ready to send after a quick human edit. Once the draft meets that bar, stop. Polishing past it rarely adds value, and the final check for accuracy and appropriateness is yours, however many rounds it took.

Key concept

Iterate with specific, targeted follow-ups that name what to keep and what to change, one meaningful change at a time, and carry lasting fixes back into the reusable prompt.

Exam scenarios show a draft that's close but not right and ask for the best next step. Strong answers give precise, actionable feedback; weak options regenerate blindly, pile on vague adjectives, or abandon a workable draft and start from scratch.

Practice scenario

ScenarioA communications coordinator asks Claude for a staff newsletter item about a new wellbeing program. The draft has accurate details and a good structure, but it's twice as long as the space allows and sounds like a press release. She is about to reply, "This is wrong, write it again."

Work it through, then open this

The draft isn’t wrong: the facts and structure are good, and those are worth keeping. “Write it again” invites Claude to change everything, including the parts that work. A better follow-up: “Keep the structure and all the details. Cut it to about 150 words and make the tone warmer and more conversational, as if a colleague were telling people about it. Here’s a past item with the tone we use: [paste].” If one sentence still feels off afterwards, she can fix it herself. And if every newsletter needs the same length and tone guidance, she should put that in her template or Project instructions.

Hands-on exercise — Run three deliberate rounds

Beginner · 15 min

What you’ll learn

How specific feedback changes the result compared with vague feedback
How to keep what works while changing what doesn’t
When to stop and edit by hand

Ask Claude for a short piece you actually need, such as a project status update. When it arrives, reply only “make it better” and note what changes.
- Why: It shows how vague feedback produces unpredictable changes.
- You should see: A different version, but not necessarily closer to what you wanted, and possibly losing parts you liked.
Go back to the first draft. Write one follow-up that names a part, the problem, the fix and what to keep (“Keep the bullets; move the risk to the top; cut the intro”).
- Why: It practices actionable, bounded feedback.
- You should see: A version that changes exactly what you asked and leaves the rest intact.
Do one more targeted round, then stop and finish any remaining wording yourself. Write down one instruction you’d add to your reusable prompt next time.
- Why: It builds the habits of stopping at good enough and capturing lasting fixes.
- You should see: A ready-to-send update and one concrete improvement for your template.

Exam traps

Giving vague feedback like “make it better”

Claude has to guess what “better” means. Name the part, the problem, the change and what to keep.

Regenerating the same prompt and hoping for a better answer

Repeated regeneration gives variations, not improvements. If the same miss recurs, change the request.

Changing several things at once so you can’t tell what helped

Bundling many changes hides cause and effect. One or two related changes per round keep it traceable.

Forgetting to say what was good, so it gets lost in the rewrite

Without “keep this,” a revision may discard the parts that already worked.

Iterating endlessly on a small wording issue faster to fix by hand

When only a phrase or a fact is left, editing it yourself is quicker and more reliable.

Not carrying a winning refinement back into the reusable prompt

If a fix will be needed again, put it in the template or Project instructions so the next draft starts better.

Sources

Anthropic, Claude Certified Associate – Foundations Exam Guide v1.0 (July 2026)
Claude Help Center
Claude Docs (includes prompt engineering guidance)