What You Need to Know
A first output, Claude's first answer to a prompt, is a draft. The prompt that set the task is the brief. Feedback is the follow-up message that you send about the output. Iteration is the cycle of output, specific feedback and better output. Iteration is the normal way of working, not a failure. The Help Center's getting-started advice says this directly: do not hesitate to refine your prompts or ask follow-up questions. People who get the most from Claude plan one or two rounds of feedback for every task. People who get the least accept a mediocre first attempt, or abandon the tool after one attempt.
This lesson uses one made-up example throughout: a dental practice. Its manager asks Claude for a letter that reminds patients that their check-up is due. The draft has the right parts but is too formal, too long and full of clinical terms.
Feedback That Names the Gap
The Help Center names feedback as an official remedy for an unhelpful answer: provide follow-up instructions, clarify, or ask for a rewrite. "Make it better and try again" reproduces the problem, because the message adds no new information to the conversation. Suppose the manager writes "Cut it to 200 words, use a warm tone, and replace the clinical terms with everyday words." That message fixes the draft, because each clause names a gap.
Good feedback does one of three things:
- It adds a missing constraint, such as a word limit, the format or the thing to avoid.
- It corrects a misreading.
- It narrows the scope.
All three depend on diagnosis: find the gap between the output and the need, then state the gap plainly. A quick method is to check the output against the five components of the original brief. The gap is usually where a component was weak or missing. The manager's prompt named patients as the readers but gave no output specification: no tone and no length.
Asking for variants is iteration too. The manager can request three versions of the opening paragraph: formal, conversational and urgent. Claude generates the options, and the manager judges them. This method is often the fastest way to find the right register, or style, when you know something is wrong but cannot yet name it.
Refine or Re-Brief?
After a weak draft you have two choices. To refine is to stay in the same conversation and send a targeted follow-up. To re-brief is to rewrite the original prompt and start a new conversation with it. Ask two questions, in this order.
1. Is the draft close? A draft is close when it does the right task, is not written for the wrong reader, and has a few gaps you can name. A generic draft that suits no reader in particular can still be close: the missing reader is then one of the gaps. Refine. The check-up letter is close: it is written for patients and has the right parts. The follow-up above names its three gaps, and one round fixes it.
2. If the draft is not close, did the brief say the right thing? A draft that is not close does the wrong task, or speaks to the wrong reader, from start to finish. Read your brief again before you choose.
- The brief was right, and Claude misread it. Refine. Everything Claude needs is already in the conversation, so a follow-up that corrects the misreading is enough. Anthropic's Claude 101 course gives this advice: if Claude goes in a different direction than you intended, steer it back. Suppose the manager's brief named patients as the readers, but the draft speaks to the practice staff. She replies: "The readers are our patients, not our staff. Redraft it for them."
- The brief was wrong, or it left out something that the whole draft depends on. Re-brief. This covers the wrong task, the wrong framing, or a reader that the brief never named and Claude had to guess. Suppose the manager asks in one line for "a notice about our new opening hours". Claude writes a rota notice for staff, but she wanted a notice for patients. The fault is in the brief, so the fix belongs in the brief. She rewrites the brief to name patients as the readers and to say what they need to know, then starts a new conversation. The rewritten brief also gives every later notice a better start.
Also re-brief when a long series of corrections has made the conversation contradict itself, even if the brief was sound. The conversation now holds instructions that conflict, and each new correction adds to them. Put the corrections that worked into one new brief and start a new conversation. The Claude 101 course says that a new chat with a clearer prompt is sometimes faster than redirecting a conversation that has gone wrong.
You can choose wrongly in both directions. You can correct, turn after turn, a draft that came from a wrong brief. Or you can abandon a conversation that one specific follow-up would have finished.
Keep the Wording That Worked
Suppose that two rounds of feedback finally produce the right output. Then the feedback that fixed the draft is exactly the instruction that the original prompt was missing. Add that feedback to the original prompt. The reusable prompt then starts the next run from the improved version, rather than from the beginning. The manager's next batch starts from a prompt that sets the 200-word limit, the warm tone and everyday words instead of clinical terms.
In the exam
The exam tests the quality of feedback. It rewards the targeted follow-up over the vague retry every time. It expects you to recognise when the re-brief is the professional choice. Questions test the wrong choice in both directions. When a task recurs every week, moving its wording into standing configuration (instructions that stay in place for every run), such as the instructions of a Project, belongs to Domain 5. Domain 1 stops at refining the current request and keeping the wording that worked.
Why start a new conversation?
Longer chats consume more of your usage limit. When a chat approaches its length limit, the documented advice is to start a new conversation or use a Project. Restarting on a wrong brief is recommended as professional working practice, because a long series of corrections gives an unclear starting point. Anthropic's Academy also says that the quality of Claude's answers can get worse in a very long conversation. Its advice is then to start fresh.
Exam Traps
The vague retry that looks like diligence
Regenerating, rewording slightly, or asking Claude to try again adds no new information to the conversation, so the next output repeats the shortfall. Feedback only works when it names the gap: the audience, the tone, the length, the constraint that was missing.
Making small fixes to a fundamentally wrong brief through many follow-up turns
When the output shows the original prompt asked for the wrong thing, each correction works against the framing instead of fixing it. Rewriting the brief and starting a new conversation is faster and gives every later batch of the same work a better starting point.
Starting a fresh chat when the output needed only one specific fix
A restart discards the context and approved material that the conversation already holds. When the draft is close, or when Claude misread a correct brief, one targeted follow-up in the same conversation is cheaper and faster.
Practice Scenario
A retail merchandiser asks Claude for product descriptions for a seasonal homeware catalogue. The first batch is structurally fine but generic and in the wrong voice. The same prompt in a new chat came back just as generic, and she wants the copy fixed in one more round. What is the most effective next prompt?
Build Exercise
Run Two Feedback Cycles
What you'll learn
- How to diagnose the gap between an output and the need
- How to write feedback that names the gap plainly
- How to use variants in named registers to choose a direction
- How to add successful feedback to a reusable prompt
- Take a recent Claude output that disappointed you, or run the five-component brief from objective 1.1 and use its first draft
Why: Feedback practice needs a real gap to diagnose; invented examples make the exercise look easier than it is.
You should see: A first draft in front of you plus the prompt that produced it.
- Diagnose the gap: name what specifically differs from what you need, such as audience, tone, length, scope or a missing constraint
Why: Diagnosis is the tested skill; feedback can only name a gap you have identified first.
You should see: One or two written sentences naming specific gaps, not a general feeling of wrongness.
- Write feedback that names each gap plainly and run the first revision
Why: Specific feedback shows what to change in the revision; vague feedback reproduces the problem.
You should see: A revision that is measurably closer on the gaps you named.
- Diagnose again and run a second cycle, asking for variants in named registers if you are choosing between directions
Why: The second cycle builds the habit: output, specific feedback, better output, and variants make you the judge rather than the editor.
You should see: A second revision, or a set of labelled variants, close enough to finish by hand.
- Write down each piece of feedback you gave, mark which of it should have been in the original prompt, and add it to the prompt as your reusable version
Why: The feedback that fixed the draft is the instruction the original prompt was missing; adding it to the prompt makes the correction reusable.
You should see: A revised prompt containing the added feedback, saved wherever you keep working notes.
Sources
- Get started with Claude — Anthropic
- My prompt isn't giving me a helpful answer — Anthropic
- How do usage and length limits work? — Anthropic
- Claude 101: Your first conversation with Claude — Anthropic
- Claude 101: Getting better results — Anthropic
- AI Capabilities and Limitations: Working Memory — Anthropic