Rubric
Contents — domains, guide and mocks

Refining outputs for the audience

CCAO-F 2.511 min read · checked 21 September 2026

Task statementEdit, adapt, refine, and compare outputs for the intended audience

The refinement loop

  1. Define the readerwho, what they know, what they decide
  2. Draftwith audience, purpose and format stated
  3. Compare to criteriafit for this reader, still accurate?
  4. Specific feedbackwhat to change, and why
  5. Final human edityour judgment, your name

Repeat compare → feedback until the draft meets the criteria

The audience is defined before the first draft, and the final pass is a human one. Everything in between is iteration against criteria, not taste.

Start with the reader, not the draft

Most “this isn’t quite right” reactions come from a draft written for no one in particular. Anthropic’s prompting guidance compares Claude to a brilliant new employee who lacks context on your norms: it cannot guess that the reader is a sceptical CFO with two minutes, unless you say so. Before asking for a draft — or a rewrite — pin down a few things about the reader.

QuestionCFOFrontline teamCustomer
What do they already know?The numbers, not the operational detailThe process, not the financesNeither; only their own experience
What do they need to do?Approve or reject a budgetChange how they work on MondayUnderstand what happened to them
What must come first?The ask and the costWhat changes and whenAn apology and what it means for them
Format and lengthHalf a page, figures in a tableShort checklistShort email, no jargon

Refining: say what to change and why

“Make it better” and “make it more professional” give Claude almost nothing to work with. Anthropic’s guidance points to three habits that make revision requests work. Explain the reason behind an instruction, because Claude generalises from the explanation. Say what to do rather than what not to do — “write in flowing prose paragraphs” works better than “don’t use bullet points”. And show examples: the guidance calls examples one of the most reliable ways to steer format, tone and structure.

Vague versus specific feedback

Vaguetext

This is too long and too technical.
Make it better for leadership.

Specifictext

Rewrite this for our COO, who will
decide on Thursday whether to fund
phase two. She knows the business well
but not the system details.

- Open with the recommendation and cost
- Keep all figures exactly as they are
- Replace technical terms with what they
  mean for operations
- Cut the implementation detail to one
  sentence; offer it as an appendix
- Maximum 250 words, plain prose
The specific request names the reader, the decision, what to keep, what to cut, and the length — so the second draft can be judged against something.

Refinement also has a limit. After two or three rounds of feedback on the same draft, it is often faster to fix the rest yourself, or to start a fresh request with everything you have learned about the reader folded into it. And the last edit is always yours: you know the relationship, the politics and the history that Claude does not.

For recurring audiences, you can set this once instead of in every prompt. Claude.ai’s personalisation features work at different scopes, and knowing which is which saves repetition.

FeatureScopeGood for
StylesChosen per conversationSwitching tone and format: presets Normal, Concise, Formal and Explanatory, or a custom style
Custom style from samplesReusable across chatsMatching a house voice by uploading or pasting approved writing examples
Profile preferencesAll your chatsGeneral preferences such as terminology or how you like answers pitched
Project instructionsChats in one ProjectA standing audience and format for a workstream, such as board papers

Adapting without drifting

Every rewrite is a chance for the facts to move. Simplifying is where it happens most: “up to 30%” becomes “30%”, “in the pilot region” disappears, a hedge is dropped because it sounded weak, or a friendly closing adds a commitment. So an adapted version needs its own check against the original — the same accuracy and completeness habits from 2.1, aimed at what the rewrite might have changed.

After any rewrite for a new audience

  • Passes: Figures, dates and names unchanged
  • Fails: Scope and qualifiers kept“some customers” became “customers”
  • Passes: Caveats and risks still present
  • Fails: No new commitments or promises“all delayed claims will be prioritised”
  • Passes: Reading level and jargon suit the reader
  • Check: Length and format suit the channel210 words; target was 150
The first four items are about the facts surviving the rewrite; only the last two are about audience fit. Both halves have to pass.

Comparing drafts

Asking for two or three versions with deliberately different approaches — one leading with the numbers, one with the customer story — is a quick way to see options you would not have written yourself. The skill is in choosing between them. Anthropic’s evaluation guidance for builders stresses specific, measurable criteria over vague impressions, and tone and style is one of the dimensions it lists. The same discipline works for a single email: decide what “good for this reader” means before you read the drafts, then score each one against it.

Choosing between two drafts

Draft A — impressive

  • Elegant opening paragraph on market context
  • Recommendation appears in paragraph four
  • 480 words; the reader has two minutes
  • All figures correct

Draft B — fit for the reader

  • Recommendation and cost in the first two lines
  • Risks in one short list
  • 220 words, figures in a small table
  • All figures correct
Draft A is longer and more polished; Draft B meets the reader’s needs. Criteria written in advance stop the more impressive-looking draft from winning by default.

Traps the wrong answers are built from

Tempting but wrongDo this instead
Asking Claude to “make it better”Name the reader, the purpose, what to keep and cut, and the length.
Choosing the longest or most polished draftScore drafts against criteria written for this reader before you read them.
Trusting a friendlier rewrite without re-checking itCheck figures, scope, caveats and commitments against the original.
Repeating the same audience instructions in every chatUse a style, profile preferences or Project instructions at the right scope.
Endless rounds of feedback on one draftAfter a few rounds, edit it yourself or restart with a better brief.

You should now be able to

  • Describe an audience in terms Claude can act on: knowledge, decision, priority, format.
  • Write specific revision requests with reasons, positive instructions and examples.
  • Check an adapted output for drifted figures, lost qualifiers, dropped caveats and added promises.
  • Compare alternative drafts against criteria set in advance, and combine their strengths.
  • Choose between styles, profile preferences and Project instructions for recurring audiences.

Practice questions

Original questions written for this lesson, in the exam’s style. Answer first, then open the reasoning — every option is explained, including why the wrong ones are tempting.

  1. Question 1

    A project manager has a detailed technical status report and needs a version for the steering committee, who will decide next week whether to extend the project’s budget.

    Which revision request is most likely to produce a useful version?

    1. A“Rewrite this so it is more suitable for senior leadership.”
    2. B“Make this shorter and more professional, and don’t use jargon.”
    3. C“For a budget decision: open with the ask and cost, keep all figures, 300 words.”
    4. D“Rewrite this in the Formal style so it reads well for executives.”
    Show answer and reasoning
    1. AIncorrect. It names an audience but not what they need, what to keep, or the length — Claude has to guess.
    2. BIncorrect. Better, but still vague about purpose and what must stay; “don’t” instructions are weaker than saying what to do.
    3. CCorrect. It states the decision, what goes first, what must not change, and the length — criteria the draft can be judged against.
    4. DIncorrect. A formal tone may help, but a style preset does not tell Claude what the committee needs to decide.

  2. Question 2

    Claude rewrites internal incident notes into a customer-facing apology email. The email reads warmly and clearly.

    Which TWO checks matter most before sending it? (Select 2.)

    1. AConfirm the scope and figures match the incident notes exactly.
    2. BAsk Claude whether the email is appropriate for customers.
    3. CMake sure the email is longer, so it seems more sincere.
    4. DCheck it makes no commitments beyond what has been approved.
    5. ESwitch to the Explanatory style so customers understand the cause.
    Show answer and reasoning
    1. ACorrect. Simplification often widens or narrows scope and rounds figures; the facts must survive the rewrite.
    2. BIncorrect. A self-assessment does not compare the email with the notes or with what has been approved.
    3. CIncorrect. Length does not signal sincerity, and customers usually need the short version.
    4. DCorrect. Friendly rewrites often add promises, which become obligations once sent.
    5. EIncorrect. Customers need to know what it means for them; a more educational style is not a check.

  3. Question 3

    A sales lead has two Claude drafts of a client proposal summary. Draft A is longer and more elegantly written. Draft B is shorter and puts the pricing and next steps first. Both are accurate.

    What is the best way to choose between them?

    1. APick Draft A, because a longer draft covers more ground.
    2. BAsk Claude which draft is better and go with its answer.
    3. CPick the draft that sounds most like the sales lead’s own voice.
    4. DJudge both against what the client needs and pick the better fit.
    Show answer and reasoning
    1. AIncorrect. Length is not fitness for purpose; the client may need the key points quickly.
    2. BIncorrect. Claude can help score drafts against criteria, but without criteria its preference is just another opinion.
    3. CIncorrect. Voice matters, but it is one criterion; what the client needs to see comes first.
    4. DCorrect. Comparing against audience criteria set in advance is the discipline; polish and length are not the test.

  4. Question 4

    A marketing team wants every Claude draft to match its brand voice without writing out the rules in each prompt.

    Which approach best fits this need?

    1. ACreate a custom style from approved brand writing samples and use it when drafting.
    2. BSelect the Concise preset style for all marketing drafts.
    3. CAdd “use our brand voice” to the start of each prompt.
    4. DPaste the full brand guide into every new conversation.
    Show answer and reasoning
    1. ACorrect. Custom styles can be built from writing examples and reused, which suits a consistent house voice.
    2. BIncorrect. Concise changes length and directness, not the team’s particular voice.
    3. CIncorrect. Claude has no way to know what the brand voice is from that phrase alone.
    4. DIncorrect. It can work but is repetitive and easy to forget; a reusable style is the better fit.

Practise every question in Domain 2 →

Sources

Drafted with AI assistance and checked against the sources above; expert review is in progress. Spotted an error? Tell us and it gets fixed, dated and listed on how this is written.