What you will be able to do
- Explain the Description-Discernment loop and why a first draft is seldom the final output
- Check a Claude draft for audience fit: accuracy, level, purpose and how readers will receive it
- Rewrite a vague critique such as 'make it better' as a specific revision instruction Claude can act on
Key concept
The Description-Discernment loop — Refining an output is a cycle. You tell Claude what you want, judge what comes back against your audience's needs, and then give a more precise instruction. Each round of judging makes the next description better.
1.Refining is a loop, not a single prompt
It is tempting to treat Claude's first answer as a finished product that you either accept or throw away. The AI Fluency material describes a different habit. Adapting an output for an audience is iterative: you describe the task, read what comes back, decide what is useful and what is not, and then describe again with more precision. The guidance says so directly: the first prompt rarely gets it right, so you refine using what each response teaches you.
This changes what counts as good editing. Accepting or rejecting a draft is not enough. The educator course describes the loop as a refinement tool in which you explain why a suggestion does or doesn't work for your particular readers. That explanation is what carries into the next draft. A reader-specific reason like "these readers are below grade level and need terms defined in context" gives Claude something to act on. A bare rejection gives it nothing.
The loop also starts before any output exists. The more context you give up front, the less correcting you do later: who the readers are, what they already know, what they will do with the text. The same guidance suggests uploading past materials such as rubrics or strong examples, so that Claude can match your voice and rigour from the first draft.
2.Judging fit: what to check before you edit
Discernment means reading the draft as the intended reader would, not as its author. The K-12 course gives a four-question checklist for this step. It is written for teachers, but each question applies to any audience: swap 'students' for customers, board members or new hires.
| Check | Question to ask of the draft | What a miss looks like |
|---|---|---|
| Accuracy | Is the content accurate for your subject area? | A confident claim you would have to correct before sending |
| Level | Did it land at the right level for your learners, with key vocabulary or concepts introduced and supported? | Jargon left undefined for novices, or basics over-explained for experts |
| Purpose | Does it support what students will do with it and where the lesson needs to go? | Well written, but not usable for the reader's next step |
| Reception | Would anything land badly with your specific students? | A tone, example or assumption that alienates this particular group |
The builder course adds two lenses that work well for written outputs too. The first is clarity: whether the output uses language and patterns the actual reader would recognise. The second is information hierarchy: whether the most important information comes first. A draft can be accurate and still fail both. A policy summary that opens with background and puts the action the reader must take in the fourth paragraph has a hierarchy problem, not a factual one.
The builder course also warns that the default AI output is often technically complete but misses the point. Complete and correct is the minimum. Fit for this reader is the standard you are editing towards.
3.Turning a critique into an instruction Claude can act on
Once you know what is wrong, you have to say it in a form Claude can execute. The builder course separates two things: a critique you would give a colleague, and a description an AI can act on. "Make it look good" is a wish, not a spec. The same applies to "make it friendlier", "make it more professional" or "make it better". Each names a feeling but gives no change to make.
The K-12 exercise shows the difference. Instead of "make it better", the teacher writes a revision that names concrete edits: shorten the steps in the directions and add a picture cue for each one. A good revision instruction names the reader, the specific change, and ideally the reason.
Claude's prompting guidance says why the reason matters. An explained instruction beats a bare rule, because Claude can generalise from the explanation. Compare a bare rule, 'NEVER use ellipses', with the version the guidance recommends:
Your response will be read aloud by a text-to-speech engine, so never use ellipses since the text-to-speech engine will not know how to pronounce them.The same guidance names two more levers for register. First, examples are among the most reliable ways to steer format, tone and structure, so pasting a short sample in the target voice often works better than describing the voice in adjectives. Second, setting a role focuses Claude's behaviour and tone; even one sentence makes a difference.
Something like: 'Rewrite this for patients with no medical background who will read it in a waiting room. Replace clinical terms with everyday words, or define them in a short phrase the first time they appear. Put what the patient needs to do first, and keep sentences short.' It names the reader, the specific changes and the reason, so Claude does not have to guess what 'simpler' means.
A product manager asks Claude to explain a new caching feature to both internal engineers and non-technical customers, but the first draft uses the same technical wording for each. What is the most effective way to adapt this output for both audiences?
Correct answer: A — Request two separate versions, matching vocabulary and detail to each reader's background
- A. Correct. Because the two audiences differ in background and needs, producing separate versions tailored to each produces content that is actually usable by both readers.
- B. Incorrect. Leaving one technical draft unchanged forces non-technical customers to filter out jargon themselves, defeating the purpose of adapting content for an audience.
- C. Incorrect. Trimming length without changing vocabulary or explanation depth does not address the mismatch between a technical explanation and a non-technical reader.
- D. Incorrect. A glossary still requires customers to translate jargon themselves rather than receiving an explanation already written in language they understand.
Exam traps
Each one states something that sounds right. Open it to see what is actually true.
1.Telling Claude to 'make it more approachable' or 'make it better' is enough to adapt the tone for a new audience.Why is that wrong?
A tone adjective is a wish. An effective revision names the audience, the concrete changes and why, and ideally includes an example of the target register.
Covered in Turning a critique into an instruction Claude can act on
2.If the first draft misses the audience, the best move is to discard it and regenerate from the same prompt.Why is that wrong?
Refinement is iterative. You use what the miss revealed to write a sharper description, instead of rerolling the same request and hoping for a better result.
Covered in Refining is a loop, not a single prompt
Sources
Every claim above is drawn from one of these pages, quoted as it was written on the date shown.
- 1.https://academy.claude.com/courses/ai-fluency-for-educators/applying-ai-fluency-to-learning-materials-and-assignmentsOfficial docs
“not just accepting or rejecting AI suggestions but explaining why they work or don't for your specific students”
↩︎ Refining is a loop, not a single prompt - 2.
“Upload past materials. Your unit plans, rubrics, or strong examples help AI match your voice and rigor.”
↩︎ Refining is a loop, not a single prompt“Did it land at the right level for your learners, with key vocabulary or concepts introduced and supported?”
↩︎ Judging fit: what to check before you edit“shorten the steps in the center directions and add a picture cue for each one”
↩︎ Turning a critique into an instruction Claude can act on“you describe what you want, the model responds, you discern what's useful, and you describe again, sharper this time.”
↩︎ Key concept“The loop is iterative. Your first prompt rarely nails it; refine using what each response teaches you.”
↩︎ Exam trap 2 - 3.https://academy.claude.com/courses/ai-fluency-for-builders/discernment-for-user-experienceOfficial docs
“Is the most important information the first thing users see?”
↩︎ Judging fit: what to check before you edit“A good critique and an actionable AI description are different artifacts. Learn to translate between them.”
↩︎ Turning a critique into an instruction Claude can act on“is a wish, not a spec. Describe experience with the same precision as a function.”
↩︎ Exam trap 1 - 4.
“The default AI output is technically complete but often misses the point.”
↩︎ Judging fit: what to check before you edit - 5.https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practicesOfficial docs
“Examples are one of the most reliable ways to steer Claude's output format, tone, and structure.”
↩︎ Turning a critique into an instruction Claude can act on“Setting a role in the system prompt focuses Claude's behavior and tone for your use case.”
↩︎ Turning a critique into an instruction Claude can act on