What you will be able to do
- Set explicit, audience-based standards before comparing two candidate outputs
- Use audience-fit checks such as voice, reader priorities, level and hierarchy to judge drafts
- Compare a revision with an original, or two audience variants, using the same standards
- Take responsibility for the chosen output as the decision-maker
1.Set the criteria before you read either draft
Suppose Claude has given you two versions of something: two pitches, two announcements, two guides. Choosing whichever feels better gives you an answer you can't explain or defend. The AI Fluency course for creative work puts the requirement directly: AI cannot serve a practice you have not described, and you cannot evaluate output against standards you have not made explicit. The aim is to turn descriptions into explicit standards you can evaluate work against, yours or an AI's.
The course lays out a sequence. First, put into words what you make and how. Next, convert those descriptions into standards a peer could apply to work that is not yours. Then calibrate: test your standards against one of your older pieces to make sure they can actually tell adequate work from excellent. For a comparison aimed at an audience, the standards come from the reader: what they say they prioritise, what they already know, and what they need to do after reading.
Checkpoint 1 of 6· Put it in order
Put the steps for building evaluation standards from the creative-work course in order.
- 1.Calibrate by testing the standards against one of your older pieces
- 2.Describe your practice: what you make, how you make it, and how you work
- 3.Convert those descriptions into standards a peer could apply to work that is not yours
The course goes from description to explicit standards, then calibrates them on older work to check they can tell adequate from excellent.
“Finally, calibrate: test your standards against one of your older pieces to make sure they can actually tell adequate work from excellent.”Source: academy.claude.com
Checkpoint 2 of 6· Exam question
A support lead finds a Claude-drafted troubleshooting article accurate but written at a reading level too advanced for the average customer. Which follow-up instruction best refines the draft without sacrificing accuracy?
Correct answer: A — Rewrite the article with shorter sentences and everyday words, keeping every step accurate
- A. Correct. Simplifying sentence structure and vocabulary while preserving the accurate steps directly targets the readability problem without losing correctness.
- B. Incorrect. Removing steps changes the article's completeness and could omit cases some customers need, which does not fix a reading-level problem.
- C. Incorrect. Adding more technical detail moves the article further from, not closer to, the audience's reading level.
- D. Incorrect. A formal specification format is written for engineers and would make the article harder for typical customers to follow.
Sources1
2.Criteria that test fit for the reader
The AI Fluency courses give concrete checks you can turn into comparison criteria. The nonprofit writing exercise lists the checks for a high-stakes draft below. Checking accuracy also belongs on the list, but whether a draft fits its reader is a separate question from whether it is correct, and these checks target fit.
| Discernment check | What it catches |
|---|---|
| Check if the language reflects your organization's actual voice or sounds generic | A draft that could have come from anyone |
| Assess whether it addresses the funder's priorities or just describes your program | A draft written about you rather than for the reader |
| Identify any deficit-based framing or problematic language about beneficiaries | Framing that would land badly with the people involved |
| Note what's missing that only you know about your work | Gaps Claude could not fill on its own |
Other courses add criteria for other audiences. The K-12 checklist asks whether a draft landed at the right level for its learners, with key vocabulary or concepts introduced and supported, and whether anything would land badly with your specific students. The builders course asks whether the output uses language and patterns the intended user would recognise, and whether the most important information is the first thing users see. Pick the criteria that match your reader, then score every candidate against the same list.
Checkpoint 3 of 6· Check yourself
Two Claude drafts of a grant proposal are equally accurate. Draft A describes the program in detail. Draft B ties each program element to what the funder says it cares about. Using the nonprofit discernment checks, which is the stronger fit?
The checklist asks directly whether a draft addresses the reader's priorities or only describes your own work. Accuracy is necessary but doesn't decide fit.
“Assess whether it addresses the funder's priorities or just describes your program”Source: academy.claude.com
3.Compare versions side by side on the same standards
Several exercises end with a direct comparison. The nonprofit lesson's stretch goal is to iterate on your prompt with specific feedback about tone or framing issues you identified, then compare the two versions, so you can see how better description changed output quality. The K-12 exercise asks you to submit one concrete revision prompt, then compare the revision to the original. In both cases the comparison is useful because both versions are held to the same criteria. If you judge each draft by a different yardstick, the result tells you nothing.
A related case is two versions written for different audiences: an expert edition and a beginner edition, or a head-office edition and a regional one. Here the criteria divide into two kinds. The K-12 checklist lists content accuracy as its own check, separate from whether the material landed at the right level. Level and register should differ between the editions, because each is meant for a different reader. Accuracy applies to both, so an edition adapted for an audience may change how it explains something but not the facts it explains.
Checkpoint 4 of 6· Check yourself
A writer has two Claude drafts of the same guide, one for experienced readers and one for newcomers. Based on the K-12 discernment checklist, which pair of checks should both drafts pass?
The checklist treats accuracy and level as separate checks. Accuracy is the same for every version, and level is judged against each version's own readers.
“Is the content accurate for your subject area? Did it land at the right level for your learners”Source: academy.claude.com
Checkpoint 5 of 6· Exam question
A marketing analyst asks Claude for a sales summary for executives who want only key trends and decisions, not methodology, but the first draft includes several paragraphs on data cleaning. How should the analyst refine the prompt for this audience?
Correct answer: A — Ask Claude to lead with the trends and decisions and drop the data cleaning explanation
- A. Correct. Executives want trends and decisions, so restructuring the summary to foreground that content and dropping methodology detail matches the stated audience need.
- B. Incorrect. Moving the methodology to a footnote still includes content the executives do not want instead of leading with what they need first.
- C. Incorrect. Expanding the methodology section moves the summary further from what a time-constrained executive audience wants.
- D. Incorrect. Uniformly shortening every paragraph keeps the irrelevant methodology section in the summary instead of removing it.
4.You make the call and own the result
A comparison ends in a decision, and the decision is yours. The K-12 course says it directly: AI accelerates, but doesn't replace expertise. You're still the decision-maker. The nonprofit lesson sets the bar for whatever you choose: own the final result, because when you submit work created with AI assistance, you should be able to stand behind every word.
Choosing doesn't have to mean picking A or B as they stand. The writing exercise asks which parts of the AI draft are usable and which need complete rewriting, so you can keep the strongest material and revise the rest. The builders course gives a final caution: AI output often passes “Does it work?” but stumbles on “Is it the right thing?” A draft can be fluent and correct and still be the wrong choice for its reader. That gap between working and right is where your judgment counts most.
Checkpoint 6 of 6· Check yourself
Which statement about choosing between Claude-generated drafts matches the AI Fluency guidance?
The guidance makes the human submitter responsible for AI-assisted work, which means standing behind every word after applying discernment.
“Own the final result: When you submit work created with AI assistance, you should be able to stand behind every word”Source: academy.claude.com
Exam traps
Each one states something that sounds right. Open it to see what is actually true.
1.Read both drafts first, decide which feels stronger, and work out the reasons afterwards.Why is that wrong?
Write down explicit standards, based on the audience's stated priorities, before comparing. Without them the judgment can't be evaluated or repeated.
2.A version adapted for a less expert audience can loosen accuracy as long as the level is right.Why is that wrong?
Accuracy and level are separate checks. Level changes with the audience, but every version must still be accurate.
Covered in Compare versions side by side on the same standards
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-creative-work/description-and-discernmentOfficial docs
“AI cannot serve a practice you have not described, and you cannot evaluate output against standards you have not made explicit.”
↩︎ Set the criteria before you read either draft“convert those descriptions into standards a peer could apply to work that is not yours”
↩︎ Set the criteria before you read either draft“you cannot evaluate output against standards you have not made explicit”
↩︎ Exam trap 1“Finally, calibrate: test your standards against one of your older pieces to make sure they can actually tell adequate work from excellent.”
↩︎ Checkpoint - 2.
“Check if the language reflects your organization's actual voice or sounds generic”
↩︎ Criteria that test fit for the reader“Iterate on your prompt with specific feedback about tone or framing issues you identified, then compare the two versions”
↩︎ Compare versions side by side on the same standards“Which parts of the AI draft are usable versus which need complete rewriting?”
↩︎ You make the call and own the result“Assess whether it addresses the funder's priorities or just describes your program”
↩︎ Checkpoint“Own the final result: When you submit work created with AI assistance, you should be able to stand behind every word”
↩︎ Checkpoint - 3.
“Would anything land badly with your specific students?”
↩︎ Criteria that test fit for the reader“submit it, and compare the revision to the original.”
↩︎ Compare versions side by side on the same standards“AI accelerates, but doesn't replace expertise. You're still the decision-maker.”
↩︎ You make the call and own the result“Is the content accurate for your subject area? Did it land at the right level for your learners”
↩︎ Exam trap 2“Is the content accurate for your subject area? Did it land at the right level for your learners”
↩︎ Checkpoint - 4.https://academy.claude.com/courses/ai-fluency-for-builders/discernment-for-user-experienceOfficial docs
“Information Hierarchy — Is the most important information the first thing users see?”
↩︎ Criteria that test fit for the reader - 5.
“AI output often passes “Does it work?” but stumbles on “Is it the right thing?””
↩︎ You make the call and own the result