Makers already know the difference between a good workshop and a bad one. A good workshop names what is working, points to specific lines or moments, and leaves the artist in charge of the fix. A bad workshop either flatters or hijacks. Default AI chat behaves like a bad workshop unless you constrain it: it praises generically, then offers a full rewrite “for clarity.”
This article gives you structured feedback prompts - a portable workshop. Use it after you draft (draft then critique), and after you lock direction (iterate variants). It complements, rather than replaces, human peers. For the wider stage map of a creative practice, see keep your taste.
Scenario
Illustrative scenario: a songwriter pastes a verse and says “give feedback.” The model replies that the verse is “powerful and evocative” and then rewrites the chorus. Useless. The same songwriter pastes the verse with three criteria and “quote before you advise.” They leave with three actionable notes and still own the melody of the lines.
The rubric (steal this)
Score or discuss only what you name:
- Clarity of intent - Can a cold reader say what the piece is trying to do in one sentence?
- Structure - Where does attention sag, jump, or repeat?
- Specificity - Which images or claims are generic?
- Voice consistency - Where does diction drift into template language?
- Risk and ethics - Any consent, likeness, or misleading-credit issues?
You will not use all five every time. Pick three.
Core prompt pack
A. Cold-reader summary
Read the piece once. In one sentence, state what you think I am trying to do.
Then list three places where a cold reader might misunderstand me.
Do not rewrite. Quote the text when you point.
B. Structure map
Outline the piece as 5-8 beats in my existing order.
Mark any beat that could be cut without losing the thesis.
Mark any missing beat the thesis seems to promise.
No prose rewrite.
C. Specificity hunt
List the most generic nouns and verbs in the draft.
For each, ask me one question that would force a more specific choice.
Do not answer the questions for me.
D. Voice drift detector
Compare the draft to these three short samples of my prior work: [samples].
Where does the new draft sound unlike me? Quote lines.
Do not "fix" my voice by rewriting.
E. Risk pass
Flag any passage that may involve someone else's private story, likeness,
or a claim that could mislead an audience about how the work was made.
Ask clarifying questions. Do not invent legal conclusions.
Save the pack in a note you control. Paste one module at a time. Multi-module dumps encourage the model to skip quoting and jump to rewrites.
Validation and fallback
Validation: every usable note should point at a quote or a beat. If the feedback could apply to any piece in your genre, it is too vague - rerun with “quote or delete the comment.”
Fallback: if the model keeps rewriting, end the chat and use a human workshop, or print the draft and annotate by hand. Tool failure is not a reason to accept ghostwriting.
Workshop pastes are still disclosures. Strip client names, unpublished embargoes, and other people’s intimate details. For photos and faces in the piece, remember photo consent.
Authority and honesty limits
AI feedback is not peer review, and it is not a rights opinion. For authorship and disclosure norms when you later publish, rely on human-accountability standards such as COPE on authorship and AI tools and the U.S. Copyright Office’s registration guidance on AI-generated material. The same office’s copyrightability analysis still centers human authorship of what you claim (Copyrightability Report, Jan 2025). For commercial claims inside the art (testimonials, product promises), keep the FTC’s consumer-protection lens in mind (FTC announces crackdown on deceptive AI claims and schemes). For EU-facing releases, plan transparency for in-scope synthetic content under AI Act Article 50 (Regulation (EU) 2024/1689; Commission FAQ).
Shown module run (illustrative)
Piece under review: a 180-word artist statement.
Module A reply shape that helps: “I think you are trying to explain why you switched from client illustration to personal ceramics.” Then three misunderstanding risks, each with a quote - for example, a line about “healing” that could be read as wellness marketing.
Module A reply shape that fails: “This is a beautiful and heartfelt statement! Here is a tighter rewrite…” - delete and rerun.
Facilitation tips for groups
If you use AI inside a human workshop:
- Agree in advance whether any member’s draft may be pasted into a tool.
- Prefer one designated facilitator account with redaction rules.
- Treat model notes as optional suggestions on a whiteboard, not as the workshop’s verdict.
- Keep human time for the emotional and relational feedback AI is bad at.
This matches the broader agency rule in what not to delegate to AI: consequence and relationship dimensions often say “assist, do not replace.”
When workshop mode is the wrong tool
Do not use these prompts to:
- Generate the piece you were supposed to write (muse boundary)
- Decide ethical consent questions for you (voice/likeness consent)
- Produce legal advice about copyright disputes
Use them to see your draft more clearly. Judgment stays yours - and so does the edit pass in edit pass that preserves your taste.
Use the pack on one piece
Pick one unfinished work. Run modules A and C only. Accept up to five changes you can defend. Full pack: workshop feedback prompt pack. If voice replacement is the real fear, continue to refuse to ghostwrite your voice.
