Budget the learning objective
Start with a teachable question, such as ‘How does changing arrangement language change a 20-second cue?’ That question gives the workshop a reason to generate only a few alternatives. A room that produces fifty files but never compares them has practised button pressing, not listening or authorship. UNESCO's guidance on generative AI in education calls for human-centred, pedagogically designed use; in music, that means the tool serves a defined listening task.
Write the outcome in human terms: each group will describe two audible differences, identify its own compositional or editorial choices, and present one version with a credit note. Avoid framing machine output as proof of talent or as a replacement for instrumental instruction. This is an arts workshop, not a claim about therapy, diagnosis or student ability.
Set a small, visible resource envelope
Use a hypothetical envelope rather than implied current prices: four groups, six generation attempts per group, one shared account operated by an adult facilitator, and a fixed 90-minute session. Reserve more time for briefing, listening and critique than for generating. The facilitator should confirm the service's current age, account and content conditions before the session; a policy page can change, so the lesson plan should not rely on remembered rules.
Give each group a paper or shared log: prompt or brief, source materials, attempt number, selection reason, human edits and unanswered rights questions. The log makes authorship discussable without pretending it settles legal ownership. It also prevents a later showcase from presenting an untraceable output as wholly made by a named student.
Make critique the main event
Run a blind listening round with the group names hidden. Ask listeners to notice form, timbre, intelligibility, pacing and fit to the stated brief. Then reveal the process notes and ask what the humans decided: the reference constraint, the rejected direction, the edited structure or the performance added afterward. A model can be one source of material; the critique should make the human judgment legible.
Keep an authorship line for every final piece: ‘human brief and selection by…; generated sketch from…; human arrangement or performance by…’ Use only claims the group can substantiate. If a final work uses a third-party recording or a recognisable voice, exclude it unless the relevant permission is clearly in scope.
Close with a revision, not a leaderboard
Ask every group to revise one choice after feedback and explain why. The final reflection can compare the first and revised brief, rather than rank who made the ‘best’ song. This turns the workshop into evidence of listening and revision, the parts learners can carry to other tools and future sessions.
Afterward, delete or retain drafts according to the host organisation's policy, and preserve only the minimal project record needed for the agreed showcase. The workshop budget is successful when it protects time for that reflection and keeps a small group from spending all its credits before anyone has listened.
Continue in Listening & research, or The Copyright Office found prompts do not control output.
Evidence & further reading
Go to the source.
Primary references checked 19 September 2026. Practical exercises are editorial guidance, not reported product tests.
- Guidance for generative AI in education and research ↗
Primary international guidance supports human-centred, pedagogically designed use and attention to privacy.
UNESCO · Source date 2023-09-07 · Accessed 2026-09-19 - Terms of Service ↗
Current terms illustrate why a facilitator must check live platform conditions before a session.
Suno · Accessed 2026-09-19