ICMD — AI Music Governance & Production Checklist (2026)

Purpose
Use this checklist to operationalize AI-generated music (e.g., Suno, Udio, Stable Audio) across marketing, product, media, or internal content—without losing provenance, brand consistency, or legal defensibility.

1) Define your usage tier (pick one per asset)
- Internal-only (training, prototypes, mood boards)
- Organic social (short-form posts, creator content)
- Paid media (ads, app install campaigns, TV/CTV)
- Monetized distribution (streaming releases, YouTube monetization, sync licensing)
Rule of thumb: the higher the tier, the stricter the provenance and review requirements.

2) Write a “Music PRD” (creative brief) before generating
- Audience and context: where will this be heard (TikTok, podcast intro, in-app, retail)?
- Target emotion: energetic, intimate, luxurious, comedic, tense, etc.
- Constraints: duration (6s/15s/30s), BPM range, instrumentation, vocal/no vocal.
- Brand exclusions: genres, instruments, lyrical topics, or vocal styles to avoid.
- Competitive references: 2–3 “vibe” references (avoid naming living artists if your policy bans it).

3) Generate with guardrails
- Do not request “sound like [living artist]” or a recognizable celebrity voice.
- Save prompts, outputs, timestamps, and tool/model version for every candidate.
- Generate in controlled batches: change one variable at a time (tempo, vocal style, arrangement density).

4) Score candidates consistently (simple rubric)
Rate each 1–5:
- Hook strength (first 3–6 seconds)
- Vocal intelligibility (if present)
- Brand fit (does it feel like you?)
- Mix clarity (especially under voiceover)
- Distinctiveness (does it feel generic?)
Keep the top 10–20% only.

5) Provenance package (required for external distribution)
Create a folder per shipped track containing:
- Final audio file(s)
- Prompt log (text file)
- Generation metadata (model/tool version, date/time)
- Usage tier + approver names
- Notes on edits (what was changed post-generation)

6) Legal/brand review triggers
Escalate for review if any are true:
- Vocals resemble a known artist archetype or celebrity
- Lyrics reference trademarks, brands, or identifiable people
- The track will be used in paid media above your spend threshold
- The track is a “signature” asset (theme, brand anthem, long-running campaign)

7) Decide: ship AI as-is vs. re-record
Ship AI as-is when:
- Low-risk tier (internal/organic) and generic bed is acceptable
- Short shelf-life creative (24–72 hours)
Re-record when:
- Signature campaign, long shelf-life, or wide distribution
- You need defensible ownership or a unique vocal identity
- You want stems and deep mix control

8) Post-launch monitoring
- Track performance deltas (CTR, watch-through, retention) across 2–4 audio variants.
- Maintain a “do-not-use” library of prompts/styles that create brand risk.
- Review policy quarterly as laws, platform rules, and vendor terms change.

Outcome
If you follow this checklist, AI music becomes a measurable production advantage—fast iteration with controlled risk—rather than an unmanaged source of legal uncertainty and brand inconsistency.