Creative credits exist to tell an audience who did what - and who answers when something goes wrong. AI complicates the “who did what” line without removing the “who answers” line. This article is the maker-facing disclosure playbook: when to name AI assistance, how to phrase it, and what not to pretend.
It extends keep your name on AI-assisted work into portfolio, exhibition, client delivery, and public posts. Workplace policy disclosure is covered separately in workplace AI disclosure when required. Stage-level taste decisions remain in keep your taste.
What the authorities already agree on
Three anchors matter for everyday creative practice:
- Human authorship for copyright claims. U.S. registration guidance requires disclosing AI-generated material along with a brief explanation of the human author’s contribution, excluding more than de minimis AI-generated material from the claim, and claiming authorship only for human-authored portions (Copyright Registration Guidance, 2023). The January 2025 copyrightability report reiterates that prompting alone is generally not enough for copyright in the output (Copyrightability Report).
- Human accountability in publishing ethics. COPE’s position on authorship and AI tools rejects listing AI as an author because a tool cannot take responsibility.
- Transparency for certain synthetic content in the EU. Article 50 of the AI Act sets transparency obligations for specified AI interactions and content, applying from 2 August 2026 (Regulation (EU) 2024/1689; Commission Article 50 FAQ). Deployer labelling duties for in-scope deepfakes and certain public-interest text remain on that date; provider machine-readable marking under Article 50(2) for generative systems already on the market before 2 August 2026 has a limited grace period to 2 December 2026 under Article 111(4), inserted into the AI Act by Regulation (EU) 2026/1744. Check whether your release format and role (provider vs deployer; professional vs purely personal use) are in scope rather than assuming a caption emoji is enough.
None of these say “never use AI.” They say: do not mislead about authorship, and do not dodge accountability.
When disclosure is usually needed
Disclose when a reasonable audience would be misled about how the work was made if you stayed silent:
- Substantial AI-generated prose, image, audio, or video ships under your name
- A client deliverable was largely drafted or visualized by a model and lightly edited
- Synthetic likeness or voice appears in the piece (also triggers consent rules - voice or likeness consent)
- A competition, gallery, publisher, platform, or grant requires disclosure by rule
Usually fine to skip a special credit for:
- Spellcheck, basic grammar, file conversion, non-generative filters
- Private sketches you never present as finished public work
- Purely internal notes that never reach an audience
When unsure, prefer a short credit. Silence is harder to unwind than a one-line note.
A generic footer that says “AI was used” while you present machine-invented facts, testimonials, or expertise as your lived experience is still misleading. Specificity matters: say what the tool did.
How to write the credit line
Keep it short, specific, and human-first.
Good patterns:
- “Essay by [Name]. Structure critique assisted by a text model; all anecdotes and final wording by the author.”
- “Cover illustration: human composition and characters; background textures generated with [tool], edited by hand.”
- “Trailer edit by [Name]. One establishing shot is AI-generated and labeled on-screen at first appearance.”
Weak patterns:
- “Made with AI” as the only author line (hides the human, or hides the split of labor)
- “AI-inspired” when the pixels or paragraphs were generated
- Dumping a full prompt log as if that were transparency (noise, not clarity)
For commercial claims inside the work, remember consumer-protection rules still apply to deceptive practices (FTC announces crackdown on deceptive AI claims and schemes).
Workflow: decide before you publish
- List AI touches on the piece (brainstorm / draft / image / voice / edit).
- Mark each as trivial vs. appreciable.
- Check venue rules (client contract, gallery, platform, grant).
- Draft the credit line before export day.
- Store the line with the project files so reprints stay consistent.
Disclosure credits should not become a dump of unpublished process details, client names, or third-party personal data. Say what kind of help you used - not every confidential ingredient that went into a prompt.
Edge cases
- Collaborations: name humans first; note shared AI tools once.
- Open-source models: still disclose appreciable generation; model license compliance is separate from audience honesty.
- Style-transfer fights: if the ethical issue is imitation, disclosure does not cure it - fix the reference practice (visual references without style theft).
Portfolio and client delivery norms
Portfolio: if a case study implies you hand-drew or hand-wrote deliverables that were substantially generated, the case study is marketing deception, not a gray area. Say what you directed, what you generated, and what you revised.
Clients: put disclosure expectations in the statement of work when AI will be material. Some clients forbid generative tools; some require them to be named; some only care about confidentiality. Guessing is how relationships break. If the client forbids generative AI, that overrides your personal comfort with silent use.
Platforms and festivals: read the rule sheet. Many festivals now ask a yes/no AI question. Answer the question they asked, not the one that flatters your process.
Illustrative scenario: a motion designer submits a festival cut with an AI sky replacement and checks “no AI” because “it was just the background.” That is the wrong bright line. Background generation is still generation. Disclose, or replace the shot with licensed/practical footage.
Content credentials and labels
Technical provenance signals such as Content Credentials can help downstream systems recognize synthetic or edited media when they are present and preserved. They are not a substitute for a human-readable credit when your audience is people. If you use watermarking or credentials features, treat them as complementary to the credit line - and remember signals can be stripped. For a primer on what those signals can and cannot prove, see Content Credentials and watermark basics. Write the human credit as if the badge might be missing.
Add the credit line before you publish
Take one piece scheduled to ship this month. Fill the decision card in creative AI disclosure card. Write the credit before you export. If you cannot explain the split of labor in one sentence, you are not ready to publish under your name.
