Why disclosure is worth doing
Audiences make assumptions about how work is made. They assume a photograph shows something that happened, that a voice belongs to the person speaking and that an article reflects a writer’s knowledge. When generated material changes any of those assumptions, telling people respects their ability to judge what they are seeing.
AI disclosure is not about confessing a wrongdoing. Using tools is normal. The point is to avoid giving a false impression, and to let your audience decide how much weight to give the work.
When disclosure matters most
There is no single rule that fits every case, and requirements vary by platform, industry and location. As a general principle, the more the generated element affects what the audience believes or feels, the more a note is warranted. Disclosure tends to matter most when:
- The material looks real. Realistic images, voices or video could pass as documentation. See the broader concerns around synthetic media.
- A real person is depicted or imitated. A simulated voice or face should rarely appear without clear consent and a clear label.
- The content makes factual claims. Readers rely on a human or an accountable source standing behind information.
- Trust is part of the product. Reviews, testimonials, expert advice and journalism lean on authenticity.
- A platform, client, school or publisher asks for it. Check their policies and honor them.
It matters less when a tool helped with minor, routine tasks. Many people do not feel the need to announce spell-check or basic color correction. Where you draw the line is a judgment call, and a reasonable test is whether an ordinary viewer would feel misled on learning about it.
What to disclose
Be specific enough to be useful but short enough to be read. Useful details include:
- What was generated. An image, a voiceover, a script draft, a translation or a background.
- How much of the work it represents. A small element or the whole piece.
- The role of a human. For example, that a person reviewed and edited the result. State this only if it is true.
Avoid wording that overclaims or underclaims. Calling a lightly edited generated draft “written by hand” is misleading, and so is hiding a heavy human edit behind the phrase “AI-made.”
Sample wording
Plain sentences work best. Hypothetical examples:
- “The illustration in this post was generated with software and then adjusted by the author.”
- “A draft of this article was produced with an AI writing tool, then reviewed, fact-checked and edited by a person.”
- “The narrator’s voice is synthetic. No real person recorded this audio.”
- “The background images in this video are generated and do not show real places.”
Notice that they say what happened without apologizing or inflating. Avoid jargon the audience may not know.
Where to put the note
A label no one sees does little. Think about placement:
- Near the content. A caption beside an image or a line at the start of an article is better than a footer.
- Inside video or audio. A spoken line or on-screen text at the beginning is clearer than a note buried in a description. Video captions and on-screen text can carry a short label.
- Consistent across your work. A standard phrase and location help regular readers recognize it.
- In accessible form. If a label appears inside an image, repeat it in alternative text or the surrounding text so that people using assistive tools get it too.
Where a platform offers its own labeling feature, using it is sensible, but still consider whether the platform’s label is visible enough for your audience.
Common mistakes
Disclosing only when asked. A note that appears after an audience member challenges you feels defensive. Early, plain disclosure reads as confidence.
Using disclosure as a shield. A label does not excuse harmful or deceptive content. A fake of a real person is still a problem with a label attached.
Over-labeling. Burying readers in notices for trivial tool use can dilute the ones that matter.
Inconsistent standards. If you disclose generated images but not generated text, a reader may wonder what else is unsaid. Choose a policy and apply it.
Claiming certainty about detection. No tool reliably proves that content is or is not generated. Avoid accusing others on the basis of a guess.
Make it part of your process
Disclosure is easiest when it is built into workflow rather than added at the end. Keep a short record of which parts of a project used generated material, even a line in a creative brief or project notes. A simple page that explains your general approach to tools can also help readers understand your standards without repeating the full explanation each time.
If you work with clients, collaborators or sponsors, agree on expectations early. Some contracts and guidelines restrict generated material or require its disclosure. Check the terms before delivery, not after. The related matter of paid relationships has its own principles, covered under sponsorship disclosure.
Rules and uncertainty
Laws, platform policies and industry standards in this area are changing and differ between regions. This article offers general principles, not legal advice. For anything with commercial or legal stakes, consult the official guidance from the relevant regulator or platform and, if needed, a qualified professional.
The steady principle underneath is simple: let people know what they are looking at. When in doubt about whether to mention generated material, the usual answer is to say so in a sentence.