Start with what your audience might reasonably assume
A disclosure is useful when it corrects a likely misunderstanding. Readers may assume an article reflects a writer’s own experience, that a photograph shows a real event, or that a voice recording belongs to the person it resembles. If AI changes one of those assumptions, say so. The aim is not to list every tool used in production. It is to give people the context they need to interpret what they are seeing or hearing.
There is a meaningful difference between using AI to suggest headline options and publishing a largely AI-written article. Likewise, removing background noise from a recording is different from generating a speaker’s voice. The more a tool shapes the substance, apparent origin or authenticity of a piece, the stronger the case for a clear disclosure. Consider the expectations created by the format, too: an obviously fictional illustration raises different questions from a realistic image presented alongside news or a personal account.
Decide what needs to be disclosed
Ask three questions before publishing. Did AI create or substantially rewrite material the audience will encounter? Could someone mistake synthetic material for a real person, place, event or first-hand experience? Would knowing about the AI contribution change how a reasonable person evaluates the content? A “yes” to any of these is a good reason to disclose, even if the finished work was carefully edited.
Disclosure is especially helpful for realistic images, cloned or generated voices, synthetic video and first-person accounts assembled by AI. It also matters when an AI system supplies explanations, recommendations or summaries that readers might otherwise take as independently checked. Minor assistance, such as spelling suggestions or formatting, generally needs less attention unless a venue’s rules call for it. If you publish through a platform, client or organization, check its requirements as well; they may be more specific than your own practice.
This is a judgment about audience understanding, not a test that a tool can settle for you. AI-detection tools can produce misleading results, and an absence of an AI label does not establish that something was made without AI.
Write a specific notice in a visible place
A useful disclosure says what AI did and, where relevant, what a person did afterward. “This article was drafted with an AI tool and edited and fact-checked by the author” tells readers more than “AI-assisted.” Only claim fact-checking if you actually checked the factual statements. For an image, a caption such as “AI-generated illustration; it does not depict a real event” addresses a different concern. For audio, identify a generated voice before a listener could reasonably mistake it for a real speaker.
Put the notice close to the content it describes. A caption works for a single image; an introduction or editor’s note can work for an article; an on-screen label and accompanying description can work for video. If a piece contains several kinds of synthetic material, one general note may not be enough. Make it easy to connect each notice to the relevant material, including when a clip or image is shared apart from the original post.
Use ordinary language rather than vague claims that imply more oversight than occurred. “Reviewed” might mean anything from a quick read to a line-by-line verification. If the distinction matters, explain it: for example, “A person edited the text for clarity; product details were not independently verified.” A disclosure should inform the audience, not function as a badge of quality.
Keep disclosure separate from accuracy and consent
Labeling content does not make it accurate. AI output can contain plausible mistakes, omit context or present uncertain information confidently. Before publishing, check claims you can verify, inspect generated visuals for misleading details and remove assertions you cannot support. If you cannot verify a consequential claim, do not rely on a disclosure to excuse it.
Likewise, a label does not resolve concerns about using someone’s likeness, voice or private information. Seek appropriate permission before creating a recognizable synthetic portrayal, and avoid entering sensitive audience or client material into a tool without understanding how it will be handled. Disclosures can explain what an audience is receiving; they cannot replace responsible choices about how it was made.
Make the practice repeatable
For a small team, a simple publication checklist helps: record which tools shaped the final piece, identify any realistic synthetic media, decide what a reader could misunderstand, and draft a notice before release. Give one person responsibility for checking that the notice matches the finished work. If the content changes substantially, revisit the wording.
The best disclosure is proportionate, prominent and truthful. It neither hides a meaningful AI contribution nor overwhelms readers with irrelevant production details. When uncertain, explain the contribution in plain terms and let the audience judge the work with that context.
