Disclosing AI-generated content

Rules on labelling synthetic media are being introduced in several jurisdictions on different timetables. Rather than tracking each, it is more useful to understand the direction and prepare for it.

What the rules are converging on

The consistent principle is deception. Where a reasonable person could take synthetic content for a record of reality, disclosure is being required. Where they could not, it generally is not.

What to do now

Keep a record. For every published synthetic asset: what was generated, with which tool, on what date, by whom, and whether it depicts a person or event.

That record takes seconds to create and cannot be reconstructed afterwards. Organisations publishing at volume without it will not be able to answer questions about their own archive.

Label the clear cases. Realistic people, realistic events, synthetic voices of real people. Visibly, near the content, in plain language.

Check platform policies. Several advertising platforms already require declaration of synthetic content in political and social advertising, and some in all advertising. Those requirements bind you regardless of the law.

How to label

Plainly and adjacent to the content. "This image was generated with AI." Not in a footer, not only in metadata, not in a hover state.

Metadata standards for content provenance exist and are worth adopting, and they are not a substitute for a visible label, because most platforms strip metadata.

The credibility argument

Set aside the legal question. Undisclosed synthetic content that is later identified damages trust disproportionately, because the concealment reads as intent to deceive.

Disclosure costs almost nothing. Audiences have shown themselves largely accepting of synthetic imagery that is labelled and hostile to synthetic imagery that is discovered.

The internal policy

One page, covering: which uses are permitted, which require approval, which are prohibited, who approves, how records are kept, and what the labelling standard is.

Prohibited should include at minimum: synthetic depictions of your own work or premises, synthetic customers or testimonials, synthetic depictions of identifiable people without consent, and synthetic imagery in any context where the audience is relying on the image as evidence.