What Should an AI Label Tell You Before You Trust a Voice or Video?

An “AI-generated” label can warn you that a voice, image or video is synthetic, but it does not answer the question most people actually have: what should I trust, and what should I check before I act?

In the European Union, Article 50 transparency obligations have applied since 2 August 2026. The European Commission says people must be informed when they are interacting directly with an AI system, and certain deepfakes must be disclosed clearly. Providers also have machine-readable marking duties, although a limited transition applies to some systems placed on the market before that date. The Commission’s current guidance is here: https://digital-strategy.ec.europa.eu/en/policies/guidelines-ai-transparency-obligations

A visible label and a technical provenance record solve different problems. C2PA Content Credentials can record an asset’s origin, edits and use of AI in a tamper-evident manifest. That can help a viewer inspect how media was made, but the standard is optional. No credential does not automatically mean something is fake, and a credential does not prove that the message inside the media is true. The C2PA explainer is here: https://c2pa.org/specifications/specifications/2.2/explainer/Explainer.html

For an important clip, a useful check has three layers: read the visible disclosure, inspect provenance information when it is available, then corroborate the underlying claim through an independent source before sharing, paying or handing over personal information. For creators, the practical standard should be equally clear: put the disclosure where people encounter the media, say what was generated or materially altered, and preserve provenance information when the tool supports it.

The difficult design question is how much detail helps before a label becomes background noise. If an AI voice or avatar disclosure appeared before you watched, which information would most affect your decision: who made it, which parts were synthetic, the tool used, a link to the original source, or something else?