Make AI Disclosure a Trust Signal, Not a Footer

The label is the interface

Your feed is filling with synthetic voices. Deezer now receives 90,000 fully AI-generated tracks per day, more than half of its uploads on peak days. The smallest UI element can carry the biggest trust decision. A footer note says the company tried to be honest. A badge on the profile says this identity is different, and it changes what the user can do.

Regulators have moved the question from policy debate to shipping requirement. The EU AI Act's Article 50 transparency rules began applying on August 2. The Act sets maximum fines at 15 million euros or 3 percent of annual turnover.

Spotify plans to begin displaying AI Persona badges on selected artist profiles in mid-September. Starting then, the badge will exclude flagged profiles' music from Discover Weekly and algorithmic recommendations unless the listener already follows the artist. That makes the badge a distribution signal: it tells the product, the user, and the ranking system that the identity needs a different default.

Verify the identity, not just the content

Most teams start with the artifact: an image, a voice, a track. The trust unit is the identity surface: the name, avatar, bio, follower graph, and the promise that the person or project behind it is real. If the identity is synthetic, the label belongs there, not only on each generated item.

Spotify lets artists self-report an AI Persona through Spotify for Artists, and it will review profiles rather than depend only on self-reporting. For a consumer product, self-reporting gives speed, review gives a defensible default, and evals catch drift when the model, prompt, or artist workflow changes.

A badge that appears only after a complaint reads as damage control. If the product cannot say why a profile is flagged, the label will read as arbitrary, and users will learn to ignore it.

Make the disclosure change the default

A label that does not change behavior is decoration. The useful version changes what the product shows, what it recommends, and what the user can choose. The pattern is clean: the default protects discovery, and explicit user choice restores access.

Build the same logic into your own surfaces. The checklist below is the order to ship it.

• Classify the identity before you classify the content. Decide whether the profile is AI-assisted, AI-generated, or a hybrid, and write that definition in the product spec. Done looks like a one-line rule a support agent can explain: this profile is synthetic because the name, image, and voice are generated.
• Verify with review or evals, not vibes. Use a human review queue for high-visibility profiles and automated checks for scale. Done means a flag state with a reason code, a reviewer or model version, and a timestamp.
• Attach the signal to the identity surface. Put the badge on the profile, not only in a metadata field or a long-form policy. The shipped state is the same signal on the avatar, the search result, and the player.
• Change the default ranking for flagged identity. If the identity is synthetic, reduce its access to discovery surfaces unless the user has already chosen it. Done looks like a documented ranking rule: lower default exposure, higher exposure after explicit follow, save, or play.
• Let explicit user choice override the default. A follow, a save, or a direct search should be enough to restore access. Done means the user can keep the relationship without re-reading a legal notice.
• Log regressions when the signal stops working. Track false positives, false negatives, user complaints, and ranking impact. The shipped state is a monthly report that says which profiles were mislabeled and which model or rule caused the miss.