Labeling AI content: C2PA, SynthID and the EU rules

Content Credentials and watermarks like SynthID are labels that record when a photo or video was made or edited by AI, and since 2 August 2026 the EU AI Act asks creators and platforms to make AI content recognizable within the EU. For you as a creator it comes down to one thing: understand how those labels work and mark your content honestly. This is general information, not legal advice.

What is a Content Credential, a watermark and a platform label?

There are three kinds of AI labels, and they work at different levels. Keeping them apart tells you exactly what is attached to your file.

• Content Credentials (the C2PA standard) are a kind of digital tag inside the file that records who made something, with which tool, and what edits were applied. It is cryptographically signed, so you can verify it is genuine. The standard was set up by Adobe, Microsoft, Intel and the BBC, among others, with Google, OpenAI and Sony joining later.
• An invisible watermark like Google DeepMind's SynthID does not sit in the metadata but is woven into the pixels, audio or text itself. It is imperceptible to the eye and stays detectable even after cropping or compression. Google reports that more than 10 billion pieces of content now carry a SynthID watermark, across image, video, audio and text.
• A visible platform label is the "AI info" or "Made with AI" tag that TikTok, Instagram and YouTube show on or next to your post.

What the EU rules ask of you since August 2026

Since 2 August 2026, the transparency rules in Article 50 of the EU AI Act apply. At their core they ask two things. Providers of AI tools have to mark their output in a machine-readable way as artificially generated. And anyone who publishes a deepfake, meaning image, audio or video that imitates a real person or scene and looks authentic, has to clearly and visibly disclose that it is AI.

Content you made before 2 August 2026 does not have to be labeled retroactively. The rules apply to the EU market and focus mainly on deception: a deepfake that makes someone appear to say or do something falls under them; a clearly fantastical image is more nuanced. When in doubt, honest labeling is always the safe choice.

Why honest labeling works in your favor

Being transparent about AI is not just a rule, it builds trust. Your audience sees through "too perfect" content faster than you think, and an honest note keeps you from looking like you got caught later.

On top of that, platforms increasingly recognize AI content themselves and slap an automatic label on it. That kind of tacked-on label often looks clumsy. If you flip the "made with AI" switch yourself and say it in your own words, you stay in control of how it comes across.

How to mark your AI content properly

Labeling costs you no quality and barely any time. These steps cover most cases.

1. Turn on the "made with AI" switch when you upload. TikTok, Instagram, Meta and YouTube all offer it at posting time.
2. Mention it briefly in your caption where relevant, for example "made with AI".
3. Be extra clear with recognizable people and with news or public-interest topics, where deception weighs heaviest.
4. Don't strip watermarks or labels to pass something off as real. That is exactly what the rules aim to prevent.

What labels do and don't hold on to

No label is watertight, and it helps to know why. Content Credentials live in the metadata and can be lost the moment someone takes a screenshot or a platform strips the metadata on upload. An invisible watermark like SynthID sits in the image itself and survives more editing, but not every platform or tool reads it out.

That is why platforms rely on several layers at once: metadata, watermark and their own visible label. So don't count on a single technique. The most reliable layer stays the one where you state it clearly yourself.

Frequently asked questions

Do I have to label every AI photo?

Not every fantasy image needs a warning, but for content that could look real, and especially with recognizable people, labeling is wise and often required. When in doubt, just label it.

Will my reach drop if I turn on "made with AI"?

No, turning on the AI switch yourself carries no penalty. In fact, a tidy self-applied label looks better than one the platform tacks on afterward.

Can I check whether an image carries a watermark?

For Content Credentials you can inspect the file through a viewer that supports the standard. For SynthID, Google offers a detector that scans image, audio, video and text for its own watermark.

Does this apply outside the EU too?

The EU AI Act covers content on the EU market. Outside the EU the rules differ, but the platform labels from TikTok, Instagram and YouTube work worldwide, so in practice it comes out about the same everywhere.

In short: make your AI content with peace of mind, and label it honestly. Want to create AI photos and videos you can publish cleanly? Create an account and start with the photo generator or the video generator.