Labelling of AI-generated content

Why this labelling exists

Artificial intelligence can now produce images and text that are hard to tell apart from human work. Anyone reading or looking at something should know what they are dealing with.

Article 50 of the European Union’s AI Act requires AI-generated content to be disclosed. It does not prescribe what that disclosure looks like. The European Commission has published a set of uniform symbols for the purpose. Using them is voluntary — we do, because one shared sign helps more than anything we could invent ourselves.

The three symbols

We use the Commission’s symbols unaltered.

AI was involved in the creation

The base symbol. It appears where artificial intelligence was involved without us quantifying the share more precisely.

Fully AI-generated

The entire content was produced by artificial intelligence. For an image that means: it shows no photograph, but a computation. For text: it was not written by a person.

Partially AI-modified

There was a human original that artificial intelligence altered. A photo that was extended or retouched afterwards. A text that was reworked.

What it looks like

Two examples from practice. The image areas here are placeholders — we do not put the label on real photographs just to demonstrate it.

Placeholder

Fully AI-generatedImage is

Example article

The body text of the article ends here. Below it follows the label, if artificial intelligence was involved in the text.

Images in the same article are labelled independently — each on its own.

Article is

Where the label appears

• —On an image the symbol sits in the bottom right corner. It appears wherever the image is shown, including in overviews and on small tiles.
• —At the end of an article the symbol appears when artificial intelligence was involved in the text.
• —Image and text are judged separately. An article can have an AI-generated image and an entirely editorial text — then the label appears only on the image.
• —Where no symbol appears, no artificial intelligence was involved as far as we know.

Limits

• —We label what we know. With material that reaches us from third parties — press photos, community submissions — we cannot rule out that artificial intelligence was involved without it being disclosed to us.
• —When an article is shared on social networks, those platforms generate their own preview image. The symbol does not appear on it. On our own site it does.
• —Plain tools do not count as involvement for us: spell checking, translation help, automatic cropping. We label when content is created or altered.