How to label AI-generated images clearly

A visible disclosure should be easy to find and understandable without specialist knowledge. Choose wording that describes the actual production process: “AI generated” for a fully generated image and “AI manipulated” when AI materially changed an existing image. “AI assisted” is usually more suitable when a person created the principal content and used AI for individual production steps.

Keep the wording specific. A generic “AI inside” provides less information than “Image fully generated with AI” or “Background modified with AI”. Avoid claims about authenticity, truth or authorship when the technical workflow does not establish them.

Text, placement and contrast

Place the disclosure where it will not be cropped or covered by common platform elements. Corner safe areas are a useful starting point, but they still need to be checked against the actual image. Text needs sufficient contrast over both light and dark regions. A semi-transparent background can stabilize legibility without completely obscuring the image.

Size the disclosure for the final publication context. Check both the original-resolution file and a reduced mobile preview. The label should remain readable after a social-media crop or responsive image transformation.

The ContentSignal Disclosure Generator supports custom text and logos as well as the official EU icons for AI-generated content. The icons are optional and already contain the visible label. When an EU icon is selected, the generator therefore renders no additional text.

Practical workflow

Establish how AI contributed to the subject, composition and editing.
Select a precise disclosure type and add custom wording when useful.
Choose between an EU icon with its integrated label and a custom text disclosure. Your own logo can accompany the text option.
Place the label with sufficient contrast inside a reliable safe area.
Export the image at the required resolution.
Inspect legibility, dimensions and metadata in the final file.
Record which version was published.

What to inspect after export

A canvas export can change embedded EXIF, IPTC or XMP data. Before export, explicitly choose whether supported metadata should be preserved, sensitive fields should be cleaned, or a disclosure should also be embedded as XMP. Pixel changes can invalidate existing Content Credentials; they should not be copied in a way that makes them appear valid.

Keep the source file separately. Inspect the actual publication file again, especially after a CMS, platform or image optimizer has processed it. This confirms whether the visible notice remains readable and which machine-readable information survived.

Visible disclosures and machine-readable provenance serve different purposes. Use both layers together where the publishing workflow supports them.