AI labelling means telling the audience that content was wholly or partly created or altered with AI. It can be a visible line of text, a symbol or an invisible, machine-readable marking in the file.
The EU AI Act's transparency obligations set out some situations where AI must be disclosed, including deepfakes. Providers of the tools must also mark generated content in a machine-readable way. But there is no general requirement to label all AI-assisted content in external communication.
For most organisations, AI labelling is therefore a choice. If everything is labelled, the label loses its meaning. If nothing is labelled, you risk the audience's trust on the day it comes out.
In practice
A useful principle is to label whenever the audience would feel misled without it. That typically applies to realistic images, synthetic voices and personal messages. Write the rule into your editorial AI policy so it is not decided task by task.
The concepts in this glossary run through my AI courses and workshops, where we work with them on your own tasks.