Instructions for notifying the use of AI and marking AI-generated content
Do I need to label my AI usage?
In teaching and research, good citation practices must always be followed. Guidance on citing AI-generated content, such as images produced by AI systems, can be found, for example, in the ImagOA guide (https://libguides.aalto.fi/imagoa_eng).
What labels does the AI Act require?
The transparency obligation - the obligation under Article 50 of the AI Act to label AI-related content or interactions - entered into force on 2 August 2026.
When the AI Act requires disclosure, the label is not optional. The label must be clear, visible, and appropriate to the medium.
Quick guide to labeling
| Content type | Where to put the label | Suggested labels* |
| AI interactions (chatbots, AI agents, avatars) | Display a notice before or at the start of the interaction. | "You are interacting with an AI system." "This is an AI-powered assistant, not a human." |
| Images | Place the label on or next to the image. | "AI-generated image" "Modified using AI" |
| Videos | Show the label at the start of the video or throughout, where appropriate. Keep it visible long enough to be read. | "AI-generated video" "Modified using AI" "This video contains AI-generated content" |
| Audio | Use a spoken or written disclosure at the start or alongside the content. | "This audio contains an AI-generated voice." "This voice was generated using AI." "This audio was modified using AI." |
| Text (public interest matters) | Place the label at the start of the article, summary, post, or publication. | "AI-generated text" "Generated by AI" |
|
Emotion recognition / biometric categorisation |
Must be considered case by case. | Must be considered case by case. |
*The EU has created a set of official icons that can be used to label AI-generated or AI-manipulated content (https://digital-strategy.ec.europa.eu/en/policies/eu-icons-labelling-ai-generated-content). Using these specific icons is optional..
Yksityiskohtaiset ohjeet
Not all AI use requires labelling or disclosure. When labeling is not required by AI Act it still might be a good practice or requird e.g. by a publisher.
The obligations in Act apply mainly to the following categories.
A. AI systems that interact directly with people
When an AI system is designed to interact directly with people — for example, a chatbot, AI agent, or AI-powered avatar — users must be clearly informed that they are interacting with an AI system, not a human. This obligation applies regardless of the content being discussed.
The only exception is where it is obvious from the circumstances and context that the person is dealing with an AI system.
B. Deepfakes
A deepfake is AI-generated or AI-manipulated image, audio, or video content that resembles existing persons, objects, places, entities, or events and could mislead people into believing it is authentic.
Examples include:
- a video of a real person saying or doing something they did not say or do
- an audio recording that imitates a real person's voice
- an image of a real person, place, or event that has been altered to appear authentic
Deployers must label deepfakes clearly and visually so that people can recognise the content as artificially generated or manipulated.
C. AI-generated text about public-interest matters
AI-generated or AI-manipulated text must be labelled if it is:
- published to inform the public
- about a matter of public interest, such as politics, public health, public security, environmental protection, consumer safety, fundamental rights or government decisions
Labelling is required unless the text has undergone substantive human review or editorial control (not merely spell-checking or grammar correction) and a natural or legal person holds editorial responsibility for the publication.
D. Emotion recognition and biometric categorisation systems
People must be informed when they are exposed to AI systems that use:
- emotion recognition — for example, tools that claim to detect emotions, mood, attention, stress, or engagement from facial expressions, voice, movement, or other data
- biometric categorisation — for example, tools that categorise people based on biometric data, such as face, voice, gait, or other physical or behavioural characteristics
This notice must be provided clearly before or at the time the system is used.
If an organisation develops or provides AI systems that generate synthetic content (images, audio, video, or text), it must ensure that the outputs are marked in a machine-readable format so they can be detected as artificially generated or manipulated. This includes the use of watermarks, metadata, or other technical solutions.
If an organisation only deploys AI systems developed by others, it should verify that the providers comply with this marking requirement.
The European Commission has published a voluntary Code of Practice on marking and labelling of AI-generated content, which sets out practical steps for providers and deployers to meet these obligations.
Labelling may not be required, although it may still be good practice, where:
- for deepfakes: the content is clearly artistic, creative, fictional, satirical, or similar, and is presented as such (note: for AI-generated text, this specific exemption does not apply — however, artistic or fictional text typically falls outside scope because it is not published to inform the public on matters of public interest)
- AI-generated text has undergone human review or editorial control, and a natural or legal person holds editorial responsibility for the publication
- the use is authorised by law, for example to detect, prevent, investigate, or prosecute criminal offences
Note that the Commission's Article 50 guidelines provide detailed definitions, practical examples, and exemptions.