People increasingly deal with chatbots, voice agents, generated content and automated decisions without being told, and both the EU AI Act and the GDPR now expect organisations to say so clearly and in time. This template gives you a set of numbered, ready-to-adapt notice sections: a short "you are talking to AI" disclosure, a content label, an automated decision notice and the supporting governance text that keeps them accurate. It is built around Article 50 of the EU AI Act, which the European Commission's AI Act Service Desk timeline shows applying from 2 August 2026, and GDPR Articles 13, 14, 15 and 22. It is for product owners, privacy teams and the engineers who build AI features and agents. For the legal analysis behind each duty see the EU AI Act Article 50 Transparency Obligations Guide; this page gives you the wording and the process.
How to use this template
- Replace text in square brackets, such as [Organisation], [AI system name] and [Privacy contact], with your own details.
- "Must" marks text tied to a legal duty described below; "should" marks recommended practice. Remove any section that does not apply, and record why in section 2.
- Use section 1 to decide which notices a system needs. Most systems need more than one: a customer service agent that also generates images and scores applications can trigger Article 50(1), 50(2) and GDPR Article 22 at once.
- Keep notices in your product's source control or content system, versioned, so you can show what users saw on a given date.
- Have the final text reviewed by your legal and data protection teams. Status line: checked against the AI Act Service Desk and EUR-Lex on 27 October 2026; AI Act dates have been amended once by the Digital Omnibus on AI, so check again before launch.
1. Scope and triggers
This notice set applies to every AI system that [Organisation] provides or deploys and that interacts with people, generates content, recognises emotions or categorises people biometrically, or takes or supports decisions about individuals.
| Trigger | Legal basis | Who must act | Notice in this template |
|---|---|---|---|
| AI system interacts directly with people (chatbot, voice agent, email agent) | AI Act Art. 50(1) | Provider designs the system so people are informed | 3.1 Interaction disclosure |
| System generates synthetic audio, image, video or text | AI Act Art. 50(2) | Provider marks outputs in a machine-readable format | 3.2 Content marking statement |
| Emotion recognition or biometric categorisation | AI Act Art. 50(3) | Deployer informs people exposed | 3.3 Exposure notice |
| Publishing deep fake media, or AI text on matters of public interest | AI Act Art. 50(4) | Deployer discloses | 3.4 Content label |
| Solely automated decision with legal or similarly significant effect | GDPR Art. 13(2)(f), 14(2)(g), 22 | Controller | 3.5 Automated decision notice |
| Decision based on an Annex III high-risk system output | AI Act Art. 86 | Deployer explains on request | 3.6 Explanation request route |
Exceptions to record, not assume: Article 50(1) does not apply where it is obvious to a reasonably well-informed, observant and circumspect person that they are dealing with AI. Article 50(2) does not cover assistive standard editing or systems that do not substantially alter the input. Article 50(4) allows a lighter disclosure for evidently artistic, creative, satirical or fictional works, and exempts AI text that has undergone human review or editorial control with someone holding editorial responsibility. If you rely on an exception, write down the reasoning and the reviewer in section 2.
2. Roles and contacts
- Notice owner: [Product owner for AI system name] keeps the notice accurate and approves changes.
- Privacy contact: [Data Protection Officer or privacy team, email] answers questions and access requests under GDPR Article 15(1)(h).
- Human review contact: [Team and channel] handles requests for human intervention and contests under GDPR Article 22(3), and explanation requests under AI Act Article 86.
- Engineering owner: [Engineering lead] makes sure the notice is shown at the right moment and that content marking is applied.
- Register entry: each AI system has a record in [AI system register] listing which notices apply, which exceptions are relied on, who approved them and the current notice version.
3. Notice texts
Article 50(5) requires the information to be given in a clear and distinguishable manner, at the latest at the time of the first interaction or exposure, and to meet applicable accessibility requirements. GDPR Article 12(1) requires information in a concise, transparent, intelligible and easily accessible form, using clear and plain language. Write for the person on the other end, not for the regulator.
3.1 Interaction disclosure (chat, voice, email agents)
Short form, shown before or with the first reply: "You are chatting with [AI system name], an AI assistant from [Organisation]. It is not a person. [You can ask for a human at any time by typing 'agent'.] [Learn how it works and how we use your data.]"
Voice form, spoken at the start of the call: "Hello, this is [AI system name], an automated AI assistant calling for [Organisation]. [Say 'person' to speak to a member of staff.]"
Agent acting by email or messaging on someone's behalf: "This message was written and sent by an AI agent operated by [Organisation] on behalf of [team or person]. Replies are [read by the agent / reviewed by staff]."
3.2 Content marking statement (generative features)
"Images, audio, video and text produced by [AI system name] are marked as AI-generated using [machine-readable method, for example C2PA Content Credentials and metadata or watermarking]. Do not remove these marks." Article 50(2) asks for marking that is machine-readable and detectable, using solutions that are effective, interoperable, robust and reliable as far as technically feasible. The Service Desk timeline gives providers of generative systems already on the market before 2 August 2026 until 2 December 2026 to comply with Article 50(2). See the C2PA Content Credentials Guide.
3.3 Exposure notice (emotion recognition or biometric categorisation)
"[Location or service] uses an AI system that [analyses facial expressions to estimate emotional reactions / assigns people to categories based on biometric data]. It is used for [purpose]. Your personal data is processed under [lawful basis]; see [privacy notice link] or contact [Privacy contact]." Article 50(3) also requires that personal data is processed in line with the GDPR and other applicable EU data protection law.
3.4 Content label (published media and text)
Deep fake media: "This [video / image / audio] was generated or altered using AI."
Creative or satirical work: "Contains AI-generated content." shown in a way that does not hamper the work, such as in credits or a caption.
Text informing the public: "This article was generated with AI." Omit only where the text has been through human review or editorial control and [named editor or desk] holds editorial responsibility, and record that in section 2.
3.5 Automated decision notice (GDPR)
Include in the privacy notice and at the decision point:
"We use an automated system to decide [what is decided, for example whether to approve your application for X]. The decision is made [solely by the system / by the system with review by staff]. It considers [main factors, for example income, repayment history and information you provide] and [how they are weighed, in plain terms]. The result can mean [envisaged consequences, for example approval, a lower limit or refusal]. We rely on [contract / law / your explicit consent] for this. You can ask a member of staff to review the decision, give your point of view and contest it by contacting [Human review contact] within [period]."
This text delivers what Articles 13(2)(f) and 14(2)(g) require: the existence of automated decision-making, meaningful information about the logic involved, and the significance and envisaged consequences. Where you rely on contract or explicit consent, Article 22(3) requires at least the right to obtain human intervention, to express a point of view and to contest the decision. Article 22(4) restricts decisions based on special category data, so state if such data is never used.
3.6 Explanation request route (AI Act Article 86)
"If a decision about you was based on the output of [AI system name] and you believe it has adversely affected your health, safety or fundamental rights, you can ask us for a clear explanation of the role the AI system played and the main elements of the decision. Contact [Human review contact]." Article 86 applies to deployers of Annex III high-risk systems (other than point 2) for decisions with legal or similarly significant effects; check the Commission timeline for when the high-risk rules apply to your system.
4. Placement and timing
| Channel | Where the notice goes | When | Evidence to keep |
|---|---|---|---|
| Web or app chat | Banner above the input box and first message | Before or with the first AI reply | Screenshot per release, UI component version |
| Voice and IVR | Spoken greeting | First seconds of the call | Call script version, sample recordings |
| Outbound email or messaging agent | Header line or signature | Every message | Template ID, sending identity |
| Generated media | Embedded metadata or watermark plus visible label where published | At generation and publication | Marking configuration, verification test results |
| Automated decisions | Privacy notice and decision screen or letter | At data collection and at the decision | Notice version linked to each decision record |
- Notices must be readable by screen readers, available in the languages the service is offered in, and not hidden behind a scroll or a "more" link.
- Do not rely on the user's terms of service alone. A clause buried in terms is not "clear and distinguishable" at first interaction.
5. Accuracy and change control
- The notice owner reviews each notice when the model, provider, data sources, decision logic or purpose changes, and at least every [12] months.
- Changes to section 3.5 decision logic text are approved by [Privacy contact] before release.
- Each notice carries a version number and effective date, kept in [repository or CMS].
- Release checklists for AI features include "transparency notices updated and tested".
6. Requests, complaints and human review
- Requests for human review, contests and explanation requests are logged in [case system] with the decision ID, received date, reviewer and outcome.
- Reviewers have real authority to change the outcome. The ICO notes that a human review is meaningful only where the reviewer has real discretion; token involvement does not make a decision less than solely automated.
- Target response time: [X working days]; GDPR access requests follow [Organisation]'s data subject request procedure.
7. AI agents and machine identities
- Every AI agent that contacts people uses a sending identity (email address, phone number, bot account) registered to that agent, so recipients and auditors can tell agent messages from staff messages.
- Agents must not be configured to claim to be human, and prompts must not instruct them to deny being AI.
- Logs link each interaction to the agent identity, model version and notice version shown. See the AI Agent Audit Trails for Regulators Guide.
8. Review
This notice set is owned by [AI governance lead] and reviewed at least annually, and whenever EU AI Act guidance, codes of practice or implementing acts on marking and labelling are published. Article 50(7) provides for codes of practice on detecting, marking and labelling AI-generated content and allows the Commission to adopt implementing acts.
Implementation guidance
- Inventory first: list every AI feature and agent, mark each trigger in section 1 and note whether you are provider, deployer or both. See the EU AI Act Provider vs Deployer Obligations Guide.
- Start with the interaction disclosure: it is the most visible and the easiest to get wrong. Ship it as a shared UI component so every team uses the same text.
- Test with real users: check that people in a pilot group can say, after one interaction, that they were talking to AI and how to reach a person.
- Align with the privacy notice: sections 3.5 and 3.6 must match the privacy notice and the DPIA. See the DPIA Template and the AI and GDPR Guide.
- Verify marking technically: run a detection check on sample outputs in each release, not just a configuration review.
- Measure: track the share of AI systems with current approved notices (target 100%), human review requests and their outcomes, the share of reviews that change the decision, and notice text versions older than 12 months.
- UK organisations: the ICO states its automated decision-making guidance is under review following the Data (Use and Access) Act 2025, which became law on 19 June 2025. Check the current ICO guidance before reusing section 3.5 for UK users. See the UK GDPR and Data Protection Act 2018 Guide.
Sources: European Commission AI Act Service Desk: Article 50 · AI Act Service Desk: Implementation timeline · AI Act Service Desk: Article 86 · EUR-Lex: Regulation (EU) 2016/679 (GDPR) · ICO: What does the UK GDPR say about automated decision-making and profiling?
This template summarises transparency duties under the EU AI Act and the GDPR, must be adapted to your organisation and is not legal advice. Related NHI Mgmt Group resources: EU AI Act Compliance Guide · EU AI Act Article 50 Transparency Obligations Guide · Global AI Regulation Map · Safe Use of AI Chatbots Guide