EU AI Act Transparency Disclosure Builder

Tick the Article 50 situations that match your AI system and get a disclosure pack: ready-to-paste wording for each duty, where each notice has to appear, and the checks that wording alone cannot do. Free. Nothing you type is sent or stored anywhere.

How it works: Article 50 of the EU AI Act lists five transparency situations. They sit on top of your risk category — a chatbot with no high-risk features still owes the first one. Work through the five cards below, tick only what is actually true for your system, and the pack at the bottom builds itself. If you have not classified your system yet, start with the free Risk Classification Self-Check; the reasoning behind every duty here is in the Transparency Obligations Guide.

Your system

These fields only label and personalise your pack. They never leave this page. The role matters because Article 50 splits its duties: duties 1 and 2 sit with providers, duties 3, 4 and 5 with deployers — the pack flags any duty that normally belongs to the other role, so you can confirm it with your vendor in writing instead of assuming.

The five Article 50 situations

Progress: 0 of 5 duties in your pack

A support chatbot, voice agent, AI receptionist or in-app assistant that people talk to directly. People must be informed they are interacting with an AI system, at the latest at the first interaction, in a clear and distinguishable way — unless it is already obvious from the context, judged from the person's side.

Our system generates or manipulates audio, images, video or text. Its outputs must be marked in a machine-readable format and detectable as artificially generated or manipulated. This builder cannot create that mark — for this duty the pack gives you the visible caption wording plus the exact written confirmations to get from the provider of the generative system.

We publish image, audio or video content that was generated or manipulated by AI and resembles real people, objects, places or events. It must be disclosed as artificially generated or manipulated. For works that are evidently artistic, creative or fictional, the disclosure can be lighter — in a way that does not hamper enjoying the work — but it does not disappear.

We deploy a system that recognises emotions or categorises people from biometric data. The people exposed to it must be informed that it is operating, and their personal data must be processed in line with GDPR and other applicable EU data law. Warning: emotion inference in workplaces and education is on the prohibited list, with narrow exceptions — a notice does not make a banned use legal.

We deploy AI to generate or manipulate text published to inform the public on matters of public interest — news-style posts, public announcements, community updates. The text must disclose that it is AI-generated, unless the use is authorised by law, or the content passed human review or editorial control and a named person or business holds editorial responsibility for it. Either way, keep a record of which route each piece took.

Your disclosure pack

Tick a situation above

Your wording, placement notes and checks will appear here as you tick the situations that apply. Duties you do not tick are listed at the end of the printed pack as checked-and-not-applicable, with today's date — that record is part of the point.

What this builder does not do

This builder does not create a machine-readable mark, does not classify your system, and does not make any wording legally sufficient for your exact surface and audience. It turns the five Article 50 situations into template wording and a placement plan based on the public text of the Act. Three traps it will not save you from: hiding a notice in your terms (the duty is clear and distinguishable, at first interaction or exposure); assuming a vendor handled your deployer duties (they stay with the business that publishes or operates the system); and treating a notice as a cure for a prohibited use (emotion inference at work or in education is banned, notice or not). Timeline as of November 2026: the Article 50 duties have applied since 2 August 2026; the grace period for machine-readable marking by generative systems already on the market ends 2 December 2026.

How to use the pack

  1. Place each notice where the interaction actually happens. The chat window, the video description, the point of exposure — not a legal page nobody opens.
  2. Send the Duty 2 confirmations to your vendor, in writing. “We handle compliance” is not an answer you can show a partner or a regulator; a named marking method is.
  3. File the pack with its date. The not-applicable list matters too: it shows the five situations were checked, not ignored.
  4. Re-run it when anything new ships. Every new chatbot surface, voice agent or content format restarts the Article 50 question.
  5. Read the reasoning. The Transparency Obligations Guide explains each duty, the provider/deployer split and the 5 mistakes that break an otherwise fine disclosure.

Want your duties mapped in writing, for your system?

The EU AI Act Classification Report maps your specific system against the Act: risk category with reasoning, the obligations that follow — including its Article 50 transparency duties — your deadlines and a step-by-step plan in plain English. $49 one-time, delivered by email. Optional Act Watch is $19/month and re-checks your classification as the rules move. Not legal advice.

Get My Report — $49 Answer the full questionnaire

Frequently asked questions

Does this builder create the machine-readable mark for AI content?

No. Machine-readable marking is a technical signal inside or attached to the output itself — metadata, watermarking or an equivalent — applied by the provider of the generative system. This builder produces the visible wording, the placement plan and the written questions for your vendor. A caption alone does not satisfy the marking duty.

Is the wording in my pack guaranteed compliant?

No. It is a plain-English template based on the public text of Article 50. Whether a notice is clear and distinguishable enough, and in time, depends on your exact surface and audience. A written classification of your specific system, or a lawyer, settles borderline cases.

What happens to what I type?

Nothing leaves your device. The builder runs entirely in your browser, sends no data anywhere and stores nothing. Close the tab and the inputs are gone — print or copy the pack to keep it.

I deploy a chatbot built by someone else. Are these my duties or the vendor's?

Both roles carry duties, for different things. The interaction notice and the machine-readable marking sit with the provider; the deepfake, emotion-recognition and public-interest text disclosures sit with the deployer. If you put a vendor chatbot in front of your customers, confirm in writing that the notice is actually shown in your surface — and keep your deployer duties either way.

When do the Article 50 duties apply?

The Article 50 transparency duties have applied since 2 August 2026. A grace period for the machine-readable marking of content from generative systems already on the market ends on 2 December 2026 (information as of November 2026).

SkillQuesters is independent and is not affiliated with the European Commission. Primary sources: Regulation (EU) 2024/1689, Article 50, and the European Commission's AI Act pages. This page contains no affiliate links for the service above — the Classification Report and Act Watch are SkillQuesters' own services.