What an Article 50 review report looks like.

1 Scope and role assumptions

Role assumptions (to confirm)

• Example Retail deploys the assistant to its customers.
• Example Retail built the assistant on a third-party model and makes it available under its own name, so it may also be the provider of the chat system. To be confirmed with legal counsel.
• The assistant interacts directly with natural persons in the EU.

2 Reviewed user-facing flow

1. S1User opens the help centre and selects “Chat with us”.
2. S2The chat panel opens and the assistant sends its first message.
3. S3User asks about a delayed order; the assistant replies.
4. S4User asks for a person; the conversation is handed to a human agent.
5. S5Same journey repeated in the iOS app support tab.

3 Findings

1. AE-F-01Missing

   Are people informed that they are interacting with an AI system, at the latest at the first interaction?

   On web, the assistant opens with “Hi, I’m Robin from Example Retail. How can I help?” The first message does not say that Robin is an AI system.

   Basis

   Article 50(1), 50(5)

   Evidence

   EV-01, EV-02

   Priority

   High

2. AE-F-02Needs review

   Is the information provided in a clear and distinguishable manner?

   An “About our assistant” link in the chat footer explains that answers are AI-generated. It is small, low in the panel, and only visible after scrolling on smaller screens.

   Basis

   Article 50(5)

   Evidence

   EV-02

   Priority

   High

3. AE-F-03Missing

   Is the information consistent across the channels where users meet the assistant?

   The mobile app widget shows neither the footer link nor any other statement about AI.

   Basis

   Article 50(1), 50(5)

   Evidence

   EV-03

   Priority

   High

4. AE-F-04Needs review

   Is the team relying on the “obvious from the context” exception, and is that decision recorded?

   The product team assumed the robot avatar makes the AI nature obvious. The assumption is not documented, and the avatar is not shown in the mobile widget.

   Basis

   Article 50(1)

   Evidence

   EV-03, EV-05

   Priority

   Medium

5. AE-F-05Needs review

   Who is the provider of the chat system, and what machine-readable marking of generated text is in place?

   Example Retail built the assistant on a third-party model and offers it under its own name. It may be the provider of the chat system for Article 50 purposes. The vendor documentation reviewed does not describe marking of text outputs.

   Basis

   Article 50(2)

   Evidence

   EV-05

   Priority

   Medium

6. AE-F-06Done

   Does the reviewed flow involve emotion recognition or biometric categorisation?

   No. The assistant processes typed text only and does not use biometric data. Recorded to show the question was considered.

   Basis

   Article 50(3)

   Evidence

   EV-05

   Priority

   None

7. AE-F-07Done

   When a conversation is handed to a human agent, is the change made clear?

   Yes. The panel shows “You are now chatting with Sam from our support team” with a distinct style.

   Basis

   Good practice, not an Article 50 requirement

   Evidence

   EV-04

   Priority

   None

8. AE-F-08Done

   Is there a record of what is disclosed, where and since when?

   Yes. Disclosure wording changes are tracked in the product changelog with dates.

   Basis

   Supports demonstrating what was done

   Evidence

   EV-05

   Priority

   None

4 Evidence notes

5 Prioritised actions

1. A1Priority: High

   Say it is an AI system in the first message, on every channel

   For example: “Hi, I’m Robin, Example Retail’s AI assistant.” Apply the same wording on web and in the app.

   Addresses AE-F-01, AE-F-03 · Owner: Product + Engineering

2. A2Priority: High

   Make the statement visible without scrolling

   Move a short AI label into the chat header, and keep the “About our assistant” link as detail.

   Addresses AE-F-02 · Owner: UX

3. A3Priority: Medium

   Confirm the provider role and text-marking approach

   Ask the model vendor what machine-readable marking is applied to outputs, and confirm the role assumption with legal counsel.

   Addresses AE-F-05 · Owner: Compliance + Engineering

4. A4Priority: Medium

   Record the decision on the “obvious” exception

   Once A1 ships, note that the exception is no longer relied on, with the date.

   Addresses AE-F-04 · Owner: Compliance

5. A5Priority: Low

   Re-check the flow after release

   Capture new evidence for web and app and update this report.

   Addresses AE-F-01, AE-F-02, AE-F-03 · Owner: Product

6 Review call

A 45-minute call with product, UX, engineering and compliance to walk through the findings, agree owners for A1–A5 and list open questions for legal counsel.

7 Limitations

This report records observations of the reviewed user-facing flow on the dates and channels stated. It is an educational and technical readiness review. It is not legal advice, not a formal audit and not a certificate of compliance.