# AI disclosure

> What the widget always shows so shoppers know they are talking to an AI, why it can't be removed, and what is rejected on save.

Source: https://shopassist.coec.dk/docs/assistant/ai-disclosure · Updated 2026-10-01

Shoppers must know they are talking to an AI (EU AI Act Article 50). ShopAssist shows it everywhere, on every plan, and you can't switch it off.

## What is shown [#shown]

* An **AI badge** on the launcher and in its accessible name.
* The assistant's name with the badge and an "AI assistant" label in the chat header.
* A first message in every conversation saying the shopper is talking to an AI and can ask for a person.
* The badge on proactive messages and on every AI-written message.
* A footer link "About this AI assistant" to a page in the shopper's language ([ai-info](/ai-info/en)).

When a person from your team takes over, the header shows the agent and keeps a line saying the earlier answers were AI-written.

## Why it can't be removed [#why]

The disclosure is a legal duty for you as the deployer and is part of the [Acceptable Use Policy](/legal/aup). The widget checks after rendering that the badge, header label and first message are visible; if a theme's CSS hides them, the widget shows a plain disclosure banner instead and reports it.

## What is rejected on save [#rejected]

| Text                                                                                                                                             | Rejection                    |
| ------------------------------------------------------------------------------------------------------------------------------------------------ | ---------------------------- |
| Instructions to deny or hide being an AI ("say you're a human", "never mention AI") in brand voice, instructions, greeting or proactive messages | `ai_disclosure_conflict`     |
| An assistant name that suggests a person ("Customer service employee Anna")                                                                      | `persona_name_implies_human` |

A photo of a person as the avatar is not allowed by the AUP; use an illustration or your logo.

## Machine-readable marking [#marking]

AI-written messages are marked in the widget's DOM, and you can verify whether a text was produced by your assistant with the [AI content verification API](/docs/api/ai-content-verification).