Does Your Chatbot Identify as AI? Free 2-Minute Test

The Question Every Company with a Chatbot Should Ask

Starting August 2026, Article 50 of the EU AI Act requires every AI chatbot to identify itself as such to the user. The question is simple: does your chatbot already do this?

The reality is that most enterprise chatbots don't meet this requirement. Many present themselves as "assistants" without specifying they're AI, and some even use human names.

What Counts as "Identifying as AI"

The EU AI Act doesn't define a specific format, but the AESIA guidance establishes three levels of disclosure:

1. Pre-Interaction

Before the user clicks on the chatbot widget, a visible indicator must exist:

Good examples:

• Widget labeled "AI Assistant"
• Robot icon + "AI Chatbot" text
• Banner: "This chat is powered by artificial intelligence"

Bad examples:

• Widget that only says "Chat"
• Human person icon
• No label at all

2. First Interaction

The chatbot's first message must contain an explicit declaration:

Good examples:

• "I'm an AI-powered virtual assistant"
• "Hi, I'm an AI designed to help you with..."
• "This is an automated chatbot powered by AI technology"

Bad examples:

• "Hi, I'm Sarah, how can I help you?"
• "Welcome to our support service"
• Emojis without AI declaration

3. Responses

The chatbot's responses should maintain transparency:

Good examples:

• Including "As an AI assistant..." periodically
• Automatic signature like "— [Company] AI Assistant"
• Disclaimer when providing sensitive information

Bad examples:

• Responses that completely mimic a human
• Never reminding the user it's AI
• Hiding its artificial nature

How the Ercel Test Works

Our automated scanner checks all three disclosure levels in under 2 minutes:

1. Visual scan: Analyzes your page's HTML looking for transparency indicators near the chatbot widget
2. First interaction: Opens a conversation and analyzes the first message for disclosure keywords
3. Response analysis: Sends test messages and checks if responses include AI indicators

The result is a report with Art. 50 compliance status: compliant, partial, or non-compliant, with specific recommendations for each checkpoint.

What to Do If Your Chatbot Doesn't Comply

If the test reveals your chatbot doesn't identify as AI, the changes are usually straightforward:

1. SaaS platforms (HubSpot, Tidio, Intercom, Crisp): Configure the bot name and welcome message from the admin panel
2. Custom chatbots: Update the welcome text and add a label to the widget
3. Direct APIs (OpenAI, Anthropic): Modify the system prompt to include self-identification

Most changes can be implemented in less than an hour. The cost of not doing so can reach 15 million euros.

Don't Wait Until August 2026

Supervisory authorities are already building their inspection capacity. National AI offices across Europe will begin proactive inspections as soon as the deadline passes.