EU AI Act: Do You Have to Label Your Chatbot? (From 2 August 2026)

Kai Viertel, Co-Founder & CMO

From 2 August 2026, the transparency obligations of the EU AI Act apply. If you run a chatbot or AI assistant on WhatsApp, Instagram or live chat, you have to tell customers clearly that they are talking to an AI. What Article 50 requires, who it affects, and how to put it in place in minutes.

Do I have to label my chatbot?

In short: yes, in almost every case. If a chatbot or AI assistant replies automatically on your channels, the EU AI Act requires from 2 August 2026 that the customer can recognise they are talking to a machine and not to a flesh-and-blood employee.

That sounds like a big deal. In practice it is a small one. The law does not require a page-long explanation and no law firm. It only requires the AI to identify itself. A single sentence in the first message is usually enough: Hi, I'm the virtual assistant for Company X.

The concern behind it is legitimate. Modern conversational AI systems write so naturally that many customers do not even notice there is no human on the other end. Avoiding exactly this confusion is what the AI Act is for. It is about honesty, not bureaucracy.

If you already work with an AI bot today, use the coming weeks to review the greeting once. If you don't run one yet, build the disclosure in from the start. The next sections show what this looks like in practice on the WhatsApp, Instagram, Messenger, Telegram and live-chat channels.

From when does the EU AI Act apply?

The transparency obligations of the EU AI Act apply from 2 August 2026. That is the date that matters for businesses with chatbots.

The AI Act does not come into force all at once, but in stages. Different parts of the law take effect at different times. For the day-to-day of a small or mid-sized business that serves customers over messaging, 2 August 2026 is the relevant date: from that day, an AI chatbot has to disclose itself to users.

There is enough time until then to set this up calmly. Unlike many data-protection topics, no complex process chain is needed here. It is enough to get the bot's first message right once and switch the setting on. After that, it runs by itself.

What is Article 50 of the AI Act?

Article 50 is the part of the EU AI Act that governs transparency when interacting with AI. For businesses with chatbots, paragraph 1 is the decisive one.

In substance, Article 50(1) says: providers and deployers of AI systems that interact directly with people must ensure that the people concerned are informed that they are communicating with an AI, unless that is clear from the circumstances.

Translated into the day-to-day of an SME, this means: a chatbot answering questions on WhatsApp counts as an AI system that interacts directly with people. So the labelling obligation applies. The customer has to know they are writing to an AI.

The breadth of the term matters. It makes no difference whether your bot is a simple question-and-answer machine or a learning AI assistant. As soon as an automated system chats with customers on its own, Article 50(1) applies. That covers bots in customer service just as much as in sales or appointment booking.

The EU Commission published its final guidelines on implementing Article 50 on 20 July 2026, days before the deadline. The core obligation under paragraph 1 is clearly worded and easy to implement.

What has to be labelled and what doesn't?

Not every use of AI and not every piece of content falls under the same obligation. The overview below separates the cases that matter for businesses with messaging chatbots.

The message of the table: for an SME with a messaging bot, almost exclusively the first row counts. The bot has to say that it is a bot. The other cases concern either synthetic media or editorial text and rarely play a role in normal customer chat.

When is the AI obvious and the obligation drops away?

Article 50(1) contains an exception: if it is obvious from the circumstances that the customer is talking to an AI, no additional disclosure is needed.

In practice, this exception is tricky. What seems obvious to you often isn't to the customer. A bot that replies under your company name on WhatsApp and writes natural, friendly sentences can easily be mistaken for an employee. That is exactly why we advise not to rely on obviousness.

The safe path is simpler than weighing it up: write the disclosure into the greeting once, and the topic is settled. A clear statement costs nothing, builds trust and pre-empts any debate about questions of interpretation. Better one friendly note too many than an argument over interpretation too few.

Deepfakes and AI text: a separate topic

Besides the chatbot obligation, the AI Act also governs the labelling of AI images, AI audio and AI video, so-called deepfakes, as well as AI-generated text on matters of public interest. These obligations under Article 50(4) are a separate field, dealt with elsewhere, and are generally not relevant for pure messaging chatbots. Important for context: for AI text, the labelling only applies where there was no human editorial control, and ordinary commercial or promotional chat content does not count as a matter of public interest. For most businesses, the bot disclosure under paragraph 1 therefore remains the only point that really counts.

Worked example: an SME with a WhatsApp bot

Take Berghoff Furniture in Galway, a fictional 30-person workshop in the EU. Berghoff gets dozens of enquiries a day over WhatsApp: delivery times, dimensions, care instructions. An AI assistant answers the standard questions around the clock and hands tricky cases to the team.

Before 2 August 2026, the bot opened with: Hello, welcome to Berghoff. How can we help? The customer could believe an employee was writing. That is exactly the situation Article 50 addresses.

After the change, the bot greets like this: Hello, I'm Lara, the virtual advisor at Berghoff, supported by AI. You can speak to a member of the team at any time. With that the disclosure is met, the customer knows where they stand, and the note about handover to the team builds extra trust.

For Berghoff, the change meant exactly one thing: rework the greeting message in the bot settings. No new tool, no project, no outside consultancy. The GDPR-compliant communication over the WhatsApp Business API was already under control. The AI disclosure was added on top as a small, clean extra.

The ready-to-use bot greeting to copy

A good disclosure does three things at once: it names the bot, makes clear that an AI is replying, and shows the way to a human. Here are three templates you can adapt right away.

Template 1: customer service

Template 2: sales and advice

Template 3: short and simple

All three variants put the disclosure in place. Which one fits depends on your brand. The only thing that matters: the note about the AI sits in the first message, not hidden in the small print. This greeting is shown to the customer as the bot's very first message.

Put it in place in SendSeven: one setting

With SendSeven, the chatbot disclosure under Article 50(1) is not a DIY job but a ready-made feature. The AI assistant has a built-in setting called AI Identity and Disclosure.

There you store the disclosure text the bot sends as its first message. It comes pre-filled with a suggestion along the lines of: Hi, I'm Lara, virtual advisor at Company X, powered by AI. You adapt the name and the wording to your brand, done.

There is also a toggle that adds the note about the human handover to the greeting: You can speak to a member of the team at any time. With that you meet the disclosure and at the same time communicate the hybrid approach, the interplay of AI and team.

This single setting applies across all the channels your AI assistant runs on: WhatsApp, Instagram, Messenger, Telegram and the website live chat. That makes the disclosure obligation easy to put in place with SendSeven and gets your bot ready for the EU AI Act. No software can or should replace a guarantee of full legal compliance; the individual review of your specific case remains your responsibility. But you'll have handled the technical part of the obligation.

Is the privacy policy enough?

A common assumption goes: we have a privacy policy that mentions the chatbot, so that's enough. This assumption is risky.

The EU AI Act and the GDPR pursue different aims. The GDPR governs how you process personal data. The AI Act governs, among other things, that the person in the conversation recognises they are talking to a machine. A privacy policy linked somewhere on the website does not meet the second requirement. Nobody reads it before asking a question in the chat.

The disclosure has to happen where the conversation happens, that is, in the chat itself and at the moment of interaction. That is why the note belongs in the first bot message and not only in a legal document. Both obligations exist side by side: you still need a clean privacy policy and, on top of that, the AI disclosure in the chat.

EU guidelines and the voluntary Code of Practice

On 10 June 2026, the EU Commission published a voluntary Code of Practice on labelling and marking AI-generated content. On 20 July 2026, its final guidelines on implementing Article 50 followed. Both are a practical way to show that transparency obligations are being met.

Important for context: the EU's Digital Omnibus of June 2026 postponed some obligations for high-risk AI, but explicitly not the Article 50 transparency obligations. 2 August 2026 stays binding for chatbot disclosure. For the machine-readable marking of AI-generated media there is a shortened transition period until 2 December 2026, which barely matters for pure messaging bots.

Voluntary means: nobody forces you to sign up to the code. Anyone who implements the core obligations under Article 50 cleanly does not depend on it. For an SME with a messaging bot, the clear disclosure in the chat remains the core.

Your checklist in 6 steps

Here is how to get your chatbot on track by 2 August 2026:

• Take stock: on which channels do you run an AI chatbot? WhatsApp, Instagram, Messenger, Telegram, live chat?
• Check the greeting: does the first bot message say clearly that an AI is replying? If not, adapt it.
• Write the disclosure text: name of the bot, note about the AI, reference to the human handover. Use the templates from this article.
• Switch the setting on: in SendSeven, store the text under AI Identity and Disclosure and turn on the handover note.
• Keep data protection running in parallel: the privacy policy is still needed on top, it does not replace the in-chat disclosure.
• When in doubt, get it reviewed: for special set-ups, such as heavily automated or cross-border applications, an individual legal assessment is worthwhile.

If your bot starts from ads, the article on GDPR-compliant Click-to-WhatsApp Ads helps. And if you're weighing up where automation should stop and a human should step in, read our comparison Live Chat or Chatbot.

Frequently asked questions

Do I have to label my WhatsApp chatbot?

Yes. An AI chatbot on WhatsApp interacts directly with people and therefore falls under Article 50(1) of the EU AI Act. From 2 August 2026, the customer has to be able to recognise that they are writing to an AI. In practice, a clear note in the bot's first message is enough.

From when does the EU AI Act apply to chatbots?

The transparency obligations of the EU AI Act, which include chatbot disclosure, apply from 2 August 2026. Until then, businesses should adapt the first bot message so the AI is clearly recognisable.

What is Article 50 of the AI Act?

Article 50 governs transparency when interacting with AI. Paragraph 1 requires that deployers of AI systems communicating directly with people inform users that they are dealing with an AI, unless that is already obvious. That is exactly what affects chatbots and AI assistants in messaging.

Is a privacy policy enough as a label?

No. The privacy policy meets GDPR requirements, not the AI disclosure of the AI Act. The note that an AI is replying has to happen in the chat itself and at the moment of the conversation, that is, in the first bot message. The privacy policy remains required on top.

Do I have to label it if it's obvious it's a bot?

In theory the obligation drops away if it is clear from the circumstances that an AI is replying. In practice that is hard to judge, because modern bots write very much like humans. The safe path is to write the disclosure into the greeting anyway. It costs nothing and builds trust.

Does the obligation also apply to Instagram, Messenger, Telegram and live chat?

Yes. Article 50(1) is channel-independent. As soon as an AI chatbot writes directly to people, the disclosure obligation applies, whether on Instagram, in Facebook Messenger, on Telegram or in the website live chat. With SendSeven, the AI disclosure works as a single setting across all channels.

What happens with AI text and deepfakes?

That is a separate topic in Article 50(4). AI images, AI audio and AI video, so-called deepfakes, must be labelled. AI-generated text on matters of public interest must be too, but only where there was no human editorial control. Ordinary marketing and service chats do not count as matters of public interest; here only the bot disclosure under paragraph 1 applies.

Do I need a lawyer for the labelling?

For the standard implementation, usually not. A clear disclosure in the bot's first message meets the core obligation under Article 50(1). For special or cross-border set-ups, an individual legal assessment is sensible. The information in this article is orientation and does not replace legal advice.