Warranty support for every website and app 🛡️

Artificial intelligence

Does your website chatbot have to say it is AI?

Since 2 August 2026, people must be told when they are talking to an AI system. What this means for the chatbot on your website, who is responsible and what the notice should look like.

Does your website chatbot have to say it is AI?

In most cases, yes. Since 2 August 2026, Article 50 of the EU Artificial Intelligence Act (AI Act) applies. Under it, an AI system that talks directly with people must be designed so that they know they are dealing with AI, unless this is obvious. In its guidelines, the European Commission names chatbots on online platforms and in customer support as an example where this is not obvious. So a chatbot on your website, in Messenger or in Viber that answers customers will usually need to say clearly that it is AI.

This article is for owners and managers who already run a chatbot or voice assistant, or plan to. It summarises the official text of Article 50 and the Commission's FAQ and transparency guidelines as of November 2026. It is a practical overview, not legal advice. In sensitive sectors, or if your role is unclear, talk to a lawyer.

What the simplification package changed and what it did not

In 2026 the EU adopted Regulation (EU) 2026/1744, known as the Digital Omnibus on AI. It postponed the rules for high-risk systems to 2 December 2027 and 2 August 2028. For transparency, the delay is narrow. According to the Commission's FAQ, a grace period until 2 December 2026 applies only to machine-readable marking of generated content, and only for systems placed on the market before 2 August 2026. There is no such period for telling people they are talking to AI. That obligation also covers chatbots that were already running on your site before that date.

Who is responsible for the notice

Article 50(1) is an obligation of the provider, meaning whoever develops the system, or has it developed, and puts it into service under their own name. A company that only uses the system is a deployer. The guidelines give examples that help you work out your role.

Your situationRole according to the guidelinesWhat to do
The bot was built for you and runs on your site under your nameA company that developed an interactive system for its own use under its own name is given as an example of a providerThe notice is your responsibility. Put it in the brief and in project acceptance
You use a vendor's ready-made chat product without changesThe vendor of the product is the provider under the ActCheck that its notice stays visible after you change the name, avatar and greeting
You took an existing system, modified it, for example with new training data, and run it under your nameYou can become the provider of the new systemTreat the notice as your obligation

The line is not always clear for heavily customised or white-label solutions. Write down, for each bot, which role you believe you hold and why. Even when someone else is the provider, visitors see your website, so it makes sense to check for yourself what the chat shows.

When "it's obvious" is not enough

The exception for obvious cases is read narrowly. The guidelines say that people's general awareness that chatbots exist does not mean they will recognise one in a specific conversation. Some features make it less obvious:

  • a human name and a photo of a person as the bot's avatar;
  • a natural-sounding voice in a voice assistant;
  • a personalised, conversational style that is hard to tell apart from a staff reply.

If the bot is called "Maria" and shows a photo of a smiling employee, it will be hard to argue that it is obvious a machine is answering. The exception may apply, for example, to tools meant only for trained specialists who know they are working with AI.

What the notice should look like

The Act does not prescribe exact wording. The information must be clear, distinguishable and accessible, and it must be given at the latest at the first interaction. The guidelines treat starting a conversation with the chatbot as the first interaction and say that a single prominent notice before it is likely to be enough in most cases.

WorksNot enough on its own
A first message such as "Hi, I'm an AI assistant, not a member of staff."Text only in the terms and conditions, privacy policy or a separate page
A short label next to the input field, for example "Answered by an AI system"Metadata or watermarks people cannot see
A persistent "AI" badge in the chat windowA vague name such as just "assistant"
For a voice assistant, a sentence at the start such as "This is an AI-powered assistant."A general phrase like "Services on this website use AI" or just "uses LLMs"

Three more rules from the guidelines matter for businesses:

  • Tell the truth when asked. If someone asks whether they are talking to a person, or seems confused, the bot must say clearly that it is AI. Cover this in the system instructions and test it.
  • Remind people in higher-risk conversations. For financial advice, insurance, legal help, health and complaints, and with children, older people and people with disabilities, a repeated reminder is likely to be needed, not just the first message.
  • Mixed conversations. If both the bot and a staff member reply in the same chat, AI replies must be marked unless a staff member reviewed them and sent them as their own. It helps to make the handover to a person visible.

If an AI agent acts on your behalf, for example booking appointments or emailing customers, the guidelines expect it to say both that it is AI and on whose behalf it acts.

What falls outside this obligation

Article 50(1) covers only direct conversation between an AI system and a person. According to the guidelines, it does not cover, for example, recommender systems, spam filters, automatic translation and transcription, search that does not generate content, or back-end systems where a person only sees the output without interacting with the system. If your staff use AI to write replies faster but send them themselves, that is not direct interaction between AI and the customer.

Data protection obligations remain separate from the Act. Chat conversations often contain personal data, and the GDPR applies to them.

Who enforces it and what the penalties are

Compliance will mainly be supervised by national market surveillance authorities. Fines for breaches of Article 50 can reach 15 million euros or 3% of worldwide turnover for the previous year, and proportionality may be taken into account for small and medium-sized enterprises. These maximums do not tell you what the actual penalty would be in a given case, but they show that the obligation is not a mere recommendation.

Checklist for your website chatbot

  • You know which bots and voice assistants run on your website, on social media and on the phone.
  • For each one, you have written down whether you are the provider or the deployer, and why.
  • The first message says clearly, in the visitor's language, that an AI is answering.
  • There is a visible label next to the input field or in the chat header.
  • The bot has no misleading human name and photo, or the notice clearly offsets them.
  • The bot answers "Are you a person?" honestly, and this has been tested.
  • Sensitive topics trigger a repeated reminder and an easy route to a staff member.
  • The handover to a staff member is visible in the conversation.
  • The notice is visible on mobile screens and readable by screen readers.
  • Someone is responsible for checking the settings after every change by the vendor.

A hypothetical example

This example is invented and is for illustration only. A dental clinic has a chat on its website called "Dr Anna", with a photo of an employee. The bot answers questions about prices, opening hours and booking, and sometimes about pain after a procedure. With this setup, the obviousness exception would be hard to rely on. The clinic renames the bot "The clinic's AI assistant", adds a first message saying that an AI is answering and a label next to the input field, and sets a rule that when symptoms come up, the bot reminds people it is AI and suggests calling the clinic.

Next step

When the bot is connected to a CRM or helpdesk, the notice, the handover to staff and the data rules are easiest to build in from the brief. In business process automation, Hostado maps out the steps, the systems and what happens when something goes wrong before building starts.