AI chatbot disclosure is no longer just a question of customer trust. For businesses operating in the EU, it is also a legal consideration. If your website uses artificial intelligence to answer enquiries, recommend products or speak to customers, should people be told they’re dealing with AI rather than a human?
The short answer is yes in many situations, although the precise legal requirement depends on where the system is used, what it does and whether your business is providing or deploying it.
Since 2 August 2026, the EU AI Act has required providers of certain interactive AI systems to ensure that people are informed when they’re communicating directly with artificial intelligence, unless that fact is already obvious. In the UK, the legal framework is different, but businesses still have responsibilities concerning transparency, personal information and potentially misleading practices.
For businesses introducing AI customer service, the more interesting question isn’t whether a disclosure will make the technology appear less impressive. It’s why a customer should have to guess who—or what—is answering them.
AI Chatbot Disclosure Under the EU AI Act
The European Union has introduced specific transparency requirements for AI systems that interact directly with people.
Article 50 of the EU AI Act requires providers to design and develop relevant AI systems so people are informed they’re interacting with artificial intelligence.
These requirements became applicable on 2 August 2026. The European Commission also published guidelines on AI transparency obligations in July 2026 to clarify how the rules should be interpreted.
For interactive AI systems covered by Article 50(1), the information must be clear, distinguishable and accessible, and provided no later than the first interaction.
Consider a business using an AI assistant to answer customer questions about orders, bookings or services. If the system is designed to hold a direct conversation with the customer, the EU disclosure requirement is likely to be relevant.
Importantly, the rule doesn’t only concern systems that use large language models. Whether a particular system is covered depends on whether it meets the AI Act’s definition of an AI system and the conditions for direct interaction.
Does Every Automated Chatbot Need an AI Disclosure?
Not necessarily under this specific provision.
The European Commission’s Article 50 guidance distinguishes AI systems designed for genuine two-way exchanges from systems that merely collect data or provide predefined automated responses.
A basic contact form that asks a series of fixed questions is not automatically the same as an AI assistant that interprets questions and generates responses.
Similarly, AI used internally to help an employee prepare a reply is different from a system that communicates directly with customers.
These distinctions matter when assessing legal obligations. However, clearly identifying an automated interaction can still be sensible even when the specific AI chatbot disclosure requirement doesn’t apply.
What If Customers Can Obviously See It’s AI?
Article 50 includes an exception where it is already obvious to a reasonably well-informed and observant person, considering the circumstances, that they’re interacting with an AI system.
That doesn’t mean businesses should automatically assume customers will recognise the technology.
An assistant with a realistic name, profile picture and conversational writing style might appear human, particularly when embedded in a website’s ordinary customer support interface.
The Commission’s guidance indicates that the obviousness exception should be interpreted restrictively.
A visible label is a relatively straightforward way to avoid relying on customers to work out how the system operates.
Does AI Chatbot Disclosure Apply to UK Businesses?
This is where the distinction between UK and EU law becomes important.
The EU AI Act does not automatically apply to every AI chatbot operating solely within the United Kingdom. The UK has its own legal framework and does not have an identical general chatbot-disclosure provision equivalent to Article 50.
However, UK businesses should not confuse that distinction with having no transparency obligations.
Under UK data protection law, businesses processing personal information must comply with requirements concerning lawfulness, fairness and transparency.
The Information Commissioner’s Office (ICO) explains in its guidance on transparency in AI that organisations need to explain how personal data is processed, including relevant purposes, retention periods and information sharing.
This becomes especially important when an AI assistant collects information such as:
- Customer names and contact details.
- Order numbers and purchase histories.
- Appointment information.
- Account-related questions.
- Complaints or other personal circumstances.
A chatbot that collects personal information needs appropriate privacy information regardless of whether a specific requirement to identify the chatbot as AI applies.
And if the chatbot makes or contributes to decisions with significant consequences for individuals, additional data protection safeguards may become relevant.
UK privacy transparency and EU AI chatbot disclosure are related issues, but they are not interchangeable legal requirements.
What If a UK Business Serves Customers in the EU?
Being based in the UK doesn’t necessarily place a company outside the EU AI Act.
Under Article 2 of the EU AI Act, the legislation can apply to providers and deployers established outside the EU where the output of an AI system is used within the Union. Other provisions address systems placed on the EU market and deployers established in the EU.
For example, a UK business deploying an AI customer service system for its operations in an EU country should assess whether the legislation applies to that arrangement.
Simply hosting a website in the UK is not enough to determine the answer. The relevant activities, role of the business and location where the AI output is used all matter.
Businesses operating across borders should assess their actual deployment rather than rely solely on the location of their registered office.
Who Is Responsible for Telling Customers They’re Talking to AI?
There is an important legal distinction between the company that provides an AI system and the business that uses it.
Under the EU AI Act:
- Providers develop an AI system, or have it developed, and place it on the market or put it into service under their own name or trademark.
- Deployers use an AI system under their authority in a professional context.
The direct-interaction disclosure obligation in Article 50(1) is specifically placed on providers. Other parts of Article 50 impose different transparency duties on deployers, including certain uses of deepfakes and biometric technologies.
A business that commissions or develops its own customer-facing AI system under its own brand may also qualify as a provider, depending on the arrangement.
For companies buying an existing chatbot service, the practical questions are straightforward:
- Does the provider support clear AI identification?
- Is the disclosure visible before or at the beginning of the conversation?
- Does the notice remain clear when the chatbot is embedded in a branded website?
- Can the business customise the introduction without concealing the AI identity?
- Who is responsible for reviewing applicable compliance requirements?
Even where a particular legal duty falls on the technology provider, a business still controls much of the customer experience.
Choosing a chatbot that identifies itself clearly is simpler than introducing uncertainty about whether customers are being misled.
What Should an AI Chatbot Disclosure Actually Say?
A useful AI chatbot disclosure doesn’t need to be a long legal warning. For most ordinary customer service interactions, the immediate aim is to make the nature of the conversation clear.
Consider this example:
Hi, I’m the AI assistant for [Business Name]. I can help answer questions about our services and guide you to the right information. If you need further assistance, you can contact our team.
The wording identifies the assistant, explains its intended purpose and points towards human assistance.
The specific phrasing should reflect what the system can actually do. Don’t promise live handover, booking capabilities or human availability unless those options exist.
There are several practical design decisions worth considering.
Identify the Assistant Before the Conversation Begins
Don’t depend on customers asking whether they’re speaking to a human.
A visible AI label in the chat heading, together with a clear opening introduction, is a practical approach. For a voice assistant, an audible introduction may be more appropriate.
Placing the disclosure only in a lengthy privacy policy is a poor substitute for informing someone at the point of interaction.
Avoid Creating a Deliberately Misleading Human Identity
Giving an assistant a name or a conversational personality isn’t inherently a problem.
But there’s a difference between making technology approachable and presenting it as a human employee.
An assistant called Sarah, accompanied by a realistic profile photograph and described as a member of the customer support team, could create the impression that a person is responding.
If the service is AI-powered, identify it as such.
Customers can still appreciate a friendly conversation without being encouraged to misunderstand who they’re speaking to.
Explain What Happens to Customer Information
The AI identity notice and privacy information serve different purposes.
Customers may reasonably want to know whether their messages are stored, used for service improvement, shared with third-party technology providers or accessible to employees.
Businesses should establish these details from their actual technical setup and relevant supplier agreements.
Don’t simply state that conversations are private, deleted automatically or never used to train models without verifying those claims.
The ICO’s guidance on lawful, fair and transparent processing explains why organisations need to be open about their use of personal data.
Does Telling Customers It’s AI Mean You Must Offer a Human?
Not automatically under the general chatbot identification requirement in Article 50(1).
The obligation to identify an AI system should not be confused with a universal legal requirement for every chatbot to offer immediate human handover.
Nevertheless, human escalation is often a sensible design choice.
An AI assistant might be well suited to answering opening hours, explaining services or locating published information. It may be less appropriate for resolving a disputed payment, handling a serious complaint or interpreting complex personal circumstances without additional safeguards.
The important question is not whether the assistant can produce a response. It’s whether it should be responsible for that response or the action that follows.
For an AI assistant connected to business systems, this also raises questions about permissions.
Should it be able to retrieve a customer’s information? Change an appointment? Issue a refund? Send a binding quotation?
Those decisions require controls beyond a disclosure notice. We’ve explored the distinction between reading information, preparing actions and carrying them out in AI Agent Permissions: What Should Your Agent Be Allowed to Do?
Transparency tells customers what they’re interacting with. Appropriate system controls help determine what that interaction is allowed to accomplish.
AI Chatbot Disclosure Checklist for Business Websites
If your business already uses an AI assistant, or is preparing to introduce one, the following checks provide a practical starting point.
- Establish whether the system qualifies as AI. Don’t assume every automated contact flow is an AI system under the relevant legislation.
- Check where it operates. Determine whether the chatbot is used in the UK, the EU or across multiple jurisdictions.
- Identify your role. Establish whether the business is a provider, a deployer or potentially both.
- Review the initial disclosure. Make sure customers can tell they’re speaking to AI without having to investigate.
- Check accessibility. The notice should be readable, understandable and accessible on relevant devices and interfaces.
- Review personal data handling. Confirm what information is collected, why it’s needed, where it goes and how long it’s retained.
- Define the assistant’s limits. Decide which subjects it can handle and what should be referred to a person.
- Test the customer experience. Check whether a typical user could mistake the assistant for a human or misunderstand its authority.
For businesses with more complex AI systems, this checklist is only a starting point. Regulated industries, sensitive information and consequential automated decisions can introduce additional obligations.
The goal is to ensure that AI chatbot disclosure is part of the product design rather than an afterthought added once the assistant is live.
Customer Trust Shouldn’t Depend on Whether Someone Notices the AI
At Nort Labs, customer-facing artificial intelligence is particularly relevant to our work on AI development and business assistants, including Lucy.
We see an important distinction between making an assistant communicate naturally and encouraging customers to believe they’re speaking to a person.
Natural language is useful because it makes software easier to interact with. It shouldn’t require hiding the technology behind the conversation.
Clear identification also helps establish realistic expectations. An AI assistant may be able to provide immediate information, but customers shouldn’t automatically assume that every answer has been checked by an employee or that every proposed action has been authorised.
The same principle applies to systems that become more capable over time. Adding access to company knowledge, customer records or operational tools increases the importance of clear boundaries.
We’ve previously discussed these kinds of deployments in AI Chatbots Tailored for Niche Industries. The technical capabilities of an assistant matter, but so does the way it is presented to the people using it.
Should You Disclose AI Even When the Law Doesn’t Explicitly Require It?
For a customer-facing AI assistant, our recommendation is generally yes.
Where Article 50 of the EU AI Act applies, the relevant legal requirements need to be met. Where it doesn’t, clear identification is still a sensible default.
There is little practical benefit in allowing customers to incorrectly assume an automated system is a human employee. It can create confusion about the assistant’s authority, capabilities and the way information is handled.
An AI chatbot disclosure doesn’t have to interrupt the conversation or undermine the usefulness of the service. It can be as straightforward as an accurate label and a clear opening message.
Customers don’t need every technical detail about the model behind the chatbot. They do need an honest understanding of the interaction they’re entering.
Businesses should design that transparency into the experience from the beginning, then ensure the assistant’s permissions, privacy arrangements and escalation process match what customers are told.
