This guide is a general technical overview. It is not legal advice and not a legal review within the meaning of the German Legal Services Act (RDG). For an authoritative legal assessment, we recommend having a lawyer review your individual case.
If your website uses a chatbot or a voice widget that talks directly to visitors, a new obligation applies from 2 August 2026: users must be able to clearly recognise that they are dealing with an AI. This obligation arises from Art. 50 EU AI Act and affects not only large corporations but every company with an AI-powered chat or voice widget on its own site – from the trades business to the law firm.
Note: This guide is a general technical overview. It is not legal advice and not a binding legal assessment within the meaning of the German Legal Services Act (Rechtsdienstleistungsgesetz, RDG). For a binding legal assessment, we recommend having a lawyer review your individual case.
Art. 50 is part of Regulation (EU) 2024/1689, better known as the EU AI Act. It sets out so-called transparency obligations – that is, rules meant to ensure that people know when they are interacting with an AI system instead of a human. In substance this is something different from the risk classification that is otherwise often at the centre of the discussion on the AI Act: the point here is not whether an AI system is classified as "high-risk", but simply that users are not deceived when they speak to a machine instead of a human. This distinction matters, because many website operators wrongly assume that transparency obligations only affect "large" or "critical" AI applications. The opposite is true: precisely because chatbots and voice widgets have become so commonplace, the legislator created a separate, broadly applicable rule here.
Specifically, Art. 50(1) requires providers and deployers of AI systems that interact directly with natural persons to ensure that these persons are informed that they are interacting with an AI system. For a website chatbot, this means at its core: before or while a visitor chats with the bot, it must be unmistakably clear that they are dealing with an AI – not with an employee. This applies regardless of how well the bot words things, how natural it sounds or how closely it resembles a human conversation partner. The more convincingly a system imitates human communication, the more important the disclosure becomes – because in exactly these cases the risk of confusion for users is greatest.
Three properties of the disclosure are particularly important when you look at the purpose of the provision:
Clear. The notice must not be implicit or indirect. A chat widget that simply carries a name like "Lisa" and gives no indication of AI does not meet the requirement – even if technically savvy users might guess. The disclosure must be explicit and must not rely on users "reading between the lines" or inferring from the context that they are speaking to a machine.
Distinguishable. The notice must be designed so that an average user can tell it apart from other content on the page and notices it before or while interacting with the system. A single half-sentence somewhere on the page that is easily overlooked will hardly meet this requirement in practice. The disclosure should stand out visually or structurally from normal chat content – for example through a fixed position, its own formatting or permanent visibility throughout the whole conversation.
Accessible. The information must be available to the user at the moment they interact with the system – not only on request, not hidden in a document that has to be opened separately, and not findable only for users with particular technical background. Accessibility also means that the disclosure is made in the language in which the rest of the communication takes place, and that it becomes visible without additional clicks.
Important: what matters first and foremost is an honest disclosure that users can perceive at the place where the AI interaction actually takes place – that is, directly in or at the chat or voice widget – not an elaborate technical procedure. For most website operators this is a question of wording and placement, not of extensive software development.
Another point that often causes confusion in practice: Art. 50 does not distinguish by whether a chatbot handles "simple" or "complex" tasks. Whether your widget only answers basic questions about opening hours or simulates complete consultations makes no difference to whether the disclosure obligation applies. The only decisive factor is that an AI system communicates live with a natural person. Whether the system is based on a large, well-known language model or on a smaller, specialised AI solution also does not change the basic obligation.
The obligation is not tied to company size, sector or turnover, but simply to whether an AI system is operated that interacts directly with people. This potentially affects every SME with a chat or voice widget on its website – regardless of whether this widget was developed in-house, purchased or embedded via a third-party provider. Whether the widget is free or paid, self-hosted or embedded via a SaaS provider also makes no difference to the basic applicability: the only decisive factor is that an AI system interacts live with visitors on the website.
In practice we see this in very different businesses:
Trades businesses. Many trades companies now use chat widgets to take appointment requests, first follow-up questions about services or quote requests around the clock – especially outside business hours, when no employee is available. If this chat answers with AI support, for example by independently answering questions about the scope of services, price ranges or availability, Art. 50 applies. This equally concerns businesses that use an AI chatbot to pre-sort incoming enquiries before they are passed on to an employee.
Law firms. Law and tax firms increasingly use chatbots for an initial assessment of client enquiries or for scheduling appointments. Precisely here, where trust and credibility count particularly, a clear AI disclosure is important – even independently of the statutory obligation, because clients who are actually speaking to an AI in what seems to be a personal initial assessment may feel deceived once they find out later.
Real estate agents. Voice or chat widgets that guide prospects to viewing appointments, property details or financing questions are widespread in the real estate business and also fall under the obligation as soon as they work with AI support. Automated answering machines or voice bots that guide prospects through first questions over the phone are also relevant here.
Medical practices. Appointment-booking chatbots or voice assistants on the phone that pre-qualify patient enquiries are a growing trend – and an area in which users should know especially clearly whether they are speaking to a human or an AI. In a healthcare context in particular, trust is a central factor, which is why transparent disclosure makes double sense here.
Service providers in general. From gyms to cleaning companies to consulting firms: wherever a chat or voice widget on the website answers customer enquiries, the question is relevant. This also applies to online shops with customer-service chatbots, to education providers with AI-powered advisory widgets and to associations or service providers that use an AI tool for member communication.
The common denominator is not the sector but the technical fact: as soon as an AI system on your website interacts directly with visitors – by text or voice – you, as the deployer, are obliged to make this recognisable. A plain contact form that sends an email to your team does not usually fall under this, because no AI-generated interaction with the user takes place. However, as soon as a system answers independently with AI support – even if only with short, automated follow-up questions – the threshold to direct interaction is, as a rule, exceeded.
Legally, a look at the allocation of roles in the AI Act is worthwhile at this point: the obligations under Art. 50(1) are addressed primarily to providers of AI systems – that is, those who develop an AI system or place it on the market under their own name. Most SMEs that merely embed a ready-made chat widget from a third party unchanged, by contrast, are typically deployers within the meaning of the regulation – in detail, other and in part leaner obligations can apply to them than to the provider of the widget itself. Where exactly the line falls in an individual case depends, among other things, on whether the widget is used unchanged or is adapted, renamed or redistributed under your own brand. A blanket statement on which party bears which specific obligations cannot responsibly be made at this point – it depends on the individual set-up and, in case of doubt, should be clarified by a lawyer.
Company size also plays no great role in the basic question of whether you are affected. A sole trader with a simple website and a third-party chat widget embedded generally operates within the same legal framework as a medium-sized company with its own development team – even if, as described above, the exact classification as provider or deployer and the resulting scope of obligations can differ in detail. For the website visitor it makes no difference anyway how large the company behind it is – what matters to them is solely whether it is recognisable who or what they are currently communicating with.
The transparency obligations under Art. 50 EU AI Act apply from 2 August 2026. This means: from that date, every affected AI system that interacts directly with users must meet the requirements for clear, distinguishable and accessible disclosure.
For website operators with chat or voice widgets this means in practice: if you already use an AI-powered widget today, you should review the disclosure now and – if necessary – add it, rather than waiting until shortly before the deadline. The technical implementation itself is usually manageable (a visible notice in or next to the widget); the time effort lies more in systematically reviewing your own website and finding the right wording. If you run several domains, landing pages or subpages, you should also remember that every place where a chat or voice widget is embedded must be checked – not only the main page.
If you are currently planning to introduce a new chat or voice widget, you should think about the disclosure obligation from the start – then the deadline is not a problem but merely an item already ticked off the checklist. This applies in particular to companies that are currently talking to agencies or providers about introducing an AI chatbot: the disclosure question is easiest to clarify directly at the concept stage, rather than planning it afterwards as an additional adjustment step.
Until the deadline there is therefore still time to review your website in peace – what matters is only that this review actually takes place and does not get lost in general website maintenance.
Art. 50 does not prescribe an exact wording. What matters is that the disclosure is clear, distinguishable and accessible to the user before or while they interact with the system. In practice this means: a visible, understandable notice directly at or in the widget – not just hidden somewhere on the website. There is therefore a certain design leeway, as long as the basic requirements are met.
Examples of sufficient disclosure:
"Hello! I am an AI-powered chat assistant and help you with first questions. For personal matters, you can reach our team at [contact]." – This notice appears directly when the chat window is opened, before the user writes a message, and makes immediately clear whom the user is communicating with.
"You are currently speaking to our AI chatbot. A human from our team will get back to you personally if needed." – As a fixed header in the chat widget that stays visible throughout the whole interaction, so that the notice is still present even when the user has already forgotten the introductory message.
"This voice assistant uses artificial intelligence to process your request." – As a spoken or written notice at the start of a voice-widget contact, before the user makes their actual request.
What these examples have in common: the AI notice comes first or is permanently visible, it is worded in normal, understandable language and it is located exactly where the interaction takes place – not on a separate subpage that the user would first have to actively visit.
Examples of insufficient disclosure:
A chat widget labelled only with a first name such as "Anna", with no further indication of AI – here the user cannot tell whether they are writing with a human or an AI, even if the name alone might be a vague indication for some users. A vague indication is no substitute for clear disclosure.
An AI notice that appears only in the legal notice (imprint) or in the terms and conditions of the website, but not at the chat widget itself – the notice exists somewhere, but is not accessible where the interaction takes place, and is never seen by most users. Such a notice may exist from the point of view of a legal "evidence document", but it does not fulfil the actual purpose of the provision: actual perceptibility for the user at the moment of interaction.
Such solutions may look like a disclosure at first glance, but they do not meet the requirement of clarity and accessibility at the moment of interaction. If you are unsure whether your own implementation is sufficient, you should always ask yourself the simple question: would an average, inattentive website visitor actually notice the notice before they start chatting?
Violations of Art. 50 fall under Art. 99 AI Act and can be punished with fines of up to 15 million euros or 3% of worldwide annual turnover – whichever is higher. For SMEs, the lower of the two amounts applies.
These fine frameworks are deliberately designed so that they can be effective for large corporations as well as for smaller companies – the distinction between the "higher" and the "lower" amount depending on company size is meant to prevent a small trades business from being confronted with the same absolute sums as an international software group. Nevertheless, it should be noted that Art. 99 is a fine framework of serious magnitude, which shows how seriously the legislator takes the transparency obligations under Art. 50.
Important for putting it in context: whether and in what amount a fine is actually imposed in an individual case depends on the circumstances of that case. This guide deliberately makes no further statements on this, so as not to suggest false certainty and not to claim any details that go beyond the established set of rules.
Besides the purely financial risk, the practical side is also worth a look: missing or inadequate disclosure can additionally damage customers' trust if they later find out that they unknowingly communicated with an AI. In trust-sensitive sectors such as law, healthcare or real estate, this can weigh more heavily in terms of reputation than the pure fine risk. The framework values mentioned, however, already clearly show on their own: missing disclosure is not a trivial matter that one should put off, but an item that should be on the compliance checklist of every company with an AI chat or voice widget.
For most SMEs, the effort to make a website compliant is manageable – especially compared with the possible fine framework. Missing disclosure can usually be added within a few hours once it is clear where exactly the notice is missing. The real effort therefore lies less in the implementation itself than in recognising at all where on your own website action is needed – and this is exactly why a systematic review is worthwhile before the deadline draws closer.
The following five questions help you assess yourself in a few minutes whether your website falls under the disclosure obligation under Art. 50. The more questions you answer with "yes", the more likely it is that you should act now.
If you answer questions 1 to 3 with "yes", you are in principle operating an AI system that falls under Art. 50. If question 4 or 5 also applies, there is an urgent need for action: your current implementation presumably does not yet meet the disclosure obligation.
Should you be unsure about one or more of these questions – for example because you do not know exactly whether your chat provider uses generative AI or fixed rule sets – it is worth looking at the technical documentation of the respective widget provider or briefly asking your agency or service provider. Many chat and voice widget providers have by now published information themselves on whether and how their systems use generative AI.
Free AI Act disclosure check
Check for free nowThe technical implementation of correct disclosure is straightforward in most cases and requires no major development work. What matters is where and how the notice is placed – not how elaborately it is designed.
Placement: The AI notice belongs directly in the chat or voice widget itself – ideally as the first message the user sees before they enter anything at all, or as a permanently visible disclosure (e.g. in the header of the chat window or directly next to the bot's name). A notice that only flashes up briefly when opening and then disappears is less robust than one that stays visible throughout the whole interaction. For voice widgets, a spoken announcement at the start of the conversation is also recommended, since users cannot read a visual notice here.
Wording: The language should be simple and unmistakable. Terms such as "AI", "artificial intelligence", "chatbot" or "AI assistant" are clearly understandable for most users and work better than vague paraphrases such as "virtual helper" without further addition. Avoid marketing language that could obscure the AI nature, even if this happens unintentionally – for example wording that strongly suggests you are speaking to a real person.
Technical integration: If you use a third-party provider for your chat or voice widget, it is worth looking at its configuration options – many providers allow you to customise a welcome message or a fixed header text, so that the disclosure notice can be built in without any programming effort. For self-developed widgets, the addition is usually a small, clearly delimited change to the chat interface. If you are setting up or revising your AI chatbot widget anyway, the disclosure can be planned in from the start instead of being retrofitted later.
Two ready-to-use text modules that you can adapt directly for your widget:
For a chat widget opening: "Hello, I am the AI chat assistant of [company name]. I answer first questions automatically. For a personal conversation, you can reach our team at [phone number/email]."
For a permanent disclosure in the chat header or as a subtitle: "AI chatbot – automated answers, no human in the chat."
Both variants are worded so that they take into account the three central requirements: they are clear (the term "AI" or "AI chatbot" is explicitly named), distinguishable (they stand out as a separate notice instead of getting lost in running text) and accessible (they appear exactly where and when the interaction takes place). Whether they are actually sufficient in an individual case additionally depends on their being placed correctly and not covered up or devalued by other design elements – the final assessment remains a matter of the individual case.
If you use a voice widget, the same principle applies to the spoken or displayed announcement: the AI notice should be given at the start of the conversation, not only on request. A simple sentence such as "You are speaking to an AI voice assistant" at the start of the announcement is a good starting point for meeting the requirement of clarity and accessibility. Whether this specific wording is actually sufficient in an individual case additionally depends on factors such as its placement in the announcement and whether callers actually hear and understand it – here too, the final assessment remains a matter of the individual case.
In implementing the disclosure obligation we keep seeing the same avoidable mistakes in practice:
Disclosure only in the legal notice (imprint) instead of in the widget itself. Many companies assume that a notice in the imprint or the terms and conditions is enough because legally required information is given there anyway. However, this does not meet the requirement of accessibility at the moment of interaction – very few users read the imprint before starting a chat, which is why the notice remains ineffective in practice.
Notices that are too small or inconspicuous. A tiny notice in light grey type at the bottom of the chat window may be technically present, but is simply overlooked by most users. This contradicts the requirement that the disclosure must be distinguishable – a notice that practically nobody reads does not fulfil its purpose, even if it formally exists.
Missing disclosure for voice widgets. While many companies now think of a disclosure for text chatbots, it is often forgotten for voice assistants on the phone or on the website – although the obligation applies here just the same. Especially for voice systems that sound increasingly natural, a clear announcement at the start of the conversation is particularly important.
Disclosure that disappears after a short time or on the next visit. Some widgets do show the AI notice at the start, but fade it out automatically after a few seconds, hide it behind a click or remember via cookie that a user has already seen it, and then do not show it again on the next visit. If a user later continues the conversation, reopens the chat window or visits the page on another device, the notice may then be missing entirely. A permanently visible disclosure, for example in the header, that is always shown regardless of visitor status reliably avoids this problem.
Anyone who avoids these mistakes and places the notice where the interaction actually takes place is already a big step ahead of many websites that are currently still live. The good news: none of these mistakes requires a fundamental redevelopment of the chat or voice widget – usually a targeted adjustment of the welcome message, the header or the voice announcement is enough to get from inadequate to clear, distinguishable and accessible disclosure.
Don't want to guess whether your chat or voice widget is affected and whether your current disclosure is sufficient? With Softstorm's free AI Act check you get an automated initial assessment in about 20 seconds: the scan detects common chat and voice widgets on your website and checks whether a recognisable AI disclosure is present.
Free AI Act disclosure check
Check for free nowThis guide is a general technical overview. It is not legal advice and not a legal review within the meaning of the German Legal Services Act (RDG). For an authoritative legal assessment, we recommend having a lawyer review your individual case.
We analyze your processes, identify automation potential and show you which digital systems really make sense.
