aiReceptionistDoes your dating chat need to disclose that it's an AI? What the AI Act requires starting August 2026
Team at aiReceptionistWritten with AI assistance19 min readIn summary
- Yes, your dating chat must inform that it is an AI. Article 50 of the European Artificial Intelligence Regulation requires informing the person that they are interacting with an AI system, clearly and, at the latest, during the first interaction (text of Article 50).
- It is enforceable from August 2, 2026. This is not a future deadline: it has already passed.
- The Digital Omnibus did not postpone this obligation. Regulation (EU) 2026/1744, in effect since July 27, 2026, postponed high-risk systems from Annex III to December 2, 2027, but left the transparency of Article 50 intact (calendar after the Omnibus).
- Applies with 3 employees just like with 300, with fines of up to 15 million euros or 3% of global turnover; for SMEs and startups, the lower of the two amounts applies (analysis of the Article 50 deadline).
- GDPR still applies in parallel: if that chat asks for name, phone, email, or the reason for the inquiry, you have personal data processing on top of the AI notice.
If you have a dental clinic, a physiotherapy center, a consulting firm, a hair salon, or a workshop, and on your website or Instagram there is a chat that attends to customers and schedules appointments, this article is for you. We wrote it from AIsToHire, where we develop aiReceptionist, an AI appointment booking and customer service chat for businesses that rely on scheduling; precisely for this reason, we had to study this regulation line by line. What follows is the result, without legal jargon: what it exactly requires, from when, how much it could cost you, and a seven-point checklist that you can review this very afternoon.
What does Article 50 of the European AI Regulation exactly require?
Article 50.1 requires that AI systems intended to interact directly with people be designed so that those people know they are talking to an AI, unless it is evident to someone "reasonably informed, attentive, and perceptive" considering the circumstances and context of use. Article 50.5 adds the when and the how: the information must be provided "in a clear and distinguishable manner, at the latest at the time of the first interaction or exposure," and respecting applicable accessibility requirements (Article 50 of the AI Act).
Translated to your business, that means three things:
- The notice comes at the beginning, not when the customer has already described their tooth problem.
- It must be visible, not a paragraph in the privacy policy or a line in the terms of use.
- The exception of "it is evident" almost never saves you. A chat that greets by name, writes naturally, and is capable of suggesting available time slots is not "obviously" a machine to a 70-year-old patient entering from their mobile. If you doubt whether it is obvious, it is not: inform.
Article 50 has other sections that may also affect you even if you don't have a chat: marking synthetic content you generate (texts, images, audio) in machine-readable format, informing if you use emotion recognition or biometric categorization, and declaring deepfakes. For an appointment business, the section that matters is 1, and the one that is advisable not to lose sight of is the one regarding generated content.
A practical recommendation about the language of the notice
Several analyses for SMEs emphasize a detail that is often overlooked: the notice must be fixed text, not improvised by the model, the assistant must respond directly if someone asks "are you a person?", and it must be issued in the user's language, avoiding ambiguous terms (obligations of chatbots and voicebots). A notice that only exists in Spanish does not comply if your chat responds in English or French.

The Digital Omnibus DID NOT postpone the AI notice: it only delayed high risk
This is the most widespread confusion of 2026 and it is worth clarifying: Regulation (EU) 2026/1744, known as the Digital AI Omnibus āpublished in the Official Journal on July 24, 2026, and in force since July 27, 2026ā, did not change the date for the transparency obligations of Article 50. What was postponed were the high risk systems from Annex III, which were moved to December 2, 2027, and those linked to regulated products from Annex I, to August 2, 2028 (obligations calendar after the Omnibus).
The only additional margin that the Omnibus granted regarding Article 50 is technical and very limited: providers of systems that were already in the market before August 2, 2026, have until December 2, 2026 to comply with the machine-readable marking of synthetic content. The duty to notify that the one responding is an AI has no extension (Article 50 deadline).
| Date | What applies | Does it affect you if you have a dating chat? |
|---|---|---|
| Feb 2, 2025 | Prohibited practices (art. 5) and AI literacy (art. 4) | Yes, basic team training |
| Aug 2, 2025 | General-purpose AI models and governance | Indirectly, via your provider |
| Aug 2, 2026 | Transparency of art. 50: notify that it is an AI | Yes, fully and from now on |
| Dec 2, 2026 | End of the transitional period for the technical marking of synthetic content | Yes, if you generate automatic texts or images |
| Dec 2, 2027 | High risk from Annex III (postponed by the Omnibus) | Normally no |
| Aug 2, 2028 | High risk from Annex I, regulated products | Normally no |
Source: consolidated calendar after the Regulation (EU) 2026/1744 and AI Act obligations for SMEs, consulted on November 11 and 12, 2026.
In other words: if someone in your sector has told you that "the entire AI Act has been postponed to 2027," they have given you incomplete information that leaves you exposed precisely in the part that does apply to you.
Does it apply to a small business? Yes, and fines reach 15 million
The Regulation does not set a threshold for staff or turnover for Article 50: it applies equally to a three-person hair salon as to a chain of 300 employees. What does change with size is the amount: non-compliance with general obligations such as transparency is sanctioned with up to 15 million euros or 3% of the annual global turnover, whichever is greater, but for SMEs and startups, the lesser of the two applies (Article 50 deadlines and penalties, obligations for SMEs).
| Type of non-compliance | General cap | For SMEs and startups |
|---|---|---|
| Prohibited practices (art. 5) | 35 M⬠or 7% of global turnover | The lesser of the two |
| Other obligations, including transparency of art. 50 | 15 M⬠or 3% of global turnover | The lesser of the two |
| False or misleading information to authorities | Specific lower tier | The lesser of the two |
Source: sanction regime of the AI Act according to the cited analyses, consulted on 11-11-2026. The penalties must also consider the "economic viability" of the sanctioned party.
For a hair salon, 3% of turnover is a perfectly real figure: it is not the fine of 15 million that should concern you, but the percentage.
And who monitors this in Spain?
In Spain, the market supervision body of the Regulation is the AESIA (Spanish Agency for the Supervision of Artificial Intelligence), based in A CoruƱa and created as the first national agency of its kind in the EU; it coordinates implementation with the AEPD regarding personal data and biometrics, with the CGPJ in the judicial sphere, and with sectoral supervisors (AESIA profile).
There is an important nuance of current Spanish affairs: the Organic Law Project for the proper use and governance of artificial intelligence (initiative 121/000096), which specifies in Spain the distribution of competencies and the details of the sanctioning procedure, expired with the end of the legislature; its profile in Congress records it as an expired initiative after the amendment period (profile of initiative 121/000096, consulted on 12-11-2026).
That does not exempt you from anything. The AI Act is a European Regulation: it applies directly in Spain without the need for a transposition law. The only thing that the expiration of the project leaves in the air is the national procedural detail, not the obligation. If you were waiting for "the Spanish law" to get started, you are three months behind.
Why does the GDPR still apply even if you comply with Article 50?
Because they are two distinct regulations that govern two different things: the AI Act regulates that you notify that there is an AI; the GDPR regulates what you do with the data that this AI collects. Complying with one does not exempt you from the other, and both authorities āAESIA and AEPDā can act cumulatively.
In a booking chat, this is very noticeable because the data you request is not innocuous:
- Name, phone number, and email are standard personal data: you need a legal basis under Article 6 of the GDPR and you must inform who processes the data and for what purpose.
- The reason for the inquiry may be health data. In a dental clinic, a physiotherapy center, or a psychology consultation, what the client writes in "why do you want the appointment?" āand sometimes the mere fact of having requested an appointment at that type of centerā falls into the special categories of Article 9 of the GDPR, with a much higher protection threshold.
- Your chat provider is a data processor. Even if they only handle name and phone number, they act on your behalf under the terms of Article 28 of the GDPR: you need a signed processing agreement before you start operating.
- No training models with your conversations. It is advisable that the contract with the provider expressly prohibits this (recommendations on chatbots and voicebots).
Add an ingredient from 2026: the AEPD published on February 18, 2026 its guidelines on agentic AI from the data protection perspective, where it recommends integrating these systems into the governance of processing, limiting access, minimizing data, controlling memory and logs, and documenting the level of autonomy of the agent (AEPD guidelines on agentic AI). If your chat not only responds but also acts āchecks your schedule and makes a reservationā, that document describes you.
7-Point Checklist for Your Dating Chat to Comply
This is the practical core of the article: seven specific checks you can do yourself, opening your own chat as if you were a client, without the need for a lawyer for the first pass.

1. The notice, in the header and in the initial greeting
Two places, not one. A permanent label in the chat header (like "AI Virtual Assistant"), visible all the time even if the client has sent twenty messages, and an explicit phrase in the first message from the assistant. Article 50.5 requires it to be "clear and distinguishable, at the latest in the first interaction": a notice buried in the terms of use or in the privacy policy does not meet that standard. And check the notice in production, from a mobile device: there are widgets that delay it or cut it off on small screens.
2. Honest identity: no pretending to be a person
The assistant can have a name, but it cannot imply that it is human. Use unambiguous terms ā"artificial intelligence", "AI system", "virtual assistant"ā and avoid vague formulas. The most important thing: if the client asks "Are you a person?", the assistant must respond no directly, always, without dodging the question. That response must be fixed text and not left to the model's discretion.
3. Exit to a human person, always available
The client must be able to exit the chat to a human at any time, and must know that they can. This is not literally required by Article 50, but it is the point where complaints concentrate and, in health consultations or claims, where you assume the most risk. Two things you need to have resolved:
- During business hours: the conversation passes to a team member, with the history visible so that nothing needs to be repeated.
- Outside of hours: the assistant must tell the truth āthat there is currently no oneā give a specific response time, and collect a means of contact. What is not acceptable is to simulate that "a colleague will assist you shortly" at two in the morning.

4. Notify what data you ask for and for what
Here comes the GDPR. Before or at the time of asking for a name, phone number, email, or reason for the consultation, the client must know who is responsible, for what those data are used (to manage their appointment and contact them), how long they are kept, and how to exercise their rights. And ask only for what you need: if for an appointment you only need a name and phone number, do not request date of birth or ID number. Minimization is not a recommendation; it is Article 5 of the GDPR.
5. Review reminders and automatically generated texts
Reminder emails, conversation summaries, and any text that the system writes on its own fall into the realm of AI-generated content. Two simple rules: it must be understood that they are automatic and not a person writing one by one, and they must not contain more data than necessary āfor example, do not send the reason for the consultation in the subject line of an email. Remember that the transitional period for the technical marking of synthetic content for systems already on the market ends on December 2, 2026 according to the post-Omnibus calendar.
6. Keep the history to be able to prove that you informed
An obligation to inform only serves if you can prove that you informed. Keep a record of conversations, the version of the notice text, and the date from which it has been active, and a screenshot for each language and channel. If you change the greeting tomorrow, document when. This is exactly the type of evidence that compliance analyses recommend for chatbots: system inventory, prompt versions, and proof of the notice by language.
7. Check the notice in all the languages you serve
This is the point that almost no one reviews. If your chat detects the client's language and responds in English, French, or German, the AI notice must exist in those languages as well. A translated banner but a greeting that remains in Spanish, or vice versa, is a non-compliance with the same penalty cap. Open your chat with the browser in another language and check it yourself, one by one. And the same goes for the data notice.
Three typical mistakes that are still out there
After August 2, 2026, these three patterns continue to be extremely common on clinic, workshop, and aesthetic center websites, and all three are avoidable in an afternoon.
The bot with a name and photo of a person. "Hello, Iām Laura š" with a portrait photo and not a mention of AI. This is the clearest case of non-compliance with Article 50.1: the design is made precisely to make the customer believe they are talking to a person, so the exception of "it is evident" works against you, not in your favor.
The "virtual assistant" that claims to be part of the team. A more subtle variant: the widget is called "virtual assistant" in the header, but when the customer asks if it is a person, the model improvises with "Iām here to help you from the clinic" or directly claims to be part of the team. The initial warning does not neutralize a subsequent false statement; the duty of transparency does not end with the first message.
The chat that leaves the customer trapped. No option to talk to a person, no visible phone, with the assistant responding the same in a loop. Technically, you may be complying with the AI notice and still generating the complaint that ends up in a claim āand on the AEPD's radar if you haven't informed about data processing in the meantime.
A fact that explains why this is so widespread: according to the Survey on the use of ICT and e-commerce in companies by INE, in the first quarter of 2026, 33.6% of Spanish companies with ten or more employees were already using AI technologies, and 21.4% of those with fewer than ten. Adoption has run much faster than knowledge of the regulation: thousands of small businesses implemented a chat before the obligation to notify existed, and they have never touched the greeting again.
How a booking chat fits that is born with the checklist done
The cheapest way is not to have to rearrange anything: choose a tool where the notice, escalation, and record come as standard instead of patching a generic widget. This is the approach we built aiReceptionist with at AIsToHire, and itās worth saying how it translates point by point, because itās exactly the checklist from above:
- It presents itself as a virtual assistant in the header and in the initial greeting, not in the terms of use (points 1 and 2 of the checklist).
- Detects the customer's language and gives the notice in the six languages it attends to, so that no language is left uncovered (point 7).
- Requests only the data that the business has configured āname, phone, email, the reason if you need itā and nothing more, which leaves you with minimization resolved by design (point 4).
- Leaves all the history recorded in an inbox like WhatsApp, which is your proof that the notice was given and when (point 6).
- Escalates to a team member as soon as the customer requests it or the conversation gets complicated, with a summary of what was discussed so you can follow up from your mobile, and you return it to the AI with a button when it's resolved (point 3).
And while it complies, it does the job for which you put it: it responds with the information of your business, checks the real availability of your schedule ācrossing your hours, your types of appointments, your blocks, and your advance rulesā and closes the booking within the conversation itself, without double bookings. You can see it in more detail on the page of aiReceptionist, the AI appointment chat for businesses.
Frequently Asked Questions
Is it mandatory to inform that a chatbot is an AI in Spain?
Yes. Article 50 of the European Artificial Intelligence Regulation is directly applicable in Spain since August 2, 2026, and requires clear information, at the latest during the first interaction, that the person is interacting with an AI system, unless it is evident to someone reasonably informed and attentive.
Wasn't the entire AI Act delayed with the Digital Omnibus?
No. Regulation (EU) 2026/1744, in force since July 27, 2026, postponed high-risk systems in Annex III to December 2, 2027, and those in Annex I to August 2, 2028, but does not delay the transparency obligations of Article 50, which remain enforceable since August 2, 2026.
How much can I be fined if my chat does not inform that it is an AI?
Failure to comply with the general obligations of the Regulation, including the transparency of Article 50, is punishable by fines of up to 15 million euros or 3% of the annual global turnover, whichever is greater; for SMEs and startups, the lesser of the two applies. Penalties for prohibited practices can reach 35 million or 7%.
Is it enough to put it in the privacy policy or terms of use?
No. Article 50.5 requires that the information be provided "in a clear and distinguishable manner, at the latest at the time of the first interaction." A notice that only exists in a legal document linked at the bottom of the page does not meet this requirement; it must be in the chat itself.
My chatbot is provided by an external company, is the responsibility theirs?
It depends on the role of each party. The obligation to design the system to inform lies with the provider, but if the chat operates under your brand and you deploy it, you can also be considered a provider, and in any case, you are responsible for data processing for GDPR purposes. Contract clauses do not eliminate the responsibilities that the law assigns objectively.
Can I give my virtual assistant a person's name?
You can give it a name, but you cannot create the appearance of a human: no portrait photographs, signatures as a team member, or evasive answers to "Are you a person?" The more human-like the design appears, the less applicable the "evident" exception is, and the clearer the notice must be.
Does the notice have to be in the client's language?
Yes, in practice. If the chat serves and responds in multiple languages, the notice must be issued in the language in which the conversation is taking place; a banner that only exists in Spanish does not comply with a client to whom the assistant is responding in English or French.
Do I have to use the European AI icon?
No. The icon is optional, not an obligation of the Regulation. What is required is that the information is clear, distinguishable, and accessible, with or without the icon.
In summary: an afternoon of work against a risk that is already open
The honest summary is that this is not a compliance project; it involves seven checks and probably two text changes in your chat. The obligation has been enforceable since August 2026, the Digital Omnibus has not granted you an extension on this point, and the fact that the Spanish bill has expired changes nothing because the Regulation applies directly. Open your chat as a client, on mobile and in every language you serve, and see if it is clear within the first ten seconds that there is an AI on the other side.
And if you are setting up that chat now, or want to change the one you have, check out aiReceptionist, the AI appointment booking and support chat from AIsToHire: it identifies as a virtual assistant in the header and greeting, informs in the six languages it serves, asks only for the data you configure, keeps all history in an inbox, and hands over the conversation to a member of your team as soon as needed. Getting started is free.
This article is general information about regulations, not legal advice. If your case involves health data or you have doubts about your legal basis, consult a professional.
Try aiReceptionist for free: a dating chat that identifies as a virtual assistant, notifies in the six languages it serves, and hands off the conversation to your team when needed.
Try aiReceptionist for free: a dating chat that identifies as a virtual assistant, notifies in the six languages it serves, and hands off the conversation to your team when needed.Keep reading
View all articles ā
How to Reduce No-Shows: 7 Data-Driven Measures and How to Automate Them

Why Your Car Insurance Increased in 2026: Average Premium of ā¬1,412 and Historic Record

Discussing via WhatsApp or talking in person: what science says about couple conflicts through messages
