AI disclosure for dental patients happens in three moments: the first message says the patient is writing to an AI assistant, a patient who asks for a person gets one, and opted-out patients are suppressed before any outreach. Article 50 of the EU AI Act, applying since 2 August 2026, asks that people be told they are interacting with an AI at the latest at the first interaction. How a practice and its vendor split that duty is a question for the practice's data protection adviser.
AI disclosure for dental patients comes down to three moments in a conversation. The patient is told, in the first message, that they are writing to an AI assistant. A patient who asks for a person gets one. And a patient who says "stop" is never contacted for that purpose again, which means the suppression list is applied before any outreach, not after a complaint.
Since 2 August 2026, Article 50 of the EU AI Act has applied. It asks that people be told they are interacting with an AI system, at the latest at the first interaction. That makes disclosure a live question for any dental practice putting an AI in front of patients, and for anyone judging whether a vendor operates one safely.
I lead operations and product at Keishal. We operate patient conversations in WhatsApp for dental clinics in Spain. This is not a legal explainer, and nothing here is legal advice. It is an operator's account of what disclosure, opt-outs and hand-offs look like inside real patient conversations, and of the questions a practice should take to its own data protection adviser.
What the AI Act asks, and what it leaves open
The part of the AI Act that matters for a patient-facing assistant is short. Article 50 asks that a person interacting with an AI system be informed of it, at the latest at the moment of the first interaction. It has applied since 2 August 2026.
What the text doesn't do is write the message for you. It doesn't say which words to use in a WhatsApp chat, whether a reminder counts as an interaction in the same way a free conversation does, or how the duty is split between the company that builds the system and the practice that uses it. Those are questions for the practice's data protection officer or legal adviser, and the answers can differ from country to country.
So the useful question for an operator is not "are we compliant?". That is for specialists to answer. It is "where, exactly, in each kind of conversation, does the patient learn what they are talking to?". That one we can answer with the product.
The first message: where AI disclosure goes
A dental practice's AI assistant has two kinds of first message, and disclosure works differently in each.
The patient writes first. Someone messages the clinic's WhatsApp at 22:40 to ask whether there's a slot this week. The first reply is where the patient learns they are talking to an AI assistant working for the clinic, before any answer about slots. An illustrative wording, not our template: "Hi, this is the AI assistant of [clinic name]. I can answer questions and help you book. If you'd rather speak to someone from the team, just say so."
The clinic writes first. A reminder the day before an appointment, an offer of a cancelled slot, a message to a patient who hasn't been in for two years. Here the disclosure has to travel with the outbound message itself, because there is no earlier moment. A patient who replies to a reminder should never find out halfway through a conversation that the "receptionist" was software.
Three design rules follow from this, whichever product you use:
- Disclosure comes before the substance. Not in a footer, not after the booking link, not only in the privacy notice.
- It names the clinic. The patient's relationship is with their dentist. The assistant writes on the clinic's behalf, in its WhatsApp number and its voice, and says so.
- It offers the way out in the same breath. Telling someone they are talking to an AI is only half the job if they don't also know how to reach a person.
In our case, patients are told they are talking to an AI assistant. We treat that as a product fact rather than a risk to manage. A common worry is that patients will stop replying once they know. Some may prefer to wait for a person, and that's their choice. A patient who feels misled about who they were talking to is a far bigger problem for a clinic than one who chooses to wait for a person.
"Am I talking to a person?"
Sooner or later a patient asks. Sometimes it is curiosity, sometimes they want to say something they wouldn't say to software, sometimes they are annoyed. There is only one acceptable answer: a plain yes-or-no, immediately, followed by the offer of a person.
That request starts a hand-off, and the hand-off is where most of the safety work sits. We covered the full list in what a dental AI receptionist actually does: anything clinical, anything the clinic hasn't approved as an answer, complaints, and decisions only the clinic can make. "I want to talk to someone" belongs on the same list, and it is honoured, not argued with.
What makes a hand-off work in practice is less obvious than the rule:
- The person receives the conversation, not a summary of it. The team member sees what the patient wrote and what the assistant answered, so the patient doesn't repeat themselves.
- The patient is told what happens next. That a person from the team will reply, and roughly when, in the clinic's own opening hours.
- The assistant stops. Once a conversation is with a person, the AI doesn't keep answering over them.
Here is the load that creates, as illustrative arithmetic you can redo with your own numbers. Take a practice whose assistant handles 400 written conversations a month. If 1 in 20 ends with a request for a person, that's 400 ÷ 20 = 20 hand-offs. At 5 minutes each for the team, 20 × 5 = 100 minutes a month.
That is under two hours, and it is the right two hours: a person dealing with exactly the patients who asked for one. If your hand-off rate is much higher, the useful question is why. Usually the assistant is being asked things the clinic never wrote down, which is a gap in the clinic's approved information, not in the patient.
Opt-outs: the suppression list comes first
Disclosure is about what the patient is told. Opt-outs are about who is contacted at all, and they matter most in outbound work: reminders, gap-filling and, above all, reactivation of patients who haven't been back in a year or more. We wrote about why dental recall fails, even with recall software; suppression is the step in that process that must not fail.
Before any outreach, Keishal automatically excludes four kinds of record:
- patients who have opted out;
- patients flagged as deceased in the practice management system;
- patients who have moved abroad, where that is flagged;
- any record the clinic has flagged as "do not contact".
The order is the point. Suppression runs before the first message is sent, not after someone complains.
To see why, size it on a database. These are illustrative figures, not averages and not Keishal results. A practice has 1,500 patients who haven't been in for 12 months or more. Suppose 4% have opted out, 1% are flagged as deceased, 2% have moved abroad and 3% are marked "do not contact". That's 4 + 1 + 2 + 3 = 10%, and 1,500 × 10% = 150 people.
Run the campaign without the list and those 150 people get a message they should never have received. The worst of them is a "time for your check-up" message to the family of a patient who has died. One message like that does more damage to a clinic's reputation than the whole campaign can recover.
Then there is the opt-out that happens inside a conversation. A patient replies "please don't message me again" to a reactivation message. That reply has to become a flag on the record, so the next campaign, months later, skips them. A list that only lives in someone's memory, or in a spreadsheet nobody re-checks, fails exactly here.
What changes by country
Opt-out rules sit in national law as well as in the GDPR, so the detail depends on where the practice is.
In Portugal, for example, the data protection authority (CNPD), in Diretriz 2022/1, reads the direct-marketing rule of Lei 41/2004 (art. 13-A) as follows. Without consent, an existing customer can be contacted about similar products or services only if a free, easy opt-out is offered at collection and in every message, with the sender identified. The guideline names e-mail, SMS/MMS and calls; WhatsApp isn't named. Whether a reactivation message falls under that rule is a question for Portuguese specialist advice, and we don't answer it here. For a market overview, we've written about the Portugal dental market.
Who is responsible for the patient data
The roles are worth stating plainly, because they decide who answers the questions above.
- The clinic is the data controller of its patients' personal data. It decides who is contacted and why.
- Keishal acts as a data processor under Article 28 of the GDPR, with a written data processing agreement (DPA) with every clinic.
- Patient data is stored in the EU, and processing is handled by EU-based providers.
- Keishal has an external data protection officer, registered with the Spanish data protection authority.
The assistant reads the practice management system for context and writes outcomes back: an appointment, a confirmation, a status change. The same principle applies to opt-outs. Wherever the "stop" is recorded, it has to be somewhere the next campaign reads; if it lives only in a chat history, it depends on someone copying it across by hand.
Questions for your DPO or adviser
We don't sentence on what a given practice must do. These are the questions we'd want a clinic to put to its own data protection officer or legal adviser before any AI talks to its patients:
- Under Article 50 of the AI Act, how should the practice and the vendor split the duty to inform patients, and what does each need to document?
- Does a reminder about an appointment the patient already booked need the same disclosure as a free conversation?
- Which outbound messages count as direct marketing under national law, and which as care communication?
- Where should the AI assistant be described in the practice's privacy notice?
- How should an opt-out given in WhatsApp be recorded, and does it apply to every channel?
- What should the data processing agreement say about where data is stored and who the processor's subprocessors are?
Questions to ask any AI assistant vendor
The same topics, asked of the company selling the assistant:
- Show me the first message. For an inbound chat and for each outbound type: where exactly does the patient learn it's an AI?
- What happens when a patient asks for a person? Who receives the conversation, how, and does the AI stop answering?
- When is the suppression list applied? Before every send, or once at import?
- Where does an opt-out go? Is it written back to the practice management system, or kept only inside the vendor's tool?
- Who operates it after go-live? If the clinic's team has to maintain the rules, the approved answers and the lists, count those hours. We explain why that question decides most outcomes in service-as-a-software in dental practices.
- Where is the data kept, and under what agreement?
Where Keishal fits
Keishal is an autonomous system that operates patient communication in WhatsApp for dental clinics: questions, booking, reminders, cancelled slots, reactivation and reviews. It tells patients they are talking to an AI assistant, runs suppression rules before any outreach, hands conversations to the clinic's team when the patient or the subject calls for it, keeps patient data in the EU, and works as a data processor under a DPA with each clinic. It reads the practice system for context and writes outcomes back; the systems it connects to today are on our integrations page.
We serve clinics in Spain, and we are researching Portugal. If you want to see the conversations themselves, book a demo.
Article 50 date: Regulation (EU) 2024/1689 (AI Act), Art. 50 and Art. 113. Portugal: Lei 41/2004 art. 13-A as read by CNPD Diretriz 2022/1. Workload and database figures are illustrative arithmetic, not Keishal results. Product description as of November 2026. This article is not legal advice.

WhatsApp for dental clinics: everyone is on it, almost nobody books through it
WhatsApp for dental clinics in southern Europe has reach but little use. In Portugal, 79% of online adults used WhatsApp in the past week (Reuters Institute, 2026), yet 65.3% of patients book by phone and only 1.8% book online through social networks, WhatsApp or a website (OMD Barómetro 2025). The bottleneck is who operates the line: usually one person, in office hours, about 32% of the week.

Dental practice management software integration: read for context, write back the outcome
Dental practice management software integration, for a system that talks to patients, means two directions: read the practice system for context (agenda, contact details, last visit, do-not-contact flags) and write the outcome back (appointments, statuses, documents). Without write-back, every outcome is re-keyed by hand: 150 changes a month at 2 minutes is 5 hours in an illustrative clinic. In a fragmented PMS market like Spain's, the answer is a modular connector layer, not one deep integration.

