WhatsApp Chatbot for a Psychology Practice: What Can Be Automated Without Touching the Session

A WhatsApp chatbot for a psychology practice answers one-off questions. An agent books the session against the practitioner's real calendar, applies a discount code where one applies, and places the appointment without anyone having to reply in person at 11 p.m. The distinction matters more here than in almost any other sector: what gets automated is scheduling and price, never the clinical conversation.

01

01

02

03

Why assessment stays out of scope

Under the EU AI Act, a mental-health chatbot that performs assessment, screening or diagnosis counts as high-risk, with mandatory risk assessment, transparency and human oversight. A tool that does pure scheduling, with no clinical decision-making, sits in a much lighter category and only has to disclose that it is AI. That single design choice — never touching assessment — is what keeps the classification honest.

It also matches what the evidence supports. A booking agent has no business inferring how someone is doing from the way they phrase a message; that judgment belongs to the person actually running the session, not to a scheduling tool reading a chat.

02

Weekly sessions that space out: a calendar that changes its own rhythm

Therapy typically starts with weekly sessions of 45 to 60 minutes and spaces out to biweekly or monthly as treatment progresses. A course of treatment lasts about three and a half months on average, and a cognitive-behavioral approach for mild-to-moderate anxiety often runs 12 to 16 sessions. That shifting cadence — weekly, then biweekly, then monthly — is exactly the pattern an automated calendar handles better than a fixed appointment book.

03

Cancellations: one fixed rule, applied every time

The common norm is straightforward: free with more than 24 hours' notice, a charge of 50 to 100% of the session for a late cancellation or no-show. An agent applies that same rule consistently, regardless of who happens to be answering WhatsApp that day.

04

01

02

03

What the agent does today — and the limits it always keeps

The agent talks over WhatsApp and Telegram, books against the real availability of a connected Google Calendar, applies discount codes, and answers with the practice's real service catalogue and real prices, at any hour. What it does not do: it does not cancel or reschedule an already-confirmed appointment, does not notify staff, and does not check whether a specific insurance plan covers a specific session — those actions stay with the practice.

Mental-health data sits in a special category under regulations like the EU's GDPR, and the agent stays strictly outside that territory: it only ever handles name, phone number, service type, availability and price. It never replaces the professional judgment of whoever is running the case — it only takes the administrative part off their plate, the part that never needed that judgment in the first place.

05

The word that stops the conversation

There is documented evidence that an AI's safety filters degrade over long conversations, and that simulated empathy can reinforce a suicidal thought rather than interrupt it. That is why the right response in that moment is not for the agent to say more — it is for it to say as little as possible. A fixed keyword match triggers a fixed response: the normal conversation halts, and the agent sends a single, fixed message pointing to local emergency services or a crisis line. That message exists to shorten the path to a professional or a public crisis service — never to replace one.

06

The administrative difference shows up most outside business hours: a practice that only answers WhatsApp while someone is at the desk loses the 11 p.m. message. With the agent, that message becomes a booked appointment — or, when it needs to, the message that points to the right help — before the practice reopens.