HealthcareIndustries and use cases
AI Chatbot for Psychologists and Therapy Practices
Most people look for a therapist late at night, and most therapy websites answer nothing at that hour. A chatbot handles fees, availability, insurance and logistics. What it must never handle is a person in crisis, and that rule shapes everything else.
The 30-second answer
A chatbot on a therapy practice website answers what stops people getting in touch: session cost, plans you take, whether you have space, whether you offer video. It captures enquiries at midnight and hands intake a complete request. It never diagnoses, never advises, and never talks about a named client. Any hint of crisis goes to a human immediately, with local crisis numbers on screen.
Start with the crisis rule, not the features
A therapy practice website attracts people in distress, and some will type something frightening into a chat box at two in the morning. That fact decides how you set the bot up, and it comes before booking or fees.
The rule that does not bend
Any message suggesting crisis, self-harm, suicidal thinking, harm to someone else, or abuse must stop the automated conversation on the spot. The bot routes to a human immediately and shows local crisis lines and emergency services on screen.
It asks no follow-up questions. It does not judge how serious the message is. It offers no coping suggestions and no screening questionnaire. A bot must never be the only responder to a person in crisis.
If you cannot put a real human on that conversation at that hour, the widget must say so plainly. An honest message with the right crisis numbers beats a friendly bot that keeps someone talking to software.
Then test it like a fire alarm. Send the ten hardest messages you can imagine through the widget yourself. Every one should escalate. If a single message gets a chatty answer instead of a handoff, the bot is not ready. Our guide to chatbot handover protocols covers the pattern.
The questions that go unanswered at 11 p.m.
Reaching out to a therapist is hard enough, and when the website answers nothing, people close the tab and try someone else. These questions stall a new enquiry, and none of them are clinical:
Every answer above already sits in your head or in an email you have sent fifty times. The bot trains by crawling your site, so the work is publishing them once, after which the same answer lands at 11 p.m. in many languages.
What the bot must never do
Write these limits into the bot instructions, then test each one. They are not suggestions:
- • No diagnosis. Not a guess, not a hedge, not a scored questionnaire in the chat window.
- • No triage of symptoms. The bot never decides how urgent anything is. A human does.
- • No treatment advice. No coping techniques, no exercises, no reading list dressed up as help.
- • No medication questions. Dose, side effects, missed doses, interactions. All of it goes to a prescriber.
- • No talking about a named client. Confirming someone is a client is itself a disclosure.
- • No promises. No timelines for feeling better and no outcome claims.
Capturing an enquiry without collecting clinical detail
A good intake handoff is short: name, contact details, preferred times, in person or video, insurance name. That is enough for a coordinator to call back with a slot. The bot writes it up and drops it into your inbox, Slack or Microsoft Teams, which work on every plan, Free included.
Resist the temptation to ask what the person is going through. A free-text box about symptoms turns an ordinary enquiry into health information, which changes what you need and what you have to sign. Keep the widget on logistics and let the clinical conversation start with a clinician.
Where patient health information changes the rules
Public information stays public: fees, cancellation policy, accepted plans, the states you are licensed in. None of that is patient health information, and a basic setup handles it. The moment chat touches health information, the requirements change, so keep clinical intake with your team. Our HIPAA and GDPR guide covers the checks worth making first.
Realistic volume for one practice website
Practice owners overestimate this. The free plan covers 500 messages a month with no credit card, two seats and one chatbot. That is roughly three chats a day, and a single-clinician site often fits. A group practice running ads may need more: Paid plans are Starter at $25 a month and Professional at $95 a month, with 20% off on annual billing. Annual billing takes 20% off, and the tiers sit on the pricing page.
A four-week setup that keeps you safe
Week 1: Write the crisis rule first
Before a single fee answer, configure the escalation and publish your crisis resources page. Then test the routing with the ten hardest messages you can write.
Week 2: Publish the answers you repeat
The bot learns by crawling your site, so the site has to say it. Put fees, sliding scale rules, accepted plans, telehealth states and cancellation policy on real pages.
Week 3: Turn on enquiry capture
Collect name, contact, preferred times, format and plan name, then send the request to whoever runs intake. One person owns that queue.
Week 4: Read every transcript
Read the first weeks of chats end to end. Fix wrong answers, add what the bot missed, and check that every borderline message went to a human.
Judge the result on two numbers: new enquiries captured outside office hours, and how fast a human picked up every escalated chat. The second matters more. A practice that answers fee questions perfectly and takes eleven hours to answer a hard message has built the wrong thing.
Frequently asked questions
Can an AI chatbot support someone in a mental health crisis?
No. A bot must never be the only responder to a person in crisis. Any message suggesting self-harm, suicidal thinking or abuse should end the automated conversation, route to a human immediately, and show local crisis lines and emergency services on screen. No follow-up questions, no screening tool, no coping suggestions.
Can the bot tell a visitor what might be wrong with them?
No. No diagnosis, no impressions, no triage of symptoms, no treatment advice, and no medication guidance of any kind. The bot describes your published specialties and your intake process, then hands the person to a clinician who is licensed to judge.
Can it book a first session?
It captures what an intake coordinator needs and routes it in seconds: name, contact details, preferred times, in person or video, and insurance name. Keep free-text questions about what the person is going through out of the widget, because that turns an enquiry form into health information.
Is a website chatbot allowed for a therapy practice under HIPAA?
It depends on what the chat handles. Fees, cancellation policy, telehealth states and accepted plans are public information. Keep patient health information out of the chat and hand clinical questions to your team.
How many chat conversations will a small practice actually get?
Fewer than most people expect. The free plan covers 500 messages a month, about three a day, and a single-clinician site often fits inside that. Paid plans are Starter at $25 a month and Professional at $95 a month, with 20% off on annual billing. Annual billing takes 20% off.
Answer the easy questions. Escalate the hard ones.
EzyConn trains on your website in minutes and answers fee, insurance and availability questions around the clock. Every message that needs a person goes to one.
Start FreeLast updated . This guide covers website chat setup only. It is not clinical, legal or compliance advice, and your regulator and insurer have the final word. View more guides.