AI Chatbots on a Therapy Website: Where the Line Is

It's 9 pm. Someone has finally decided to look for a therapist, and your office is closed. A chatbot that can answer "Do you take my insurance?" feels like a gift.

It can be. It can also cause harm if it drifts into territory it shouldn't. The difference is a line you draw before you switch it on.

What a chatbot is good for

A chatbot works best on questions that have factual answers already sitting on your website:

• Whether you're accepting new clients
• Fees, insurance and payment
• Hours, location, and online or in-person sessions
• What a first session involves
• How to book or send an inquiry
• Which clinician works with which concerns, taken from your own pages

The bot isn't adding knowledge. It's finding what you've already published, faster.

What it must never do

• Answer clinical questions, like "Do I have anxiety?" or "Will therapy work for me?"
• Ask about symptoms, history or risk
• Suggest exercises or coping techniques
• Interpret anything the visitor shares
• Promise outcomes

Any of this can edge toward practicing without a license, and a bot can't read context the way you can. It can miss risk in someone who writes one quiet sentence. Keep it out of that room entirely.

Crisis: plan for it, don't improvise

Some people who open your chat will be in distress. The bot isn't there to counsel them. It should do three things:

• Show crisis resources in plain view at all times, such as the 988 Suicide and Crisis Lifeline in the US, or the local equivalent where you practice
• Respond to words that suggest risk by showing those resources and a way to reach a person, then stop
• Say clearly that it can't provide emergency help

Test this yourself with a few phrases before launch.

Privacy: people overshare

A stranger's general question about fees usually isn't protected health information. But people don't stay general. Someone may type a diagnosis or their story into the box, and from that moment the transcript is sensitive.

• Put a short notice above the chat telling people not to share personal health details
• If the chat could capture health information, ask the vendor whether they'll sign a business associate agreement. A "HIPAA-ready" label with no signed agreement protects nobody
• Find out where transcripts are stored, for how long, who can read them, and whether they're used to train models
• Never connect the bot to your notes or client records

HIPAA is a US law. Elsewhere, data protection rules such as GDPR apply, so check what governs your practice.

Be honest that it's a bot

Say it's an AI at the start. Offer a human option in every conversation, and make sure someone actually answers. A bot that ends in silence is worse than no bot.

Do you need one?

Maybe not. A clear FAQ page, visible fees and a quick human reply cover most of what a chatbot does. A bot earns its place when inquiries arrive outside office hours, or when a group practice has an admin team stretched thin. If you're a solo clinician who replies within a day, start with the FAQ.

We build chatbots as one of our services, so weigh our view with that in mind. Our view is that they suit fewer practices than the sales pages suggest.

Before you switch one on

1. Write down what the bot may answer and what it must not.
2. Load it only with your public website content.
3. Add the crisis notice and the "don't share health details" notice.
4. Get the business associate agreement question answered in writing.
5. Test it with 20 real questions, including awkward and distressing ones.
6. Check where transcripts go, and decide who reads them.
7. Review conversations weekly for the first month.

Thinking about adding one? Book a free call and we'll help you decide whether you need it at all.