HIPAA-compliant chatbots: how to choose website chat for a practice

What 'HIPAA-compliant' means for website chat

HIPAA applies to how your practice and its vendors handle protected health information (PHI), not to software in the abstract. Website chat becomes a HIPAA question the moment a visitor types something like: "Hi, I'm Dana Ruiz, 555-0142, I need to see someone about my knee."

That is a name, contact details and a reason for a visit. HHS's business associate guidance (opens in a new tab) now lists a "third-party vendor Artificial Intelligence (AI) chatbot on a provider's patient portal that provides services involving the patient's PHI such as symptom assessment, medical reminders, and appointment scheduling" as an example of a business associate.

What good looks like:

• The vendor signs a BAA, and you can show it.
• Chats are encrypted and stored where you know, for as long as you decide.
• Only the staff who handle requests can read them.
• The bot sticks to front-desk questions and hands anything clinical to a person.
• Nothing from the chat leaks to ad or analytics platforms.

General-purpose assistants are a different question; see is ChatGPT HIPAA compliant and the wider HIPAA-compliant AI guide.

Follow the data: from chat window to your inbox

Every hop a chat takes is a place PHI can sit. Map yours before you sign anything.

Three things in that table catch practices out:

• Encryption doesn't remove the BAA. HHS says a cloud provider that stores PHI is a business associate "even if the CSP processes or stores only encrypted ePHI and lacks an encryption key" (HHS cloud guidance (opens in a new tab)).
• Subcontractors count. If the chat vendor sends conversations to an AI model provider, that provider is the vendor's subcontractor and needs its own BAA with the vendor.
• Tracking scripts. HHS's online tracking bulletin (opens in a new tab) says tracking technologies that collect "an individual's email address, or reason for seeking health care" while someone books an appointment are receiving PHI. A federal court vacated part of that bulletin in June 2024 (American Hospital Association v. Becerra), to the extent it treated an IP address plus a visit to a public page about a health condition as enough to trigger HIPAA. HHS still publishes the rest of the bulletin, including the appointment example, as of October 2026. HHS also says cookie banners are not a HIPAA authorization.

This is general information, not legal advice. Confirm how it applies to your practice with a qualified advisor.

Vendor checklist: BAA, encryption, retention, access logs

Use this on every demo. A "no" or a vague answer on the first item ends the conversation.

• Signs a BAA for the plan you're buying, not only for an enterprise tier you aren't buying. Compare it against a BAA requirements checklist.
• Encrypts in transit and at rest. The Security Rule lists encryption as an addressable safeguard for stored data and for transmission (45 CFR 164.312 (opens in a new tab)).
• Tells you where data lives, including the country, and where backups go.
• Lets you set retention and delete transcripts, and confirms deletion in writing when you leave.
• Doesn't train models on your chats. Read the terms, not the sales deck.
• Role-based access and an audit log of who opened which conversation. The Security Rule requires audit controls.
• Human handoff with a clear path to a person during and after hours.
• Emergency language shown before and during the chat: "If this is an emergency, call 911 or go to the nearest emergency room."
• AI disclosure, so visitors know they're talking to software.
• No third-party marketing pixels inside the widget.

Red flags

• A claim to be "certified" for HIPAA. No such certification exists.
• A BAA only on request, after you sign.
• Transcripts emailed in full to a shared Gmail address by default.
• The bot offers symptom advice in its demo.
• No answer to "which AI model provider do you use, and do they have a BAA with you?"

What a practice chatbot should and shouldn't answer

The full list of limits, plus a test plan, is in what medical chatbots should do.

Disclosure laws

Some states regulate bots directly. California makes it unlawful to use a bot that misleads people about its artificial identity to drive a sale, unless the bot discloses it is a bot (Business and Professions Code 17941 (opens in a new tab)). California also requires an AI disclaimer and a way to reach a human when a physician's office uses generative AI for patient messages about clinical information; administrative matters such as scheduling are excluded (Health and Safety Code 1339.75 (opens in a new tab)). Check your own state before launch.

Contact form, live chat or AI chat: choosing the option

A HIPAA-compliant contact form is the simplest, and for many small practices it is enough. Live chat suits practices that have someone free to answer within a minute. AI chat earns its keep when many visits arrive in the evening and on weekends.

Cost drivers

• Seats. Live chat is usually priced per agent.
• Volume. AI chat may be priced by conversations or messages.
• The HIPAA tier. HIPAA is often sold higher up the price list. Two examples from other categories: Otter lists HIPAA as an Enterprise add-on (Otter pricing (opens in a new tab)), and Jotform offers HIPAA features only on its Gold and Enterprise plans (Jotform pricing (opens in a new tab)).
• Integration. Sending chats into an EHR or CRM can add setup fees.
• Staff time to read, respond and review transcripts.

Setting it up: scripts, handoff and after-hours

1. Write the opening message. Illustrative example for a fictional clinic: "Hi, I'm Harbor Family Medicine's virtual assistant. I can help with hours, insurance and booking requests. I can't give medical advice. If this is an emergency, call 911."
2. Limit what the bot collects. Name, phone, email, new or existing patient, preferred times, and a short optional reason. No symptom questionnaires.
3. Define handoff triggers: any clinical question, an upset visitor, billing disputes, or a request for a person.
4. Plan after-hours. Say when the office will call back, and give the after-hours number for urgent clinical questions.
5. Route requests to one place, ideally a patient CRM with access controls, not a shared mailbox.
6. Keep notification emails thin. "New request from the website chat" plus a secure link, not the conversation itself.
7. Remove tracking pixels from pages where the widget loads, and check after every website change.
8. Review transcripts weekly for wrong answers and fix the source information.

A free marketing audit flags privacy risks such as pixels on patient forms, alongside your website, SEO and AI visibility scores.

Medfy's HIPAA AI chat widget

Medfy offers a HIPAA AI chat widget as a $49/month add-on to the Medfy plan ($349/month). It is covered by the BAA Medfy signs with your practice. Chat data is stored in Medfy's separate, encrypted regional databases in the US or Canada, and new-patient requests land in the patient CRM and inbox, next to your website form submissions.

See HIPAA AI chat widget pricing, and how Medfy protects chat data for the safeguards behind it. If you already have a chat vendor, use the checklist above to compare.

Frequently asked questions

Is a website chat widget subject to HIPAA?

Usually, yes. Once a visitor types a name, a phone number and why they want an appointment, the chat holds health information about a prospective patient, and the vendor storing it is acting for your practice. HHS lists a third-party AI chatbot that handles appointment scheduling for a provider as a business associate. Confirm your specific setup with your compliance officer or attorney.

Do I need a BAA with my chatbot vendor?

If the vendor receives, stores or transmits chats that contain patient information, yes, and HHS says that holds even when the vendor only stores encrypted data it can't read. The BAA must cover the exact plan you buy. Any subcontractor that touches the chats, such as the AI model provider, needs its own BAA with the vendor. Have your attorney review the agreement.

Should a chatbot answer medical questions?

No. A practice chatbot should answer practice questions such as hours, services, insurance accepted and directions, then collect contact details so staff can call back. Diagnosis, dosing and treatment advice belong with a clinician. Every chat window should also tell visitors to call 911 or go to the nearest emergency room if they are having an emergency.

Where are website chat transcripts stored?

On the chat vendor's servers, and often in a second place: your email inbox, a CRM or a helpdesk the vendor forwards them to. Ask each vendor where data is hosted, how long transcripts are kept and how deletion works. Every system that holds a transcript needs a BAA and access controls, including the inbox that receives notifications.

Is live chat safer than an AI chatbot?

Not automatically. Live chat puts a trained person on every reply, which removes the risk of a wrong automated answer, but transcripts and notifications still need a BAA, encryption and access controls. An AI chatbot can answer after hours with tightly limited scope. The safer option is whichever one you can configure, staff and monitor well. Check your choice with your compliance officer.