Resources
HIPAA-Compliant Chatbots: What a Clinic Website Chat Needs (and When It Should Never See PHI)
When a clinic website chatbot handles PHI, what a BAA and safeguards must cover, how to keep PHI out by design, and the questions to ask any chatbot vendor.
By ClinicSite Editorial Team, Healthcare Practice Web Architecture · Published July 21, 2026
This article is general information, not legal advice. Ask a compliance adviser or attorney about your specific setup.
Two different things: staff AI tools and a chat widget on your website
Search results for "HIPAA-compliant chatbot" mix two separate questions:
Staff-facing AI tools. Can your team paste patient notes into an AI assistant to draft a letter or summarize a chart? That is a question about a vendor your staff uses, and it usually turns on whether the vendor signs a business associate agreement (BAA) for that specific service.
A patient-facing chat widget on your website. A visitor types into a box on your homepage. You do not control what they type, and they are often not yet your patient.
This guide is about the second. The risks are different: staff tools handle PHI on purpose, while a website chat often handles it by accident.
When does website chat handle PHI?
HIPAA's definitions live at 45 CFR 160.103. Three pieces matter here:
Covered entity. A health care provider is covered if it "transmits any health information in electronic form in connection with a transaction" covered by HIPAA, such as electronic insurance claims. Many practices that bill insurance electronically qualify. Some cash-only practices may not. Ask your adviser which you are.
Health information includes information "created or received by a health care provider" that relates to a person's "physical or mental health or condition," or "the provision of health care to an individual."
Protected health information (PHI) is individually identifiable health information, meaning it identifies the person or there is "a reasonable basis to believe" it could.
Put together: if your practice is a covered entity and a visitor types "Hi, I'm Dana Ruiz, dana@example.com, I've been having panic attacks and want an appointment", your practice has likely received PHI through the chat. A visitor asking "Do you take Blue Cross?" or "Is there parking?" has not shared health information.
The trouble is that you cannot stop the first kind of message from being typed. The design question is what your chat does when it happens.
If your chat is meant to handle PHI
Some practices want chat to do intake, triage or appointment requests with symptoms. That is a legitimate choice, but it makes the chat vendor part of your HIPAA obligations:
A BAA. A vendor that "creates, receives, maintains, or transmits" PHI on your behalf is a business associate (45 CFR 160.103). HHS's cloud computing guidance says a cloud provider that stores PHI for you is a business associate, and that this "is true even if the CSP processes or stores only encrypted ePHI and lacks an encryption key for the data." Ask for the BAA before turning the feature on, and check that it covers the chat product you are using, not just some of the vendor's services.
Every vendor in the chain. If the chat vendor sends messages to an AI model provider, that provider is a subcontractor handling PHI too. Ask whether the vendor's own agreements with its AI provider cover it.
Security Rule safeguards. HHS's tracking bulletin summarizes them: risk analysis, access controls, audit controls (45 CFR 164.312(b)), authentication, and encryption in transit and at rest where reasonable and appropriate.
Breach notification. An impermissible disclosure of unsecured PHI can trigger notification duties. Ask how the vendor will tell you about an incident and how fast.
Retention you control. Find out how long transcripts are kept, where, and whether you can delete them.
None of this produces a certificate. HHS says it "does not endorse or otherwise recognize private organizations' 'certifications'" regarding the Security Rule. A "HIPAA compliant" badge on a vendor's site is marketing; the BAA and the safeguards are what count.
If your chat should not handle PHI: design guardrails
For most small practices, the questions coming through website chat are not clinical: insurance, hours, parking, whether you see children, how to book. A chat built only for those questions can be designed to stay out of PHI:
Scope it to public practice information. The bot answers only from what you would publish on your website anyway: services, hours, address, insurers, booking link.
Refuse clinical questions. No symptom assessment, no "is this normal?", no medication advice. The bot redirects to a call or appointment.
Do not ask for intake details. No date of birth, no reason for visit, no insurance member numbers. If the bot never asks, fewer visitors volunteer.
Warn before people type. A short line above the input box: "Please don't share health details here. For anything about your care, call [number]."
Hand off to a person. When the bot cannot answer, the visitor asks for a human, or something sounds urgent, it stops and routes the conversation to your team or tells the visitor to call.
Show crisis information. For therapy and behavioral health practices especially, include the 988 Suicide & Crisis Lifeline (call, text or chat, 24/7) and tell anyone in immediate danger to call 911.
What to do when visitors type PHI anyway
Guardrails reduce PHI in chat; they do not eliminate it. Plan for the message that arrives anyway:
Do not reply with more detail. Move the conversation to the phone or your patient portal.
Know where the transcript lives. If it sits in a vendor inbox, know who on your team can read it and how long it stays.
Decide on deletion. Ask whether you can delete a transcript that contains clinical detail, and whether deletion also removes it from backups and any AI provider logs.
Talk to your adviser. Whether a scoped, PHI-free chat still needs a BAA, given that some visitors will type health details, is a judgment call to make with your compliance adviser, not a vendor's sales team.
Questions to ask any chatbot vendor
Ask in writing and keep the answers:
Will you sign a BAA, and does it cover this chat product specifically?
Which AI model provider processes the messages, and do your agreements with it cover PHI?
Are messages used to train any model? Can that be turned off?
Exactly what is logged: message text, visitor IP address, browser details, page URL?
How long are transcripts and logs kept, and can we shorten that?
Can we delete a single conversation, and does deletion reach backups?
Who at your company can read our conversations, and is that access logged?
Where are the servers, and is data encrypted in transit and at rest?
What does the bot do with clinical questions, urgent messages and requests for a person?
Can we restrict its answers to our own content only?
How will you notify us of a security incident, and how quickly?
What happens to our transcripts if we cancel?
Vague answers to questions 4 and 5 are a reason to slow down. For a wider view of what a practice website should keep off its pages, see our HIPAA website requirements guide.
How the ClinicSite assistant is scoped
ClinicSite offers a website assistant on the Elite plan. Its feature page describes it as answering "only from your own practice information": a knowledge base of services, hours, address, phone, email, insurers, booking link and question-and-answer pairs you can edit. According to that page, it is instructed to give no medical advice, diagnoses or treatment recommendations; it is instructed "not to collect or store personal health information"; and it hands off to your clinic when it cannot answer, when a visitor asks for a person or wants to book by phone, or when a request is urgent or complex.
The widget runs on Chatwoot, an open-source messaging platform, with a separate account and inbox per clinic. Handed-off conversations stay open in your clinic's inbox for your team to take over. The feature page also advises treating the chat "as a public channel" and moving anything clinical to a phone call.
What we do not claim: we do not describe the assistant as HIPAA compliant, and ClinicSite publishes no HIPAA certification and no SOC 2 report. The feature page does not state a retention period for chat transcripts, so ask us, as you would ask any vendor, using the questions above. The assistant is not a triage tool, and the handoff goes to your clinic; ClinicSite does not staff your inbox. Plans are listed on the pricing page.
To see what your own site would look like rebuilt, you can paste your website's address for a free preview.
Related Practitioner Guides
Draw a clean architectural boundary between public pages and clinical intake in our patient portal links vs. marketing website guide.
Review BAA boundaries for online patient intake in our HIPAA rules for medical website forms.
Understand why third-party ad pixels create regulatory exposure on clinic pages in our Google Analytics and tracking pixels HIPAA analysis.
See how source-grounded generation prevents invented credentials in our medical practice AI website accuracy guide.
Sources
45 CFR § 160.103 — HIPAA General Definitions (Covered Entity, Business Associate, PHI)
HHS OCR Guidance on HIPAA & Cloud Computing Business Associates
HHS Office for Civil Rights: Use of Online Tracking Technologies by HIPAA Covered Entities
HHS HIPAA FAQ: Are Organizations Required to Certify Compliance With the Standards?
Frequently asked questions
- Is there such a thing as a HIPAA-compliant chatbot?
- Not as a product label. HIPAA compliance depends on how a covered entity and its vendors handle PHI: a BAA, safeguards and policies. HHS says it does not endorse or recognize private organizations' HIPAA certifications. A chatbot can be part of a compliant setup, but the badge alone tells you little. Ask for the BAA and the safeguards.
- Does my website chatbot need a BAA?
- If the chat vendor creates, receives, maintains or transmits PHI on your behalf, it is generally a business associate and you need a BAA. If the chat is scoped to public questions and refuses clinical detail, the answer is less clear, because some visitors will type health information anyway. Make that call with your compliance adviser.
- Can a chatbot on my website answer medical questions?
- It can be built to, but for a small practice that creates two risks: confident wrong answers published under your name, and clinical details collected in a chat log. Many practices instead scope the bot to logistics such as hours, insurance and booking, and have it redirect clinical questions to a clinician by phone or appointment.
- What should a chatbot do if a visitor describes a crisis?
- It should stop answering, show crisis information, and route the conversation to a person. For behavioral health practices, that means naming the 988 Suicide & Crisis Lifeline (call, text or chat, 24/7) and telling anyone in immediate danger to call 911. Ask any vendor exactly how its bot detects and handles urgent messages.
- Can I use a general consumer AI chatbot on my clinic website?
- General-purpose chatbots will answer almost anything, including symptom questions and insurance details they cannot verify. Unless the vendor signs a BAA covering the product and lets you restrict answers to your own content, it is a poor fit for a patient-facing clinic site. Ask the vendor the questions above before installing anything.
- How long should chat transcripts be kept?
- There is no single number that fits every practice. What matters is that you know the retention period, can shorten it, and can delete individual conversations that contain clinical detail. Ask your vendor in writing, and check with your compliance adviser whether state record-keeping rules affect your decision.
- Does a cash-only therapy practice need to worry about HIPAA for chat?
- Possibly not under HIPAA itself, because providers are covered entities only if they conduct certain transactions electronically, such as electronic insurance claims. But state privacy laws, professional ethics and client trust still apply. Confirm your status with an adviser, and keep clinical details out of website chat either way.
Related
- Webflow vs Framer for Medical Practice Websites: 2026 Costs, Redirects and HIPAA LimitsWebflow Premium is $25/mo, Framer Pro $30/mo plus $40 per editor, and Framer Basic has no 301 redirects. A 2026 cost, SEO and HIPAA breakdown for clinics.
- Psychiatric and Ketamine Clinic Website Examples: 6 Blueprints That Book Safely6 annotated psychiatry, TMS, Spravato and IV ketamine website blueprints: REMS 2-hour monitoring copy, off-label wording, 988 placement and LCP under 2.5s.
- New Medical Practice Website Launch Checklist (2026)A 30-day website launch checklist for new medical practices: domain, core pages, online booking, local SEO, accessibility, and QA before opening day.
- How Much Does a Medical Practice Website Cost in 2026? Three-Year Totals by ApproachWhat a U.S. medical, dental or therapy practice website costs over three years by approach, with dated vendor prices, hidden costs and a staff-time worksheet.
- All resourcesEvery ClinicSite guide in one place.
See what your clinic site could look like
Paste your current website address. You get a full generated replacement to look at, free, with no account and no card.
Free preview with no card. Previews are deleted after 7 days.