Search for a HIPAA compliant AI chatbot and you get lists of twenty tools, each with a shield icon. That does not help you decide. This page gives you the straight answer, the three things that make an AI tool safe for patient information, and the eight questions to ask any vendor before you share a single record.
BAA
the agreement that comes first
3
layers a compliant setup needs
8
questions to ask any vendor
14 days
about, for a Done-For-You build
The short answer
HIPAA does not approve or ban software. It sets rules for how patient information is protected, and those rules apply to your practice and to any vendor who handles that information for you. So the real question is not "is this AI compliant?" It is "can this vendor, on this plan, set up this way, handle our patients' information under HIPAA?"
That comes down to three layers. You need all of them.
01
A signed BAA
The vendor agrees in writing to safeguard patient information. No BAA, no patient data.
02
HIPAA-eligible infrastructure
Hosting, storage and processing that can be covered by that agreement, end to end.
03
The right setup
What goes in, what it answers, what it hands to staff. This is where most tools go wrong.
Most tools fail on the first layer: they will not sign a BAA for the plan you are on. Some pass the first two and fail the third, because the AI answers from its own general knowledge and says things your clinicians never approved. A compliant setup gets all three right.
What HIPAA asks of an AI vendor
If an AI tool creates, receives, stores or sends patient information on your behalf, the vendor is a business associate. HHS is clear about what has to happen before you share anything. In its guidance on business associates, it says the HIPAA Rules permit a covered entity to disclose PHI to a business associate "if the covered entity obtains satisfactory assurances, in the form of a contract or other written arrangement (collectively referred to as a 'business associate agreement,' or BAA), that the business associate will appropriately safeguard the information".
The same page adds: "A business associate is also directly liable for complying with certain provisions of the HIPAA Rules." So the BAA is not a formality. It is the vendor putting its own name on the line.
In practice, that gives you a simple first test. Ask for the BAA. If a vendor will not sign one, or will only sign one on a plan you are not buying, stop there. Nothing else on their sales page matters.
What counts as PHI
PHI is broader than most people think. The HHS summary of the Privacy Rule says it covers "individually identifiable health information" held or transmitted by a covered entity or its business associate, "in any form or media, whether electronic, paper, or oral."
That includes a chat. If a patient types their name and asks about their own symptoms, dose or lab result, the transcript is PHI. It does not matter that it arrived through a website widget rather than a chart.
The other side matters too. The same summary says: "There are no restrictions on the use or disclosure of de-identified health information." Your patient education, the handouts you give everyone, is not about any one person. That is why it makes a good training set for a clinic clone, and why patient records do not need to go in.
Usually not PHI
- Your side-effect FAQ
- Lab preparation sheets
- Opening hours and policies
- A meal guide you hand out
- Pricing and what a first visit covers
PHI
- A patient asking about their own symptoms
- Their name with their dose or plan
- A lab result, a chart note, an intake form
- Billing for their care
- A chat transcript that identifies them
Why a badge on a website is not enough
"HIPAA compliant" is easy to write on a landing page. It tells you nothing about the plan you are buying, how the data flows, or what the AI will say to a patient at midnight.
Be careful with three patterns. A general-purpose chat tool that only offers a BAA on an enterprise tier, while the plan you can sign up for today has none. A chatbot builder that is HIPAA-ready but leaves the setup to you, so compliance depends on settings you may not know exist. And a tool that answers from the open internet, which can be fully secure and still tell a patient something your clinicians would never say.
None of them is the same thing as a compliant setup for your practice. The checklist below is how you tell the difference.
Self-serve vs Done-For-You
Most AI clone and chatbot products are self-serve. You sign up, upload some material and go live when you are ready. That is fine for a coach answering questions about a course. It is the wrong shape for a clinic.
We draw the line plainly on our own plans. Self-serve Personify, Free and Pro, is not HIPAA compliant and must not be used with patient data. Healthcare clones are Done-For-You builds only.
| Self-serve (Free, Pro) | Done-For-You healthcare build | |
|---|---|---|
| Who it is for | Coaches, creators, educators | Clinics and health coaches |
| BAA signed | No | Yes |
| HIPAA-eligible infrastructure | Not for patient data | Yes |
| Patient information | Must not be used | Covered by the BAA |
| Who sets it up | You | We build it, your clinicians review it |
| Time to launch | When you are ready | About 14 days from the discovery call |
Personify's own plans. Self-serve is not HIPAA compliant and must not be used with patient data.
The reason is not just the paperwork. A healthcare build needs someone to decide what goes in, what the clone never answers, and where clinical questions go. That is a conversation with your clinicians, not a settings page.
What to ask a vendor before you share anything
Take these eight questions into any vendor call. The first three cover the BAA and the infrastructure. The other five cover how the AI behaves, which is where most of the real risk sits.
| Ask the vendor | A good answer | A red flag |
|---|---|---|
| 1Will you sign a BAA with us? | Yes, before any patient data is shared | "We are HIPAA compliant" with no agreement |
| 2Which plan does the BAA cover? | The exact plan you are buying | Only an enterprise tier you are not on |
| 3Where is our data hosted and processed? | HIPAA-eligible infrastructure, explained to your IT contact | Vague answers, or "the cloud" |
| 4What does the AI answer from? | Only your approved material | The open internet or its own general knowledge |
| 5What happens with a clinical question? | It hands it to your staff | It answers anyway |
| 6What happens in an emergency? | It points to 911 or a crisis line | Nothing defined |
| 7Who reviews answers before launch? | Your clinicians | Nobody, it just goes live |
| 8Who keeps it current? | Named, with a process | You, with a login and no help |
Eight questions. Take them into any vendor call, ours included.
Get the answers in writing. Your compliance or IT contact should see the technical detail before anything is built, not after launch.
What a compliant clone should and should not do
Protecting the data is half the job. The other half is what the AI says. A clinic clone should answer only from your approved material: plans, handouts, prep instructions, policies. It should never diagnose, prescribe or change a dose. Clinical questions go to your staff.
It is not a crisis service either. In the US, anyone describing an emergency should be told to call 911, and the 988 Lifeline is available 24/7/365 for anyone in crisis.
A compliant clone protects the data and knows when to stop talking.
For a worked example, including a real hand-off on a Sunday evening, see AI patient follow-up for weight loss, GLP-1 and TRT clinics.
How Personify handles it
Here is what we do, so you can hold us to the same checklist.
- Healthcare clones are Done-For-You builds only.
- We sign a Business Associate Agreement with your practice.
- We build on HIPAA-eligible infrastructure, and walk your compliance or IT contact through it on the discovery call.
- The clone is trained only on your approved material, never on patient records.
- It never diagnoses, prescribes or changes a dose, and hands clinical questions to your staff.
- Your clinicians review it before launch. It is live in about 14 days from the discovery call.
- Builds are priced by scope, and we quote after the call.
If you run a clinic or a health coaching practice, see how healthcare builds work and book a discovery call.
Frequently asked questions
Is there a HIPAA compliant AI chatbot?+
Yes. An AI chatbot can be used in a HIPAA-compliant way when the vendor signs a Business Associate Agreement, runs on HIPAA-eligible infrastructure, and the tool is set up to handle patient information properly. No chatbot is compliant on its own name alone.
Are general-purpose AI chat tools HIPAA compliant?+
Ask the vendor of any general-purpose chat tool whether they will sign a BAA for the exact plan you are on. If the answer is no, do not put patient information into it.
Is self-serve Personify HIPAA compliant?+
No. Healthcare builds are Done-For-You only. The self-serve Free and Pro plans are not for patient data, so please do not use them with patient information.
Do you sign a BAA?+
Yes. We sign a Business Associate Agreement with your practice as part of every healthcare build.
Can a HIPAA compliant AI give medical advice?+
Compliance is about protecting patient information, not about what the AI is allowed to say. A clone we build never diagnoses, prescribes or changes a dose. Clinical questions go to your care team, and emergencies go to 911.
Can I train an AI clone on patient records?+
We train clinic clones on patient education, not patient records: your handouts, FAQs, prep sheets and policies. If you would not hand a document to a patient at the front desk, it does not go in.
How long does a HIPAA compliant build take?+
About 14 days from the discovery call to launch, including your clinicians' review.
What does it cost?+
Healthcare builds are priced by scope, so we quote after the discovery call. The call is free.
A HIPAA-compliant clone of your practice
We sign a BAA, build on HIPAA-eligible infrastructure, and train it only on your approved material. Tell us about your practice and we'll show you what it would cover.