Somebody at your front desk has probably already tried it. A patient sends a long, upset email, and a staff member pastes it into ChatGPT to help write a calm reply. It works beautifully. It may also have just sent protected health information to a company your practice has no agreement with.
Owners of dental offices, physical therapy clinics, chiropractic practices and med spas ask the same handful of questions about this: Is ChatGPT HIPAA compliant? Do I need a BAA for an AI tool? If we take the name out, is it OK? Which AI jobs are safe to start with? This guide answers each one in plain language, using guidance from the U.S. Department of Health and Human Services (HHS). It’s general information, not legal advice. For decisions about your own practice, talk to your compliance advisor or healthcare attorney.
The short answer: patient data plus an outside tool means a BAA
If an AI tool will create, receive, store or send protected health information (PHI) for your practice, the company behind it is a business associate. HHS defines a business associate as someone who performs work on your behalf “that involve creating, receiving, maintaining, or transmitting PHI.” Before any PHI goes in, you need a signed business associate agreement (BAA) with that company.
The BAA spells out what the vendor may do with the data and requires it to protect the data and report security incidents to you. No BAA, no PHI. The rest of this article is about applying that rule.
PHI is more than the chart
Many owners picture PHI as X-rays and treatment notes. It’s broader: information that connects a person to your practice and their care is PHI, including everyday front-desk details.
HHS’s de-identification guidance lists 18 types of identifiers that must be removed under its “Safe Harbor” method before data stops counting as PHI. They include:
- Names
- Street addresses, cities and ZIP codes
- All dates except the year, such as birth dates and visit dates
- Phone numbers and email addresses
- Medical record and account numbers
- Full-face photos
So a list of patient phone numbers with appointment times is PHI. So is a patient’s email asking to reschedule. Deleting the name before pasting a message into a chatbot usually isn’t enough, because the email address, the date or the details in the story can still identify the person.
“HIPAA compliant” on a website isn’t a BAA
Plenty of AI tools say “HIPAA compliant” in their marketing. What matters is whether the company will sign a BAA for the exact product and plan you use.
HHS is strict about this. Its cloud computing guidance says a cloud provider that handles electronic PHI 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.” Encryption is good practice, but it doesn’t replace the agreement.
Which popular AI tools offer a BAA?
This changes often, so check the vendor’s current page before you rely on it. As of November 2026:
- ChatGPT: OpenAI’s BAA help article says, “Currently, we do not offer a BAA for ChatGPT.” OpenAI does offer a BAA for its API platform, which developers use to build tools, and only for API endpoints eligible for zero data retention.
- Claude: Anthropic offers a BAA for its API and for Enterprise plans once HIPAA coverage is turned on by the account owner. Consumer plans aren’t covered, and some features are excluded.
- Google Workspace with Gemini: Google’s HIPAA page says customers without a signed BAA “must not use PHI” in Workspace, and that certain Gemini features in Gmail, such as Help me write and smart replies, are covered as HIPAA included functionality.
- Microsoft 365 Copilot: Microsoft says Copilot follows the existing Microsoft 365 compliance commitments, including HIPAA. If your practice already has a BAA with Microsoft, confirm with Microsoft or your IT provider that Copilot is covered for your plan.
The pattern: free and personal accounts are out for patient data. Business tools can be in, but only with a signed BAA and the right settings. (For the non-health side of the same question, see our guide on whether it’s safe to put customer data into ChatGPT.)
AI jobs that don’t need patient data at all
The easiest way to stay safe is to start where there’s no PHI. These are useful, and none of them require patient information:
- Writing blog posts, social posts and newsletters
- Drafting templates for reminders, review requests and recall messages (with placeholders, not real patients)
- A website assistant that answers general questions: hours, parking, insurance accepted, new-patient paperwork
- Turning your office procedures into a training manual
- Job postings and interview questions
Note the difference between drafting a template and sending it. Writing the review-request wording is PHI-free. The system that texts it to real patients is not, so that system needs a BAA.
AI jobs that touch PHI (BAA required)
These often save the most time, and they’re worth doing with the right vendor:
- Appointment reminders, rescheduling and waitlist fill
- Missed-call text-back to callers and existing patients
- Online intake forms and insurance details
- AI note-taking or scribe tools
- Recall and lapsed-patient reactivation campaigns
- Drafting replies to patient emails and portal messages
HHS allows providers to communicate electronically with patients “provided they apply reasonable safeguards,” according to its email FAQ. The automation itself isn’t the problem. The vendor relationship and the safeguards are.
A five-step check before you turn on any AI tool
- List the data it touches. Write down exactly what goes in and what comes out. If any of it is on the identifier list above, treat it as PHI.
- Get the BAA first. Confirm it covers the specific product, plan and features you’ll use, and ask how the vendor’s own subcontractors are covered.
- Set it up on purpose. Use a business account, limit who has access, and check the data retention and model-training settings.
- Add it to your risk analysis. HHS’s risk analysis guidance calls the risk analysis “foundational.” A new AI tool that handles PHI belongs in it.
- Train your team and keep a person in the loop. Give staff a one-paragraph rule: which AI tools are approved, what may never be pasted into anything else, and who to ask. AI drafts; a person reviews before anything goes to a patient.
That last step is where most practices slip. When Hub Media rolled out AI tools across a client’s whole company, every department got a permissions and data audit before launch, and training happened in small groups of five or fewer so people could ask real questions. The same approach works for a three-chair dental office.
What’s changing
HHS proposed a major update to the HIPAA Security Rule in December 2024, with stricter requirements for things like security planning and technology inventories. As of this writing a final rule hasn’t been issued; HIPAA Journal reports that the federal regulatory agenda now targets July 2027. Florida practices should also know the state’s own data breach law, the Florida Information Protection Act (Fla. Stat. 501.171), which sets its own notice rules. Either way, the habits above (BAAs, a current risk analysis and trained staff) are what regulators already expect.
How Hub Media helps practices use AI safely
Hub Media has worked with more than 80 businesses since 2013, including healthcare practices such as The Dental Lounge, where we ran lead acquisition. For a multi-location spa and wellness membership business, we built AI-assisted matching of member records to service usage across systems. Our rule for healthcare work is the one described here: start with PHI-free wins, and put patient data only into tools covered by a signed BAA.
If you run a dental, physical therapy or chiropractic practice, or a med spa or IV lounge, our AI implementation work starts by sorting your ideas into “safe today” and “needs a BAA first.”
Not sure which AI jobs are safe for your practice?
Our AI Opportunity Audit is a short, fixed-scope review of how your practice gets and handles leads and admin work. It ends with 2–3 recommended automations, with a clear note on which ones involve patient data and what each would need. Request your audit.
Frequently asked questions
Is ChatGPT HIPAA compliant?
Not for patient data in its standard plans. OpenAI’s help center says it does not currently offer a BAA for ChatGPT, so practices shouldn’t put protected health information into it. OpenAI offers a BAA for its API platform, which developers use to build tools. Vendor terms change, so check the current page before relying on this.
Do I need a BAA to use an AI tool in my practice?
Yes, if the tool will create, receive, store or send protected health information for your practice. HHS treats that vendor as a business associate, and a signed business associate agreement is required before any patient data goes in. Tools that never see patient data, like one that drafts blog posts, don’t need one.
If I remove the patient’s name, can I paste their message into ChatGPT?
Usually not. HHS’s Safe Harbor method lists 18 types of identifiers, including dates, phone numbers, email addresses and record numbers, that must all be removed before data stops being PHI. A message with the name deleted often still identifies the patient.
What can a practice use AI for without patient data?
Plenty: blog and social posts, templates for reminders and review requests with placeholders, a website assistant for general questions like hours and insurance accepted, training manuals and job postings. Sending those messages to real patients is a separate step that needs a vendor with a BAA.
If an AI vendor encrypts our data, do we still need a BAA?
Yes. HHS guidance says a cloud provider that handles electronic PHI is a business associate even if it stores only encrypted data and has no encryption key. Encryption is a good safeguard, but it doesn’t replace the agreement.

