ChatGPT for Chiropractors: What Is Safe Without a BAA?

Without a BAA (business associate agreement), a chiropractic clinic can use ChatGPT for work holding no patient information: marketing posts, generic exercise handouts, policy and job-advert drafts, template emails and staff training. Anything containing protected health information, such as notes, names with appointments or report text, needs a tool whose vendor has signed a BAA, set up as the vendor requires.

The detail that surprises many owners is that the business plan is not the answer. OpenAI's help centre says it does not offer a BAA for ChatGPT Business, which is the plan clinics most often choose because it is shared, billed per seat and does not train on your content by default. A BAA is a different question from training: a vendor can promise not to train on your chats and still not have signed the contract HIPAA expects before patient information goes in.

Follow me on Instagram@sagnikteaches

What a BAA is and why privacy settings do not replace it

If your clinic is a covered entity under HIPAA, a vendor that creates, receives, maintains or transmits protected health information (PHI) on your behalf is your business associate. HIPAA requires a written agreement, the BAA, setting out what the vendor may do with that information and what safeguards it must have. Without one, putting PHI into the vendor's tool is a compliance problem however good the tool is.

Connect on LinkedInSagnik Bhattacharya

Three things people mistake for a BAA:

Subscribe on YouTube@sagnikteaches

• "We don't train on your data." A promise about model training says nothing about security obligations, breach notification or permitted uses.
• "It's encrypted." Encryption is one safeguard. The BAA is the contract that makes the vendor accountable.
• A business-tier plan. Plan names do not determine HIPAA coverage. OpenAI lists its HIPAA-eligible products separately.

OpenAI's own HIPAA page, as checked in November 2026, lists ChatGPT for Healthcare, ChatGPT for Enterprise with a Regulated Workspace, ChatGPT for Clinicians, ChatGPT FedRAMP and the API with modified retention as available with a BAA. Its separate help article says eligible organisations can accept a standard BAA for the API in a self-serve flow (which requires an established history of API usage), that a BAA for a sales-managed ChatGPT Enterprise or Edu account goes through sales, and that eligible individual clinicians on ChatGPT for Clinicians have their own in-product BAA flow. Product names and eligibility change, so read the HIPAA eligible products page on the day you decide. Legal questions about whether your clinic is a covered entity, or what your agreement covers, belong with your compliance adviser.

The safe list: jobs with no patient information

The test is whether the prompt, the files you attach and the output you expect contain anything about a real patient. If they do not, a clinic can use ChatGPT like any small business. These jobs fit.

Marketing and local content

Clinic web pages on posture and desk work, social posts, seasonal messages, newsletter copy and website pages all use general knowledge. The care needed is accuracy and claims: do not let a draft say a treatment "cures" anything, and check any health statement against a source you trust. A prompt:

Write a 250-word newsletter item for a chiropractic clinic about simple desk-posture habits. Plain English, friendly. No claims that treatment cures or prevents any condition. End with an invitation to book a consultation. Do not mention any patient.

A typical draft (illustrative) offers five practical habits and a soft call to action. What you fix: remove any sentence that promises results, and check any stated time interval, such as "stand every 30 minutes", is something you are comfortable giving as general advice.

Generic patient education

Handouts on how to sit, stretching ideas, what to expect at a first visit or how to use an ice pack are general. Keep them general, and have a clinician approve them. A handout tailored to one patient's condition is no longer generic.

Policies, templates and the business side

Cancellation policy wording, a new-starter checklist, job adverts, interview questions, a front-desk phone script and staff meeting agendas have no PHI. The same goes for questions about spreadsheets, pricing models and marketing plans. Our tutorial on AI receptionist costs for a chiropractic office is the next stop if phone handling is the pain.

Reply templates, not replies to real people

Build a bank of generic replies: "How do I book?", "Do you take my insurance?" (answered from your own page), "How long is a first visit?". Do not paste an actual patient message. When you answer one, add the details yourself.

Staff training and learning

Using ChatGPT to explain a spreadsheet formula, draft a training quiz about your own procedures, or role-play a difficult phone call with an invented caller is fine. Invent the caller entirely; a real patient's story with the name removed is not invented.

The never-paste list, for the front desk and treatment room

Notice that every row is something a conscientious person does to save time. A never-paste list works because it names those specific moments. Print it, pin it up, and add your clinic's own recurring examples.

One clinic, one week of ChatGPT requests

Suppose a three-chiropractor clinic with two front-desk staff runs a ChatGPT workspace for admin. Here are ten requests that came up in a week, with the decision on each. (An illustrative scenario.)

Four of the ten were safe as asked, four were safe with a change of approach, and two belonged elsewhere. The spreadsheet row is the sneakiest: a column of names three tabs away from the formula is still PHI if it is uploaded. For general spreadsheet help, see our ChatGPT and Excel guide, and build a dummy file with invented rows before asking about a formula.

De-identifying: harder than deleting the name

Many clinics assume that removing the name makes a note safe. HIPAA's de-identification standard is stricter. The Safe Harbor method requires removal of eighteen categories of identifiers, including names, geographic subdivisions smaller than a state, all elements of dates except the year, phone numbers, email addresses, medical record numbers, full-face photographs and any other unique identifying number, characteristic or code, and the clinic must have no actual knowledge that the rest could identify someone. "A 54-year-old roofer who attended on the 3rd of March with right shoulder pain" fails on several counts and, in a small town, may identify a person to anyone who knows them.

A practical rule for a busy clinic: do not rely on de-identifying real notes as a daily habit. Instead, invent the scenario. If you want to see how a model might structure a patient letter, write an imaginary patient with a different age, job and complaint, and use that. If a task truly needs real patient content, the answer is a covered tool, not cleverer editing. If you do believe a dataset meets the standard, have your compliance adviser confirm it in writing first.

When the work does involve PHI: three routes

1. Use the tools already inside your practice system. Some chiropractic EHRs now build AI into the record itself; ChiroTouch, for example, describes its Rheo assistant as automating SOAP notes inside its EHR. An EHR vendor that holds your PHI should already be under a BAA, so ask in writing whether that agreement covers the AI feature and any outside AI provider it relies on. Our tutorial on what chiropractic EHR AI features really do compares them. This is usually the lowest-effort route because the data never leaves the system you already trust.
2. Use a scribe or assistant whose vendor signs a BAA. Ask for the signed agreement before you record a single visit, and ask the questions in our vendor security questions. For therapy-style clinics, this list of vendors who sign BAAs shows what that looks like.
3. Use a covered OpenAI product. If you want ChatGPT specifically, the eligible products are the ones on OpenAI's HIPAA page, and a clinician-focused offering is one of them. Eligibility, price and setup requirements apply, and a BAA is only the start: the workspace has to be configured as the vendor's HIPAA guide describes. Treat it as a project with an owner, not a switch.

Whichever route, put the agreement in your compliance folder, record who may use the tool for what, and tell staff plainly which tool is for patient work and which is not. It is far safer for a clinic to have one covered tool for PHI and one general tool for everything else than to hope that everyone remembers the difference.

How the ChatGPT plans differ for a clinic

Plan details change often, so use this as a map of questions rather than a price list. Check OpenAI's pricing and help pages on the day you decide.

For a typical three-chiropractor clinic, the sensible pattern is two tools: a two-seat business workspace for non-patient work, and the AI inside the practice system or a covered scribe for anything clinical. Do not try to make one tool do both jobs by loosening the rules.

Before and after: a review reply that stays safe

Review replies are where owners most often leak PHI without noticing, because the reviewer's post feels public. Suppose a reviewer writes, "I had five visits and my neck is no better. Wasted money." The unsafe reply, drafted by someone in a hurry, says: "We are sorry, [name], we reviewed your five visits and our notes show you missed your last two appointments." That confirms the person was a patient and discloses treatment information.

The safe route is to ask ChatGPT for a generic reply without the review text or any details:

Write a short, courteous reply for a chiropractic clinic responding to a negative online review. Do not confirm the reviewer was a patient and do not mention any treatment, visit or record. Say we take feedback seriously and invite them to contact the clinic directly on [number or email]. Under 70 words.

An illustrative draft: "Thank you for taking the time to share your feedback. We take every comment seriously and would like to understand more. Please contact our clinic manager on [number] so we can talk this through privately." The owner checks it for any accidental confirmation, adds the real contact, and posts it. Details of the person's care are never discussed in public, and are never typed into an unapproved tool.

A filled-in one-page policy to adapt

This is the shape of a policy a small clinic can adopt this week. The bracketed items are decisions for you.

AI USE POLICY - [Clinic name] - version 1, [date]

1. Approved tools
   - [ChatGPT Business workspace]: general admin and marketing, NO patient information.
   - [Practice system AI feature / covered scribe]: patient notes, under the signed BAA.
2. Never paste into any other tool
   Names with care details, notes, findings, appointment lists, insurance details, photos, patient messages.
3. Testing and learning
   Use invented scenarios only.
4. New tools
   Only the owner signs up for tools. Ask before you try one.
5. If you slip
   Delete the chat, tell [owner] the same day, write down what was pasted.
6. Review
   Monthly prompt check; policy reviewed every [6] months.

Signed: ____________ Date: ______

Walk through it with the whole team for ten minutes and ask each person for the task they would be most tempted to shortcut. Add those to the never-paste table. Expect answers such as insurance letters, review replies and "just summarising" a note, which is why those already appear above.

Claims and accuracy in clinic marketing

Even the safe jobs carry a different risk: health claims. A draft may state that spinal adjustment "relieves migraines", "corrects" posture or "boosts immunity" because such phrases appear across the web. Advertising rules for health claims are set by regulators and professional bodies, and a claim you cannot support is your problem, not the tool's. Give the model a rule such as "do not claim to treat, cure or prevent any condition; describe what the clinic offers and what a visit involves", and then read for verbs like "fixes", "heals" and "guarantees". Have a clinician sign off anything that describes a condition. For a wider view of AI-drafted marketing in a similar clinic, see marketing a physio clinic with AI.

When patient details slip into a chat anyway

Even with a policy, someone will eventually paste the wrong thing. What matters is what the clinic checks in the first hour, because a chat is not always as private as it feels. Three places to look before you delete anything:

• Shared links. If the chat was shared with a link, that link has to be removed too. OpenAI's data controls let a user review and manage shared links.
• Shared Projects. If the chat sits inside a shared Project, other members can see and use it. Delete it there and tell the members it existed.
• Memory. If memory is on, check whether anything from the chat was saved as a memory, and remove it.

Whether the slip counts as a reportable breach is a question for your compliance adviser, not for the person who made it. Your job on the day is to record the facts: what was pasted, into which account, when, and what was deleted.

A sample slip, handled

Suppose a receptionist pastes a patient's appointment history into a free ChatGPT account to write a recall message, then realises. The steps: delete the conversation and check whether history or memory kept anything; tell the owner at once; write a dated note of what was pasted and into which account; ask the compliance adviser whether it needs further action; and retrain without blame. Do all of this the same day. The slip is the signal that the template route (placeholders, then merge in the practice software) needs to be easier than the shortcut.

A final practical point for owners who share the workspace: name the shared Project clearly ("Admin and marketing, no patient data") and put the never-paste list at the top of its instructions, so the rule is the first thing anyone sees when they open a chat. Small cues like that do more than an annual training session.

Checking the line is holding

A shared admin Project makes the check easy, because everyone who has access can see the chats in it. Once a month, open the Project, read the twenty most recent chats and note any names, dates, phone numbers or text pasted from a patient email. Then ask each person, briefly, whether they used a personal account for clinic work that month. Count how many chats used placeholders, and ask what the team finds slow about the safe route. If staff keep reaching for real notes because the template route is clumsy, fix that, because a rule that is awkward in practice does not last. If the clinic wants a mapped plan, that is the kind of work a 1:1 call is for, via the contact page.

This is practical guidance, not legal advice. Whether and how HIPAA applies to your clinic, and what your agreements cover, is a matter for your compliance adviser and the vendor's current documents.

Chiropractors' questions about ChatGPT and BAAs

Does ChatGPT Business give me a BAA?

No. OpenAI's help centre says it does not offer a BAA for ChatGPT Business. Plans that can be covered are listed on OpenAI's HIPAA page and include ChatGPT for Healthcare, ChatGPT for Clinicians and enterprise workspaces set up for regulated use. Check that page for the current list before buying.

Is a patient's first name PHI?

A name tied to health information, appointments or treatment is PHI. Do not rely on 'first name only'. Dates, phone numbers, email addresses and many other details also count. When in doubt, treat the text as PHI and keep it out of any tool without a BAA.

Can I paste de-identified notes into ChatGPT?

Only if they genuinely meet HIPAA's de-identification standard. Under the Safe Harbor method that means removing all eighteen listed identifiers and having no actual knowledge that the remaining text could identify the person. A paragraph with the names deleted often still fails. Ask your compliance adviser before treating anything as de-identified.

Does switching off training make ChatGPT HIPAA-safe?

No. Turning off training is a privacy setting. Whether a tool is appropriate for PHI depends on having a BAA and the right configuration. A tool without a BAA stays off limits for patient information whatever the settings.

Can my staff use their own ChatGPT accounts?

They can for tasks with no patient data, if your policy says so. Personal accounts give you no admin view, so write the never-paste list into the policy and review it with staff. For any PHI work, use only the tool you have signed a BAA for.

Further reads

• AI Software for Chiropractors: A Data Export Checklist Before You Switch — Moving between chiropractic systems without losing data.
• Is It Safe to Put Customer Data Into ChatGPT? — The general customer-data rules behind this clinic list.
• Does ChatGPT Remember Conversations? Memory Settings for Business — Memory and history settings that affect what lingers.
• Is an AI Scribe Worth It for a Small Private Clinic? — Whether a covered scribe pays for itself in a small clinic.
• AI Tools for Physiotherapy Clinics: What Each One Actually Does — What each AI tool does in a close sibling setting.
• HIPAA-Compliant AI Assistants for a Medical Office: BAA by Plan — Which plan of ChatGPT, Claude, Gemini and Copilot can carry a BAA for a small practice, what each excludes, and a tracker to record it.
• AI Marketing for Chiropractors: Claim Limits and Review Replies — Use AI for chiropractic posts and review replies without outcome promises: ASA and FTC limits, a claims register, safe reply wording and prompts.
• AI Tools for Chiropractors: Report of Findings, Care Plans and Handouts — Draft a plain-language report of findings, a care-plan summary and exercise handouts with AI, while keeping outcomes unpromised and every word reviewed.
• Is ChatGPT HIPAA Compliant? Free, Business and Clinician Plans — Which ChatGPT plans can sign a BAA, which cannot, and the do-not-paste rules to give your front desk and clinicians.
• Is Gemini HIPAA Compliant? What a Workspace BAA Covers — The Workspace BAA covers the Gemini app and Gemini in Workspace apps, not Gemini in Chrome. Here is the scope and the admin step that comes first.
• AI Notes for Chiropractors: Consent and Data Checks Before Recording — Before an AI note tool records a chiropractic visit: the consent wording, the data terms to check by plan, an opt-out workflow and the clinician review step.
• AI Search Tools for Medical Providers: How to Check the Citations — A five-step routine for checking an AI search answer's citations before you act, plus a look at the tools that cite and what never goes in a prompt.
• Chiropractic AI for Clinic Owners: Which Admin Jobs to Start With — A clinic owner's order of play for AI: bookings and recalls first, notes with review, marketing within claim limits, no AI diagnosis.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI Help Center articles on Business Associate Agreements, HIPAA eligible products, data controls and Projects; OpenAI ChatGPT Business pricing page; HHS guidance on de-identification and business associate contracts; ChiroTouch AI assistant page. Checked November 2026.