If your team is already using ChatGPT on their phones for work, you don’t have an AI strategy. You have an unmanaged one. The goal isn’t to ban the tools. It’s to give people a clear path that doesn’t create risk.
@w_btariq The clinic is still responsible for how they use it. But if your platform is the one generating the agent and handling the calls, you’re part of the data flow. That usually means a BAA is still required on your side, not just the clinic’s.
HIPAA-safe AI workflow for a front desk: Staff writes the message in plain language with no patient identifiers. AI cleans up the wording. Staff adds the patient name and sends it from the practice system. The AI never sees the name.
@TimBorovkov Sovereign is the right long-term structure. Most practices aren’t there yet though. The practical line right now is still “don’t send audio or identifiers into a consumer model at all.” Scheduling and forms can stay useful without crossing it.
@IsharSandhu Capturing the basic details is fine. The moment the agent asks “how bad is the pain” or logs a symptom, you’re in PHI territory and need the proper vendor setup.
@MohsinFBA The 10-hour number is real for reminder and intake work, but only if the agent is behind a BAA and doesn’t dump transcripts into ChatGPT. That’s the difference between useful and risky.
@SmashingApps Agree that the hard part isn’t the agent itself — it’s knowing which steps can touch PHI and which must stay human. Offer a short checklist of the first three things to automate safely.
@OsteoBunni@precisionfires1@dfeuling_ That's fair, and honestly it's the right setup. The pharmacy's compliance team should make that call, not each tech. If they ever do approve something, ask what it's cleared for before you put anything with a patient name into it.
A lot of “AI tools for dentists” are just wrappers around ChatGPT. That’s fine if you’re careful. It’s a problem if the tool is storing the full prompt and you didn’t notice. Ask the vendor one question: “Do you train on or retain the data we send?”
Common request I hear: “Can we use AI to write our recall messages?” Yes. But don’t paste real patient names or phone numbers into the tool to do it. Write the template first. Merge the names later, outside the AI.
Looking for clinics that already have public signals of friction (phones not answered, no online booking, mixed reviews about wait times) is a reasonable research approach. When you turn that into an offer, the HIPAA conversation should happen early. An AI receptionist or follow-up system that touches patient names, phone numbers, or appointment history needs a BAA and a clear data-flow map. Most dental owners will ask about that once they see the demo; having the answer ready shortens the sales cycle.
Unanswered phones cost real money, and an AI receptionist can close that gap after hours. For dental specifically, the implementation details matter: does it book into the real schedule, or does it just take a message? Can it confirm insurance without storing the call? The practices that do this cleanly usually start with new-patient and recall calls only, then expand once the team trusts the handoff process. Demo-first is the right sales motion.
Patients asking ChatGPT “who’s a good implant dentist near me” is already happening. The practices that show up tend to have clear procedure pages and consistent local signals. On the HIPAA side, the risk is usually internal: staff using AI to rewrite a review response or a treatment-plan email and accidentally including chart details. Public marketing AI is low risk; anything that touches an individual patient’s information is not. Separate those workflows on purpose.
Sample calls are smart. Dental offices hear the same questions every week—hours, insurance, new-patient availability, “do you take my plan.” An AI that can handle those cleanly reduces front-desk load. The HIPAA line is usually crossed when the caller starts describing symptoms or asking about a specific existing patient. Clear handoff rules (“I’m transferring you to the office for that”) keep the AI useful without turning it into an unauthorized clinical conversation.
“It depends on the deployment, not the tool” is the sentence most dental teams need to hear. Once PHI is in the prompt, the tool is effectively part of your covered entity’s operations. I tell practices to treat the BAA as the floor, not the ceiling. Then look at retention, subprocessors, and whether staff can accidentally use a personal account. The architecture decisions matter more than the vendor’s homepage claim.
Zero-retention + PHI recognition is the right direction for voice tools. For dental practices, the next question is usually about the front-desk use case: after-hours calls, confirmation texts, or recall. Even with a compliant platform, the script and the escalation rules still matter. An AI that correctly identifies PHI but still promises clinical advice creates a different problem. Technical compliance and clinical scope are separate controls.
AI receptionists that write straight into the schedule can be useful, but the integration is where the HIPAA questions live. Before you demo any of these, ask three things: Do they sign a BAA? Can you turn off any training on call audio or transcripts? What happens to the data if you cancel? Open Dental integrations make the operational case easy; the compliance case still needs to be verified in writing for that specific configuration.
You don’t need a full AI policy on day one. Start with three things: what staff can paste into AI tools, what they can’t, and where the output is allowed to go. That’s enough to stop the most common mistakes.4