PHI-Safe Prompting: How Dental Professionals Can Use AI Without Putting Patient Data at Risk
- Kyle Summerford
- Mar 23
- 3 min read
Updated: Mar 24
Dental teams across the country are already using AI. They are asking ChatGPT to help write patient letters. They are using Claude to draft insurance appeals. They are running team meeting agendas through Gemini. And most of them are doing it in ways that create real HIPAA compliance risk without realizing it.
PHI-safe prompting is the skill that closes that gap. It is the foundational AI competency that every dental professional needs, and it is the signature differentiator in everything Kyle Summerford teaches about AI in dental practice management.
What Is PHI-Safe Prompting?
PHI stands for Protected Health Information. Under HIPAA, PHI includes any information that could identify a patient and is connected to their health, treatment, or payment history. Names, dates of birth, treatment codes, insurance information, appointment dates tied to a patient identity, all of these are PHI.
General AI tools like ChatGPT, Claude, and Gemini are not covered entities under HIPAA and do not automatically have Business Associate Agreements in place with your practice. Entering patient information into these tools without a BAA creates potential HIPAA exposure. PHI-safe prompting teaches you how to get the full benefit of these tools without ever putting patient data at risk.
The Core Rules of PHI-Safe Prompting
Never enter identifying patient information into a general AI tool. This means no patient names, dates of birth, insurance IDs, treatment histories, or any other information that could be traced back to a specific individual. The AI does not need this information to help you do your job.
Use role-based and scenario-based prompting instead. Instead of entering a specific patient situation with identifying details, describe the situation in general terms. Instead of writing a letter for John Smith with Delta Dental who had a crown on tooth 14, write a letter template for a patient with a major restorative procedure whose insurance paid less than expected. The AI gives you a template you then personalize outside the tool.
Build templates, not individual communications. PHI-safe prompting shifts your AI workflow from creating individual patient communications to building reusable templates that your team personalizes manually. This approach is actually more efficient, because one well-built template serves hundreds of patients instead of one.
Always review AI output before using it. AI tools make mistakes, misunderstand context, and sometimes produce language that does not fit your practice's tone or situation. A human review step before any AI-generated content reaches a patient or insurance company is non-negotiable.
What PHI-Safe Prompting Looks Like in Practice
A dental office manager needs to write an insurance appeal for a crown that was denied as cosmetic. The PHI-unsafe approach is to paste the patient's explanation of benefits with their name, ID number, and treatment details into ChatGPT and ask it to write an appeal. The PHI-safe approach is to ask the AI to write a template appeal letter for a crown procedure that was incorrectly denied as cosmetic, including the clinical justification language that typically supports medical necessity for this procedure. The manager then adds the specific patient and insurance details manually before sending.
Same result. Zero compliance risk. And the template the manager just created can be used for every similar appeal going forward.
Where to Learn PHI-Safe Prompting
PHI-safe prompting is a core module in the DOMA AI Certification and a central component of every AI speaking engagement Kyle delivers. If you want to build this skill inside your practice, the certification is the most comprehensive path available.
Kyle also teaches PHI-safe prompting at DOMA Academy Live events and can deliver this content as a standalone workshop for dental organizations, DSOs, and associations looking to build AI competency across their teams safely.
Comments