The slow part of a treatment plan is not the dentistry. It is explaining the plan you already decided so the patient understands why the crown, why now, and what happens if they wait, in language that informs rather than sounds like a sales pitch. That translation is exactly the kind of work AI is good at. The trap is that the same tool will happily read a radiograph and suggest a diagnosis, propose a procedure you never recommended, or invent a billing code, and the plan goes to the patient under your name. This is a plain look at where AI genuinely helps with patient explanations, where it does not, and a set of prompts you can paste in and test on your own cases.
One ground rule before any of it: patient records are PHI. Keep identifiers out of a public chatbot, test with de-identified plans, and for real charts use a private tool that keeps the data inside your own systems. More on that at the end.
The honest picture
AI is genuinely useful for turning the plan you made into clear prose:
- What AI does well today: take the diagnosis and treatment you decided and write a plain-language explanation a patient understands, the reason for treating now in your words, a readable visit-by-visit sequence and cost summary from the figures your office entered, all at a calm reading level that supports case acceptance without overselling.
- What AI does not do: diagnose, read radiographs or photos, choose a procedure, or assign a code. It cannot examine the patient, and it cannot sign the plan. The diagnosis and the plan are yours and you are accountable for them. AI can explain a decision you made; it must never read an image to find a problem, recommend a treatment, or generate a CDT or insurance code.
The right way to think about it: AI is a fast writer working from your plan, not a second clinician. The calls are yours. The writing is what you hand off.
The line: it will diagnose from an image if you let it
The specific failure to watch for is fabrication, and it shows up three ways:
- From an image. It is tempting to upload a radiograph and ask "what do you see." That is the exact line not to cross. A model guessing decay or bone loss from a picture is making a diagnosis, and the patient relies on the plan as if a licensed dentist examined them. Use AI to explain the plan you made, never to read the film.
- From a thin note. Give it "#19 needs crown" and it may write a confident paragraph about the tooth's condition and prognosis you never assessed. Plausible clinical detail you did not diagnose is a finding you did not make.
- From a code request. Never ask it to assign or "look up" a CDT or insurance code. A guessed code is a billing problem waiting to happen. Codes come from your team, not the model.
The fix is the same in every case: AI writes only what your plan states, with the fees and procedure names your office entered, and flags anything it cannot tie to your plan rather than filling it in.
The setup that keeps the plan yours
Three habits make AI much safer here, and the prompts below build them in:
- Give it your diagnosis, your proposed treatment, and your office's figures. Tell it to use only those, copy fees and procedure names verbatim, and never add a procedure, a diagnosis, or a code.
- Make it flag, not fill. If something is missing, it lists it under "needs my input" and asks, rather than inventing the rest.
- Keep the data yours. De-identified plans for testing in a public tool; real patient charts only in a tool that stays inside your own systems.
How to test it on your own work
Do not trust a polished demo, including this one. Pull two or three of your own recent plans, de-identified, with the diagnosis and treatment you decided and the fees your office entered, a timer, and the prompts below. Rate each output 1 to 5 on usefulness and accuracy, and compare the time against how you write patient explanations today. Keep what wins.
Paste-ready prompts
Copy these as written. Bracketed text is what you swap per case.
Test 1: Explain the proposed plan to the patient (text model)
I am giving you the diagnosis and treatment plan I already decided for one
patient, plus the fees and notes my office entered. Write a plain-language
explanation the patient can read. Rules:
- Use only the diagnosis and treatment I wrote. Do not add a diagnosis, propose a
different or additional procedure, or interpret any image. If something is
missing, list it under "Needs my input" and ask, do not fill it in.
- Copy all fees and procedure names exactly as entered. Do not generate or guess
any insurance or CDT code.
- Write at about a 7th-grade reading level, honest and calm, not a sales pitch.
Diagnosis and treatment plan: [paste]
Fees and notes (as entered): [paste]
Watch for: did it add a procedure or read into the case, or stick to the plan you set? Every added procedure and guessed code is the work it cannot do for you.
Test 2: "What happens if you wait" in your words (text model)
Here is the reason I gave for treating now and what I told the patient about
waiting: [paste]. Write two or three plain sentences explaining what untreated
looks like, using only my stated rationale. Do not add a prognosis, a statistic,
or a consequence I did not write, and do not exaggerate the urgency.
Watch for: did it invent a scary statistic, or keep to the reasoning you gave?
Test 3: Sequence and cost summary (text model)
Turn the treatment plan, visit sequence, and fees my office entered into a clear
summary for the patient: what happens at each visit and the cost of each step,
with the total. Use only the figures and procedure names I provide. Do not add
procedures, do not estimate a fee I did not give, and do not generate any code.
Plan, sequence, and fees: [paste]
Watch for: did every number match what your office entered, or did one drift?
Test 4: Consistency audit (text model)
Review this draft patient explanation against my source plan for internal
problems only. For each issue, quote the exact line and say what is wrong.
1. Any procedure, diagnosis, or clinical claim in the draft not present in my
plan.
2. Any fee or total that does not match the figures I entered.
3. Any generated code, or any language that reads as interpreting an image.
Do not fix or add anything. Only flag.
My plan and fees: [paste]
Draft explanation: [paste]
Watch for: a procedure or fee that drifted from what you entered. Run it on an explanation you already gave a patient.
What success looks like, and where it could go
If your own testing shows real time savings, the next step is a small pilot: run a week of plans through the prompts and measure the minutes saved per case and whether patients understand the plan better. If that holds up, the natural next step is a simple agent, running on your own practice's cloud, that you use in plain language. The most useful version takes your diagnosed plan and your office's fees, writes the explanation at a fixed reading level, copies every fee and procedure name verbatim, never touches images or codes, and cites the plan line behind each sentence, surfacing anything it cannot tie to your plan. You read, correct, and hand the patient a clear page in a couple of minutes. Because it runs in your own systems, PHI never leaves for a public chatbot.
The principle holds the whole way through: AI gives you a faster draft of the explanation and a second set of eyes on consistency. It does not read the radiograph, choose the treatment, or assign the code. The dentist diagnoses every case; the writing is what you hand off.
This is general information about workflow tools, not clinical, billing, or compliance advice.
Want a straight answer for your practice?
I build practical AI and custom software for businesses, on Google Cloud. If you want a second set of eyes on how AI could fit your practice's workflow, or on a tool you are considering buying, tell me what you are working with. No pitch, just a straight answer.