Dentistry's most visible AI is clinical — FDA-cleared radiograph analysis from companies like Overjet and Pearl. But the AI most dentists actually touch every week is less glamorous: drafting insurance narratives, cleaning up clinical notes, and writing the patient letters nobody has time for. Roughly 35% of US practices have implemented some form of AI, and about 77% of those report a positive impact — yet only around 22% of dentists use AI tools weekly, so most of the gains are still on the table.
The pattern in practices that stick with it is consistent. General-purpose tools like ChatGPT and Claude handle de-identified writing work — claim narratives that used to take 15-25 minutes now take under 3, and a full post-op letter library gets built in a week — while BAA-covered, purpose-built tools handle anything that touches the actual chart.
The guardrails matter as much as the wins. Consumer ChatGPT accounts do not come with a Business Associate Agreement, so patient-identifiable information stays out of them, full stop. The ADA has started formalizing the space — ANSI/ADA Standard 1110-1:2025 is the first US standard on AI in dentistry — and every AI draft, from a claim narrative to a review reply, becomes the dentist's own words the moment it is signed or posted.
About 35% of US dental practices have implemented some form of AI, and roughly 77% of those report a positive impactSource ↗
Dentists using ChatGPT report treatment narrative writing time dropping from 15-25 minutes to under 3 minutesSource ↗
ANSI/ADA Standard 1110-1:2025, approved in 2025, is the first US standard on artificial intelligence in dentistrySource ↗
A 2025 study in the Journal of Medical Internet Research found dental radiology reports simplified with ChatGPT significantly improved patient comprehension and readabilitySource ↗
Claim denials for crowns, buildups, SRP, and implants eat hours of front-office time, and writing the narrative is the bottleneck — dental teams cite it as the single most time-consuming part of billing. Drafting narratives and appeals is the most widely reported AI win in dental offices, cutting a 15-45 minute task to a few minutes.
Prompt
You are an experienced dental billing coordinator who writes claim narratives that pass carrier review. Write a claim narrative for this procedure: {{procedure}}. Clinical findings from the chart (already de-identified): {{clinical_findings}}. Carrier context or denial reason, if any: {{carrier_notes}}.
Requirements:
- Keep the narrative under 150 words, in plain clinical prose (no bullets), third person ("the patient presented with...").
- Use only the findings I gave you. Do not invent measurements, dates, symptoms, or radiographic findings. If a fact this carrier typically wants is missing (probing depths for SRP, extraction date for a bridge, cusp involvement for a crown), insert [NEED: description] so my team can pull it from the chart.
- State clinical necessity plainly: what failed, why a lesser treatment is not adequate, and the prognosis without treatment.
- If I included a denial reason, write an appeal letter instead: one page, professional tone, quote the specific denial language, and rebut it point by point using only the documented findings.
- End with a short checklist of attachments this claim type usually needs (radiographs, perio chart, intraoral photos).
Fill in your details and the prompt updates live — then copy.
Dentists commonly finish the day with an hour of charting, and thin notes are both a top reason claims bounce and a liability problem if a board complaint lands. Turning quick operatory shorthand into a complete, structured note is one of the most repeatable AI wins — vendors in this space report saving clinicians 1-2 hours of documentation a day.
Prompt
You are a dental scribe who writes complete, defensible clinical notes. Convert my shorthand into a structured SOAP note for a {{visit_type}} visit.
My shorthand: {{shorthand}}
Format:
- S (Subjective): chief complaint in the patient's words, relevant history.
- O (Objective): clinical and radiographic findings, existing conditions.
- A (Assessment): diagnosis.
- P (Plan): treatment completed today — anesthetic type and amount, isolation, materials and shades, occlusion check, post-op instructions given — plus the next visit.
Rules:
- Use only what is in my shorthand. Where a standard element is missing (anesthetic carpules, isolation method, consent), write [VERIFY: element] instead of inventing it.
- Factual, past tense, no opinions about the patient.
- Standard dental abbreviations are fine, but expand anything ambiguous.
- After the note, list anything a claims reviewer or state board would expect for this visit type that is still missing.
Fill in your details and the prompt updates live — then copy.
Patients nod along at "D4341, scaling and root planing" and then decline at the front desk — poor understanding is a known case-acceptance killer. A 2025 peer-reviewed study found that dental reports simplified with ChatGPT scored significantly better on readability and patient comprehension, and dentists report explanations that took 15-25 minutes to write now take under 3.
Prompt
You help a dental practice explain treatment to patients. Rewrite the following treatment plan in plain language at a 6th-grade reading level, as a one-page patient handout.
Treatment plan (de-identified): {{treatment_plan}}
The patient's main concern: {{patient_concern}}
Structure:
1. What we found — the condition in everyday words. No CDT codes or jargon; if a technical term is unavoidable, define it in parentheses.
2. Why it matters — what happens if we wait, stated factually, no scare tactics.
3. What the treatment involves — visit by visit, including numbing, time in the chair, and how it will feel afterward.
4. Common questions — three short Q&As tied to the patient's stated concern.
Rules:
- Never promise outcomes ("this will fix it"); say what the treatment is designed to do.
- Do not add treatments, alternatives, or prognoses I did not list.
- Warm but direct tone. Under 350 words so it fits one page.
- End with: "Questions? Call us — we would rather explain twice than have you worry once."
Fill in your details and the prompt updates live — then copy.
The average practice sits at 40-50% recall compliance, and identical "you're due for a cleaning" blasts get ignored. Teams now use AI to write segmented, multi-touch sequences — perio maintenance patients get a different message than someone 18 months overdue for a checkup — then load them into their reminder system.
Prompt
You write patient recall messages for {{practice_name}}, a general dental practice. Write a 3-touch outreach sequence for this segment: {{patient_segment}}.
Touches:
1. Text message — under 160 characters, friendly, one clear ask, includes {{booking_method}}.
2. Follow-up text 2 weeks later — a different angle (health benefit, or use-your-insurance-benefits), not a repeat of touch 1. Under 160 characters.
3. Email — subject line under 45 characters, plus a 90-120 word body explaining in plain language why this specific recall matters for this segment (for example, why perio maintenance runs every 3-4 months, or what a year of missed checkups can hide).
Rules:
- No guilt-tripping ("we noticed you've been avoiding us"), no fake urgency, no exclamation-point pileups.
- No clinical or account details in the texts — a text can be read by anyone near the phone.
- Use [FIRST_NAME] and [DUE_DATE] as merge fields; never write real names.
- Include "Reply STOP to opt out" on the first text.
- Give me two variants of touch 1 to A/B test.
Fill in your details and the prompt updates live — then copy.
Prospective patients read reviews before booking, and practices that respond consistently earn several times more reviews — but a review reply is a public statement, and regulators have fined dental practices whose responses disclosed patient information. AI drafts the reply fast; the constraint list is what makes it safe.
Prompt
You are the office manager of {{practice_name}} responding publicly to a Google review. Draft a response.
Review ({{star_rating}} stars): {{review_text}}
Hard rules — these protect us legally:
- Never confirm or deny that the reviewer is a patient. No "thanks for visiting us" or "at your appointment."
- Never mention any treatment, date, tooth, payment, or insurance detail — even details the reviewer disclosed themselves.
- Never argue, correct their story, or explain what "really happened."
For positive reviews: thank them warmly in one or two sentences, reference something generic they praised (friendly team, gentle care), no medical specifics.
For negative reviews: acknowledge the frustration without admitting fault, state our general standard ("we aim for every visit to run on time and every bill to be clear"), and move it offline: "Please call our office manager at {{phone}} so we can look into this."
Keep it under 80 words, no emojis. Give me two versions: one warmer, one more formal.
Fill in your details and the prompt updates live — then copy.
Every practice sends the same 15 or so documents — extraction post-op, temporary crown care, new-patient welcome — and most were written a decade ago at a reading level patients skim past. Practices building a letter library with AI report saving 8-16 hours of writing in the first week alone.
Prompt
You write patient instructions for {{practice_name}}, a general dental practice. Create a post-op instruction sheet for: {{procedure}}.
Format, in this order:
1. "The first 24 hours" — the 4-6 things that matter most today, as short bullets.
2. "What's normal" — expected sensations (soreness, oozing, sensitivity) with realistic timelines.
3. "Call us if" — clear warning signs, each on its own line, with our number {{phone}}, and an instruction to call 911 for emergencies like trouble breathing or swallowing.
4. "Do / Don't" — two short columns covering eating, brushing, smoking, straws, and exercise as relevant to this procedure.
Rules:
- 6th-grade reading level. Short sentences. No dental jargon without a plain-word explanation.
- For pain medication, write only: "Take medication exactly as prescribed or as Dr. [DOCTOR] directed" — do not invent drug names or doses.
- Under 400 words so it prints on one page.
- Then give me a 2-3 sentence version of the key points, sized for a follow-up text message.
Fill in your details and the prompt updates live — then copy.
Common questions from dentists
Is it a HIPAA violation to use ChatGPT in my dental practice?
Using it with patient-identifiable information is — OpenAI does not sign Business Associate Agreements for consumer ChatGPT accounts, and removing the name is not enough when dates, tooth numbers, and clinical details can still identify someone. The compliant pattern is to fully de-identify before pasting, or use a BAA-covered tool for anything that touches the chart.
Can AI write insurance narratives that actually get claims paid?
It drafts them well and fast — teams report cutting narrative time from 20+ minutes to under 3. But the narrative must match what is documented in the chart; an AI-embellished finding is a fraud exposure, and carriers increasingly run their own AI on the claims you submit. Draft with AI, verify against the chart, and have the dentist sign off.
Do I need expensive dental AI software, or is a $20-a-month chatbot enough?
For writing work — narratives, letters, review replies, recall copy — a general-purpose subscription covers it, and practices report it pays for itself within the first week. Purpose-built tools earn their cost where they integrate with your PMS or analyze radiographs; do not buy them to solve a writing problem.
Can I rely on AI to read X-rays or make diagnoses?
Only FDA-cleared systems like Overjet or Pearl are built for that, and even they work as a second set of eyes — the diagnosis remains yours. General chatbots are not diagnostic devices, and the ADA's emerging standards (ANSI/ADA 1110-1:2025) exist precisely because unvalidated image analysis is not ready for clinical reliance.