ChatGPT prompts for dental offices

You are helping a US dental office manager get ready for an AI phone receptionist. Using only the practice details I give below, draft a one-page call-handling playbook a vendor could configure from. Include: (1) which calls the AI takes (e.g., overflow during business hours, all calls after hours) and which must go straight to a person; (2) each appointment type with its length, which provider or chair it needs, and whether new patients may book it; (3) emergency red flags the AI must always escalate (e.g., facial swelling, trouble breathing or swallowing, uncontrolled bleeding, a knocked-out tooth) and the after-hours escalation path; (4) approved answers about the insurance plans we accept, and what to say when the AI isn't sure; (5) what the AI must never do (give clinical advice, quote fees not on our fee list, cancel an appointment without confirming); (6) a weekly report we want from the vendor: calls answered, calls transferred, transfers not returned, appointments booked, appointments kept, production booked. Ask me for anything missing before you draft. Do not include any patient names or patient information.

Practice details: [office hours] [number of locations] [practice-management system] [appointment types and lengths] [insurance plans accepted] [emergency/after-hours contact procedure]

From the case: AI receptionist for overflow and after-hours calls: Wolfe Dental's results. Never paste patient details into a general AI tool.

Run these on a trial line connected to a test patient in your real schedule (or a sandbox copy). Make each call once during business hours and once after hours. Write down pass or fail for each.

1. Ask for a person. Say: "Can I speak to someone at the front desk?" Pass: during hours it puts you through live. After hours it tells you when you will be called back and creates a task. Fail: it says staff are busy, takes a message and moves on.

2. Move an existing appointment. Ask to move your test patient's visit to a specific open slot. Pass: when you open your practice management system right after the call, the appointment has moved. Fail: it confirms on the phone but only sends your team a message.

3. Book as a new patient with a hard name. Spell an unusual last name once, then correct one letter. Pass: the record in your system is spelled correctly.

4. Do the whole call in Spanish (or your patients' second language), including spelling a name. Pass: same result as in English, and it never pushes you back to English.

5. Book for someone else. Say: "I'm calling to make an appointment for my mother." Then add a second family member in the same call. Pass: each appointment is under the right patient, with your number saved as the contact.

6. Say: "My face is swollen and it really hurts." Pass: it follows your emergency protocol (a same-day slot or an alert to whoever is on call) and gives no medical advice. Fail: it offers the next routine opening.

7. Call when you are closed and ask to book. Pass: it offers a real open slot during the call. Fail: it says you are closed or sends you to a website.

8. Act misheard. Repeat yourself twice as if it did not understand. Pass: it offers a person or a callback instead of looping.

9. Ask the vendor in writing: "What is your task completion rate for practices like ours? Does a message count as completed? What share of calls get transferred to a person?"

10. After the trial, read the full transcript of every call marked incomplete, not just the summary email, and sort the failures by cause before you decide.

From the case: AI receptionist audit: LA dental office's 15-month call log shows 41% task completion. Never paste patient details into a general AI tool.

I manage a dental office. Our [phone system / practice platform] vendor, [VENDOR], is offering an AI receptionist add-on. We use [PMS: e.g., Open Dental, Dentrix, Eaglesoft]. We only want it for overflow (after [X] unanswered rings) and after hours. Our hours: [HOURS]. Locations: [LOCATIONS]. Emergency policy: [WHO IS ON CALL AND HOW TO REACH THEM]. Appointment types we'd let it book: [LIST WITH LENGTHS, PROVIDERS AND OPERATORIES].

1. Write 10 test calls I can make myself before go-live. For each, give a short caller script and the exact result I should see in our PMS and call log. Include: an after-hours patient in severe pain who needs the on-call doctor; a new patient asking for a cleaning; an existing patient rescheduling; a caller asking whether a specific insurance plan is in network; a caller who insists on talking to a person; a patient who gives a nickname instead of the first name on their chart; and a request for a time or location we don't offer.
2. List the questions I should ask the vendor, grouped under: write-back to [PMS] (does it create the appointment, request it for approval, or just leave a note/task?), call reporting (how AI-answered calls are counted, dashboard, where recordings and summaries are kept), transferring callers to staff during business hours, emergency routing, pricing (per location? minute caps?), and contract length, auto-renewal and cancellation notice.
3. Give me a one-page scorecard to fill in after the test calls and again after the first 30 days.

Don't assume any feature exists. Mark everything I need to confirm with the vendor in writing.

From the case: Dental Intel AI receptionist vs. Mango Voice Margo: what dental offices report. Never paste patient details into a general AI tool.

You are helping a US dental front desk check the quality of an insurance verification. Below is a benefits breakdown with ALL patient identifiers removed (no name, date of birth, member ID or subscriber details). Compare it against this checklist and return a table with three columns: Field | Found (yes / no / unclear) | What to ask the payer.

Checklist: annual maximum and amount remaining; deductible and amount met; coverage % for preventive, basic and major; frequency limits for exams, prophy, bitewings, FMX/pano and perio maintenance; last-service dates for those procedures; SRP and perio maintenance rules; missing-tooth clause; waiting periods; downgrades (e.g., posterior composites); age limits for fluoride and sealants; coordination of benefits.

Then list the 3 questions most worth asking the payer before the appointment. Do not guess any value that is not in the text; mark it 'unclear' instead.

[Paste the de-identified breakdown here. Use only an AI tool your practice has a BAA with.]

From the case: AI insurance verification, fifth tool tried: Overjet reports 15+ hrs/week saved. Never paste patient details into a general AI tool.

First remove the patient's name, date of birth, member/subscriber ID, group number and any other identifiers. Paste only the plan-level benefit breakdown and the planned procedure codes. Do not paste patient information into a general chatbot.

Prompt:
You are helping a US dental front office double-check an insurance benefit breakdown before we give a patient a cost estimate. No patient-identifying information is included. Below are (1) the benefit breakdown our eligibility tool returned and (2) the procedures we plan to quote, with CDT codes and our fees.

For each planned procedure:
1. State the coverage percentage, deductible and annual maximum that apply, and quote the exact line of the breakdown you used. If the breakdown does not state it, write NOT STATED. Do not assume a category (for example, do not assume a crown is 'Major' at 50%).
2. Check whether the breakdown answers each item that commonly changes coverage for that procedure: frequency limits and last date of service; missing-tooth clause; waiting periods; alternate-benefit or downgrade rules; replacement limits for crowns and bridges; implant coverage or exclusions; perio rules (scaling and root planing per quadrant, periodontal maintenance vs. prophy frequency); pre-authorization requirements; remaining annual maximum. Mark each one ANSWERED or MISSING.
3. Flag any contradictions in the breakdown, and anything that looks like a generic default rather than plan-specific data.
4. Give each procedure one verdict: OK TO QUOTE, VERIFY BY PHONE OR PORTAL, or DO NOT QUOTE YET, with one sentence explaining why.

Then write (a) a short list of questions for our staff to ask the payer about every item marked VERIFY, and (b) one plain-language sentence for the patient explaining that the estimate is based on the information the insurer gave us and is not a guarantee of payment.

Benefit breakdown:
[paste here]

Planned procedures (CDT code and fee):
[paste here]

From the case: Dentist drops DentalXChange Eligibility AI within 3 months over crown error. Never paste patient details into a general AI tool.

I run billing for a dental practice with [number] locations. Below is a de-identified export of our denied and rejected claims from the last 90 days. It has no patient names, dates of birth or member IDs. The columns are: payer, CDT code, denial or rejection reason, billed amount, and days to resolution.

1. Group the denials by reason and by payer. Show the 3 reason + payer combinations that cost us the most in dollars and in rework.
2. For each of those 3, tell me whether it looks preventable before submission (missing attachment or narrative, eligibility, frequency limit, coding or bundling) or only fixable after the denial.
3. Recommend ONE workflow to fix or automate first, and explain why in plain language.
4. Write a short pre-submission checklist my team can start using this week for that workflow.
5. List 5 questions I should ask any AI claims vendor. Include how they will measure our baseline denial rate and days to payment before go-live, and how much of the work is software versus people.

If the data is too thin to support a conclusion, say so instead of guessing.

[paste de-identified data here]

From the case: Signature Dental Partners automated clean claims first, reports denials down 50%+. Never paste patient details into a general AI tool.

I'm the office manager at a dental practice with [number] hygienists using [practice management system]. We're evaluating a voice perio charting tool. Write me a 2-week trial plan I can run before signing a contract. Include: (1) a baseline log sheet to fill in the week before, covering minutes per full-mouth perio chart and how often a second team member is pulled in to type; (2) the same log for the trial weeks; (3) an accuracy spot-check where each hygienist compares 10 voice-entered charts with their own readings and notes any errors; (4) a setup timeline with dates so every hygienist is charting by day 10; (5) a list of questions for the vendor about price per hygienist, microphone costs, contract length, cancellation terms and how data is written into our practice management system. Give me a one-page checklist plus one simple table. Use no patient names, dates of birth or other identifiers anywhere.

From the case: Voice perio charting at Heartland Dental: what a 27-office Bola AI pilot reported. Never paste patient details into a general AI tool.

I'm a dentist setting up an AI scribe. Below is one of my existing chart-note templates (for example: limited exam, new-patient exam, or consult). It contains no patient information.

Rewrite it as instructions for an AI scribe that will listen to the appointment and fill in this template. Follow these rules:
1. Keep my headings, their order and my wording exactly. Do not add or remove sections.
2. Under each blank, say in plain words what the scribe should listen for.
3. If something was not said out loud during the visit, the scribe must write [NOT DISCUSSED]. It must never guess or insert a 'normal' default.
4. Tooth numbers, surfaces, quadrants, perio readings, medications and doses are written only exactly as heard. If anything is unclear, write [CHECK].
5. End every note with: 'Draft - reviewed, edited and signed by Dr. ____'.

Then list the 5 places in my template where an AI scribe is most likely to make a mistake, so I know what to check first when I review each note.

[Paste your template here]

From the case: AI scribes for dental chart notes: dentists say templates first. Never paste patient details into a general AI tool.

You are helping a US dental office draft a reusable patient dismissal letter TEMPLATE. Do not ask me for, and do not include, any real patient name, date of birth, account number, balance, appointment history or treatment details. Use bracketed placeholders only.

Write a short, courteous, impersonal letter (under 200 words) using these placeholders: [DATE], [PATIENT NAME], [PATIENT ADDRESS], [PRACTICE NAME], [DENTIST NAME], [PRACTICE PHONE], [EFFECTIVE DATE], [NUMBER OF DAYS].

The letter must:
1. Give the reason in one neutral, factual sentence, chosen from [REASON: repeated missed appointments / unpaid balance after repeated notices / conduct toward staff / breakdown of the dentist-patient relationship]. No blame, no adjectives, no clinical details.
2. Say clearly that the patient should find another dentist, and offer help with a referral [OPTIONAL: local dental society contact].
3. Say we will remain available for emergency care for [NUMBER OF DAYS] days from [EFFECTIVE DATE].
4. Explain how to have a copy of their records sent to the new dentist: [RECORDS REQUEST PROCESS].
5. Encourage the patient to continue any needed care with the new dentist, without describing that care.
6. End politely with a signature block for [DENTIST NAME].

After the letter, give me a separate checklist of what I must confirm before using it: our state's required emergency-availability period, that the reason is not discriminatory, whether the patient is mid-treatment or has an urgent problem, how we will send it and prove delivery, what we will note in the chart, and whether our malpractice carrier or attorney should review the template. Do not give legal advice; flag anything I should check with them.

From the case: 5 office tasks dentists hand to ChatGPT and Claude, and the rules they set. Never paste patient details into a general AI tool.

I own a dental practice. Our website was rebuilt or moved to a new domain around [month/year] by an outside vendor. Since then [describe what happened, e.g. fewer calls or consult requests from the website, or 'I just want to confirm it was done right']. Act as an independent auditor who works for me, not for the vendor. Do not guess. If you can't confirm something from what I give you, say so and tell me exactly how to check it myself, step by step.

What I'm giving you:
1. Our OLD page URLs (from the old sitemap.xml, a Google Search Console Pages export, or the Wayback Machine): [paste]
2. Our NEW sitemap or page list: [paste]
3. If I have it, the result of testing each old URL (status code and final destination): [paste]
4. Our Google Business Profile details: website link, primary and secondary categories, listed services: [paste]
5. How we track leads: analytics tool, call-tracking numbers, form or online-booking conversions: [paste]
6. What the vendor says they did and what their monthly report shows: [paste]
7. The patients and treatments we want more of (e.g., veneers, Invisalign, implants): [describe]

Check and report on:
- Redirects: does every old page that had traffic or links permanently (301) redirect to the closest matching new page, not just the homepage? List any 404s, redirect chains or loops. Are the www/non-www and http/https versions covered? If the domain changed, was a Change of Address submitted in Search Console?
- Indexing: does robots.txt allow the new site? Are there any stray noindex tags? Is a sitemap submitted in Search Console? Are our main service pages indexed?
- Google Business Profile: does the website link point to the new domain (ideally with UTM tags)? Do the categories and services match the treatments in item 7?
- Tracking: do analytics, call tracking and form/booking conversions work on the new site, so website leads can be told apart from other sources? Are our practice name, address and phone the same everywhere?
- Content fit: do the main pages and recent posts target the patients in item 7, or generic dentistry?

Then give me:
A) A prioritized fix list (critical / important / nice to have), each item with how I can check it's done.
B) Plain-language questions to send the vendor this week.
C) Specific benchmarks the vendor should commit to at 30, 60, 90 and 180 days (indexed pages, impressions, clicks, calls, consult requests), so that waiting for SEO comes with checkpoints.

Do not ask for or include any patient information.

From the case: Cosmetic dentist says AI found the redirects her website rebuild missed. Never paste patient details into a general AI tool.