How Dental Practices Use AI for Recalls and Appointment Reminders

Dental practices use AI in three places around recalls and reminders: drafting and personalising the messages, handling patients' replies (confirming, cancelling, asking to move) by two-way text or an AI receptionist, and flagging patients likely to miss or drift so a person phones them. The sending itself still runs from your practice software or patient messaging platform.

The part most practices underestimate is the list. AI can write a lovely recall message, but it can't tell that a patient's mobile number changed three years ago, that their recall interval was never set, or that they died last spring. So the first job is cleaning data, and the second is deciding what the AI may say when a patient texts back "can I come Tuesday instead?" or, more awkwardly, "my face is swollen". Get those two right and the messaging almost takes care of itself.

Follow me on Instagram@sagnikteaches

Stage 1: clean the recall list before any AI touches it

Run a report from your practice management system of every active patient with a recall due in the next 90 days. The exact report name varies by system, but you want these fields: last examination date, recall interval, recall due date, mobile number, email, contact preference, usual clinician, and family or household link.

Connect on LinkedInSagnik Bhattacharya

Then look for the problems that make recalls fail. An illustrative first pass on a three-surgery practice's 90-day list of 1,400 patients might turn up:

Subscribe on YouTube@codingliquids

Allow two to three hours for this the first time, then half an hour a month. It is the least glamorous stage and the one with the biggest effect.

The shared-mobile row is the easiest to find with a spreadsheet. Export the list, sort by mobile number, and add a column with =COUNTIF(E:E,E2) (where column E holds the numbers) to count how often each one appears. Filter for counts above 1, then look at the surnames. Three records sharing a number with the same surname is a family to link. Nine records with nine different surnames is usually a care home manager's work phone, and without the fix that manager gets nine recall texts in one morning, none of which the residents see. Those patients need a recall arrangement agreed with the home, not a text.

Stage 2: map the reminder timeline so nothing sends twice

Most practices have reminders coming from two places: the practice software's built-in reminders and a separate messaging platform. Before adding anything, write down every message a patient can receive and when. A typical set looks like this:

The reason to map it on paper first is a failure that only shows up when a patient walks in. An illustrative case: a patient moves an appointment from 9:00 to 11:30. The messaging platform picks up the change, but the practice software's own reminder was queued the week before, so the evening before the patient gets two texts, one saying 9:00 and one saying 11:30. They trust the first and arrive two and a half hours early. If your map shows two systems sending the 48-hour reminder, switch one off before you add anything new.

Patients who drift past the phone-list stage become lapsed patients, a different problem with different messages; winning back lapsed dental patients picks up from there.

Stage 3: write the message variants with AI

This is where a general assistant such as ChatGPT or Claude helps, and it needs no patient data at all: you're writing templates, and your platform fills in the name and date. One recall message for everyone reads flat, so ask for variants by patient type.

Write 4 dental recall text messages, each under 160
characters including a [booking link] placeholder.
Versions: (1) adult, routine check-up; (2) parent booking
for a child; (3) patient who usually sees the hygienist too;
(4) patient who has cancelled twice before.
Tone: warm, plain, no exclamation marks, no scare tactics,
no discounts. Use [first name] and [practice name].
End each with how to reply if they'd rather be phoned.

An illustrative reply for version 2:

Hi [first name], [child name]'s check-up at [practice name]
is due. Book a time that fits round school: [booking link].
Prefer a call? Reply CALL.

Good, but check two things before using it. Is "[child name]" a field your platform can actually fill? And does "fits round school" suit the months you'll send it? For version 4 the AI might write "we've noticed you've had trouble keeping appointments", which reads as a telling-off. Rewrite it to something like "We know diaries get busy. Pick a time that suits you, or reply CALL and we'll find one together."

Set the new wording against what many practices send now and the gain is obvious. Before: "Dear Patient, you are now due for your dental examination. Please contact the surgery to arrange an appointment." After, for a routine adult: "Hi [first name], your check-up at [practice name] is due. Pick a time here: [booking link]. Prefer a call? Reply CALL." The second one names the patient, removes the phone call as the only route, and gives a one-word way out.

Count characters after the placeholders are filled, not before. Over 160 characters, a standard text is sent as two or more linked parts, and many platforms charge for each part. A template that measures 150 characters with "[practice name]" in it can pass 160 once a 28-character practice name and a long booking link replace the placeholders. Paste one fully filled example into a character counter before loading the template, and ask the AI for a version under 140 characters if the real one runs long.

Stage 4: let AI handle replies, inside strict limits

Two-way texting is where AI saves the most reception time. Patient messaging platforms such as Weave now include an AI receptionist that answers calls and replies to texts, and NexHealth offers one-click recall booking from a message. Whatever the tool, write down exactly what it may and may not do.

The AI may: confirm an appointment; cancel when asked; offer the next available slots with the same clinician; answer practical questions from a fixed list (parking, opening hours, what to bring).

The AI must pass to a person, immediately and during opening hours: any mention of pain, swelling, bleeding, a broken tooth or an accident; questions about fees, treatment or results; complaints; anyone who says they're anxious or distressed; any message it isn't sure about.

Out of hours, messages mentioning swelling, bleeding or trauma should get a fixed reply pointing to your out-of-hours emergency arrangements, never a booking link. An illustrative exchange handled correctly:

Patient: Can't make Thursday, any chance of next week?
AI:      No problem. Dr [name] has Tue 9:20, Wed 14:40 or
         Fri 11:00. Reply 1, 2 or 3.
Patient: 2. Also my gum's been bleeding a bit
AI:      Booked Wed 14:40. I've passed your note about your
         gum to the team and someone will call you today.

The second patient message is the test. A tool that simply booked the slot and ignored the bleeding would be doing its job badly. Include messages like that in every test before you go live, and see which dental front-desk tasks AI can take over for the wider picture.

A test sheet makes the go-live decision a matter of record rather than a feeling. Send each message from a staff phone set up as a test patient, and note what came back. An illustrative sheet from a first run:

Three failures on seven tests is a normal first run. Each gets a rule change, then the whole sheet is sent again, not just the failed lines, because a new rule can break an old pass.

Stage 5: flag the appointments most likely to be missed

Some platforms offer no-show "risk scores". Whether or not yours does, a few plain rules find most at-risk bookings, and you can see why a patient was flagged:

• Failed to attend or cancelled late at least twice in the last two years.
• Booked more than eight weeks ahead.
• New patient's first appointment.
• Long treatment appointment (over 45 minutes) where an empty chair costs most.
• Didn't reply to the seven-day confirmation.

Each morning, reception phones the flagged patients for the next two days. A phone call from a person tends to land better with a patient who has ignored two texts than a third text would. If you want AI to rank the list, export it without names (appointment type, lead time, history counts) and ask it to sort by risk and explain each ranking; then check the top ten make sense to someone who knows the patients.

Read the explanations, not just the order. An illustrative line from that kind of ranking: "Row 14: high risk. Booked 11 weeks ahead, two late cancellations, Monday morning appointment (Monday slots have higher no-show rates)." The first two reasons came from your data. The third came from nowhere: nothing in the export said anything about Mondays, and the AI supplied a general-sounding pattern to justify its answer. Delete that kind of reason, add "use only the columns provided; do not add reasons from general knowledge" to the prompt, and rerun. If your own records do show Monday problems, add a day-of-week column so the pattern is evidence rather than invention.

Stage 6: measure four numbers every week

The escalation line catches the quietest failures. Suppose week two shows 4 escalations out of 380 replies, about 1%. That looks efficient until someone reads the conversations and finds the AI had answered "does a filling hurt?" and "is it safe to have treatment while pregnant?" itself, in friendly but clinical terms. Both belonged with a person. After tightening the rules, week four shows 19 escalations out of 360, about 5%, which is more work for reception and the right answer.

Keep a note of your starting figures before you change anything, or you'll never know whether the AI helped. Reducing no-shows with AI reminders and automatic rebooking covers filling cancelled slots from a waiting list, the natural next step.

Three months of recalls in a three-surgery practice

Take an illustrative practice with three surgeries, four dentists and two hygienists, about 5,000 active patients and a receptionist spending most mornings on recall calls. Before: recall texts went out from the practice software, confirmations came from a separate messaging platform, and some patients received both. The failed-to-attend rate sat at about 7% of appointments.

• Weeks 1 and 2: the list clean-up found the problems in the Stage 1 table. The practice switched off the software's duplicate reminders and kept the platform's.
• Week 3: four recall message variants and two confirmation variants were drafted with AI, edited by the practice manager and loaded into the platform.
• Weeks 4 to 6: AI replies were switched on for confirmations and rebooking only, with every escalation rule from Stage 4. Reception reviewed every AI conversation daily at first, then weekly.
• Week 7 onwards: the flagged-patient phone list replaced the old approach of calling everyone overdue.

After three months, suppose the failed-to-attend rate came down to around 5% and recall calls fell from most mornings to about an hour a day. At an average appointment value of $150, two percentage points fewer no-shows across roughly 1,600 appointments a month is about 32 appointments, or nearly $5,000 of chair time a month that is used rather than wasted. Your own figures will differ; the point is to measure them.

Where recall automation goes wrong in dental practices

• Messages to patients who have died. The most upsetting failure, and it happens when a death is noted in the clinical record but the patient isn't marked inactive. Make "check for deceased" part of every monthly list review.
• Family accounts. Three children on one parent's mobile get three separate texts an hour apart. Most platforms can group family members; switch that on.
• Clinical questions answered as admin. The swollen-face text that gets a booking link. This is why Stage 4's escalation list comes before anything goes live.
• Marketing slipping into care messages. A recall text that adds "ask about whitening" may turn a care communication into marketing, which many places treat differently under consent and opt-out rules. Keep recalls about the recall, and check with your data-protection adviser if you want to combine them.
• Nobody reading the AI's conversations. A monthly sample of 20 AI text conversations, read by the practice manager, catches drift before a patient complains.

If you're weighing up an AI receptionist to handle calls as well as texts, what an AI receptionist costs a dental practice sets out the pricing, and dental practice AI mistakes around consent and data covers the risks beyond recalls.

Further reads

• Explaining Dental Treatment Plans to Patients With AI — The next patient-communication job AI helps with once recalls run smoothly.
• Patient Data and AI: A Confidentiality Checklist for Small Practices — Checks to run before any AI tool touches patient records.
• AI Receptionist vs Front Desk Hire: A Dental Practice Comparison — Whether an AI receptionist can take on more of the front desk.
• What to Ask Before Buying Any AI Tool for a Medical Practice — Questions to ask any vendor selling AI into a practice.
• Can AI Handle Vaccination and Check-Up Reminders for a Vet? — How vets handle the same recall problem, with useful parallels.
• How a Clinic Can Use AI to Cut Missed Appointments — The wider clinic view of cutting missed appointments with AI.
• How Independent Opticians Can Use AI for Recalls and Bookings — Four kinds of optician recall, the wording that gets them booked, an assistant for booking questions, and the figures to track, for independent practices.
• Can AI Make Outbound Calls for Your Business? Costs and Rules — What AI calling platforms charge per minute, what 300 confirmation calls a month really cost, and the consent, disclosure and calling-hours rules to check.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Weave dental product page; NexHealth patient reminders and recall pages (checked September 2026).