AI Marketing Automation for Doctors: Recall, Reminder and Review Flows

Set up three flows in this order: appointment reminders with a confirm or cancel reply, recall messages for patients overdue for a check or review, and a review request after suitable visits. Use the messaging tools in your existing practice software, get consent and opt-out handling right first, and use AI only to draft and test wording.

Reminders come first because they cost the least and show results within a fortnight. Recalls bring back patients who would otherwise drift away. Review requests come last because they carry the most risk: they touch patient privacy and Google's rules, so they need exclusions and a human reading every reply that goes public.

Follow me on Instagram@sagnikteaches

Why these three flows and not a full marketing funnel

A medical practice has a short list of repeat jobs that happen to be perfectly suited to automation: tell people about an appointment, call back people who are due, and ask satisfied people to say so publicly. Each runs from data you already hold in the booking system (appointment time, last visit, visit type). None needs a newsletter, an advertising account or a content calendar.

Connect on LinkedInSagnik Bhattacharya

AI plays a modest part, and that is deliberate. The flows run on rules inside your practice software: send this message, this many days before, unless the patient has opted out. AI helps with the words, with making a message readable at a school-leaver reading level, with a second language version a staff member then checks, and with sorting a month of results. It should not decide who gets contacted or read symptom descriptions in replies. If you want a wider picture of where AI fits in a clinic, see how clinics cut missed appointments with AI.

Subscribe on YouTube@sagnikteaches

Consent and opt-outs before any message goes out

Settle this on paper before you build anything. Rules on contacting patients by text, email and phone differ between places and between care messages and marketing, so the guidance below is a checklist to take to a qualified adviser, not legal advice.

• Record how each patient agreed to be contacted. A tick box on the registration form, with the channels listed (text, email, phone), stored in the patient record. If you cannot show consent, do not message that patient by that channel.
• Separate care messages from promotional ones. An appointment reminder or a letter saying a blood-pressure review is overdue is part of care. A message offering a new service or a seasonal promotion is marketing and generally needs explicit opt-in. Keep recall wording clinical and free of offers.
• Make opting out one step. Every text ends with a plain instruction such as "Reply STOP to stop these messages", and your software must act on it automatically. Check that the opt-out flag is shared across all three flows, so a patient who stops reminders is not still getting recalls.
• Keep content minimal. No diagnosis, test result or treatment name in a text. "You are due a review at the practice" is enough. A message that reveals a condition, read over someone's shoulder, is a breach.
• Check the messaging provider's terms. Bulk text services can require registered sender names or numbers, and your software vendor may charge per message. For example, Cliniko lets a reminder go out up to nine days before the appointment, at a time you set and with an option to skip weekends; SMS credits are bought separately, and a text over 160 characters uses an extra credit. Check your own software's current help pages, because the options change.

The tutorial on appointment reminder automation and consent goes deeper on the consent side.

Flow 1: appointment reminders with a confirm-or-cancel reply

This is the flow to switch on first. A typical sequence is a message 72 hours before the appointment, a second 24 hours before, and a same-day nudge only for appointment types that often fail. Each message includes the date, time, clinician and location, a one-word reply to confirm, and a number to call to change it.

The wording, drafted with AI and checked by you

Write three versions of an appointment reminder text for a medical practice.
Constraints: under 300 characters, plain words, no diagnosis, visit reason or
treatment name, includes [date], [time], [clinician], a way to confirm (reply
C), a way to cancel or change (call [number]) and "Reply STOP to opt out".
Tone: warm and brief. Do not use emojis or exclamation marks.

Illustrative output you might pick from: "Hello [first name], a reminder of your appointment at [practice] with [clinician] on [date] at [time]. Reply C to confirm or call [number] to change it. Reply STOP to opt out." What you fix: remove any version that names the visit type, and check the character count against the segment limit your provider bills on, because long messages can cost double.

A worked month of reminders

Say a three-doctor practice books 1,400 appointments a month and 9 per cent are missed without notice, which is 126. After reminders with a confirm reply go live, suppose that falls to 6.5 per cent, or 91, so 35 more appointments are kept. At an average fee of $75, the recovered revenue is about $2,600 a month. Message costs depend on your provider; if each reminder pair costs a few cents, 1,400 appointments cost well under $100, but price your own. Those percentages are assumptions to measure against your own before-and-after figures. Record the no-show rate for the two months before launch so the comparison is honest, and look only at appointment types that were reminded.

Use the replies to fill gaps. A cancellation reply should trigger a rule that offers the slot to a waiting-list patient who has consented to short-notice contact. That is a rule in the software, not an AI decision. For the analysis side, analysing no-show patterns with ChatGPT shows how to find which appointment types and times are worth an extra nudge.

Flow 2: recall messages for patients who are due

A recall flow finds patients who are overdue for something clinical and invites them back: a blood-pressure check, an annual review, a vaccination booster, a repeat-prescription review. It needs a list, a rule and a message sequence.

Build the list from the booking system, not from a guess

Run a report for patients whose last visit of a given type was more than N months ago and who have no future appointment. Start with one group. Say the practice has 6,200 active patients and the report for "annual long-term-condition review, last seen over 13 months ago, nothing booked" returns 540. Remove anyone flagged as deceased, moved away, deregistered or opted out of contact, which leaves 486. Have a clinician read ten names to check the rule has not caught patients who were reviewed elsewhere or who need a phone call rather than a text.

The three-touch sequence

1. Touch 1, text or email: "Hello [first name], our records show you are due a review at [practice]. You can book online at [link] or call [number]. Reply STOP to opt out."
2. Touch 2, after 14 days if not booked: a shorter reminder with the same link.
3. Touch 3, after another 14 days: a phone call from reception or a letter, because some patients never act on texts. Mark the patient "recall completed" only when they book or decline.

AI helps write the three versions and adapt them: one at a plain reading level, one for patients who prefer a longer explanation, a translation that a bilingual team member checks before it is used. A realistic mistake is the over-friendly draft: "We have missed you at the surgery!" does not suit a patient who has been avoiding a difficult appointment, and it can read as pressure. Ask for neutral, respectful wording and cut anything that sounds like a sales message.

Illustrative result: of 486 contacted, say 29 per cent book after touch 1, another 14 per cent after touch 2 and a further 8 per cent after the call, about 248 patients, so around half. Your own figures will vary with the type of recall, the patients and the channel. Track by touch, so you can drop a step that adds nothing. The tutorial on drafting patient recall letters with AI has wording for the letter that serves as touch 3.

Flow 3: review requests, with exclusions that protect patients

A review request goes to patients after a visit and invites them to leave a public Google review. This is the flow where practices get into trouble, so build the exclusions into the rule before the message exists.

Who not to ask

• Visits for sensitive reasons (mental health, sexual health, fertility, anything the patient might not want associated with the practice name in public).
• Children, or parents of children, unless you have decided a parent-facing request is appropriate.
• Anyone with an open complaint or a flagged incident.
• Patients who opted out of non-essential messages.
• Anyone who may be vulnerable or who lacks capacity to consent to a public post.

Google's rules you must not break

Google's Maps user-generated content policy prohibits offering incentives such as payment, discounts or free goods or services in exchange for a review, and it prohibits discouraging negative reviews or selectively soliciting positive ones. That rules out the common trick of sending a private satisfaction survey first and passing only the happy responses on to Google (often called review gating). It also forbids "leave us five stars and get a free X-ray". Read the live policy page yourself, because wording changes. The request must be neutral: invite feedback, give the link, do not steer the rating.

Write a review request message for a medical practice.
Rules: neutral wording, no request for a particular star rating, no offer of
any reward, no mention of the visit type or any health information, under 280
characters, ends with "Reply STOP to opt out". Include [Google review link].
Give two versions, one text and one email.

Illustrative text version: "Thank you for visiting [practice]. If you would like to share your experience, you can leave a review here: [link]. Reply STOP to opt out." What you fix: strip "we hope you were happy" or similar phrases, which lean towards a good rating.

Replying to reviews without confirming anyone is a patient

Do not publish a reply that confirms the reviewer was a patient or refers to their care, even if they wrote about it themselves. Even confirming a relationship can disclose health information. Use a general reply: "Thank you for your feedback. We take all comments seriously. Please contact the practice manager on [number] so we can look into this." Let AI draft the general replies, and have a person publish each one after reading it against the review. The tutorial on patient reviews and confidentiality has more sample replies.

Worked numbers, illustrative: the practice sends review requests to 400 eligible patients a month after exclusions, from about 1,100 completed visits. If 6 per cent respond, that is 24 new reviews a month. A rule that you cannot count on is how many will be positive, and you should not try to influence it. What you can track is volume, average rating over time, and the share of reviews that receive a reply within three working days.

Build it in the software you already use

Most practice management systems have appointment reminders built in, and many offer recall or "patient due" reports, message templates and patient-level contact preferences. Start there before buying a separate marketing tool, because patient data stays in one system and the opt-out flag is shared. Check these before you build:

1. Does the software store consent per channel, and does it block sending when consent is missing?
2. Can you build a recall list from visit history with a "nothing booked" condition?
3. Do templates include merge fields for name, date, time and clinician?
4. Does a STOP reply update the patient record automatically?
5. Can you see message delivery and booking results by flow?

If the answer to several is no, a separate patient-messaging tool may be worth a look, but ask for its data-handling terms first and check whether your privacy rules allow patient contact details to leave the practice system. The consent checklist for clinic chatbots lists the questions to ask a vendor, and they apply here too.

Which recall groups to automate first

Not every recall is worth a flow. Rank candidate groups by how many patients they cover and how much a missed recall matters clinically, then start with the group that scores best on both. An illustrative ranking for a three-doctor practice with 6,200 active patients:

Notice the last row has the most patients but the lowest weight. Large and low-priority groups are tempting because the list looks impressive, yet a clinical recall that reaches 250 of 486 patients does more for patient care than a routine health-check email reaching 640. Build the first group, run it for six weeks, fix what went wrong, then add the second.

Different appointment types need different reminder rules

One reminder template for every appointment is a common shortcut and it causes trouble. Adjust by type, and use the booking system's appointment-type field to pick the rule:

• New-patient visit: send a reminder with what to bring (photo identification, medication list, referral letter) and the time to arrive early to complete forms. This one is worth a longer message.
• Follow-up of 10 minutes: a short text 24 hours before is enough. Add a note that a missed slot is released to others.
• Procedure with preparation (fasting, a full bladder, stopping a product): do not automate the clinical instruction by AI. A clinician writes the preparation text, you store it as a fixed template and the flow sends it. AI can tidy the layout, never the content.
• Telephone or video consultation: include the call window and the number the clinician will ring from, so patients answer unknown numbers.
• Long-gap bookings made months ahead: add a "still need this appointment?" message two weeks before, because these are the likeliest to fail.

A realistic mistake here: the flow sends the generic "arrive 10 minutes early" to a telephone appointment, so patients drive to the practice for a call. Appointment-type rules catch that. Test the flow by booking a dummy patient of each type with your own phone number and reading every message that arrives. Ten minutes of testing beats a morning of corrections.

Before and after: a review reply

Illustrative review: "Waited 40 minutes and the doctor did not listen to my back problem." A first AI draft often reads: "We are sorry to hear about your back problem and the wait. Dr [name] always takes time to listen." That reply names the clinician, mentions the condition and confirms the person was seen: three leaks in one sentence. The safe version: "Thank you for taking the time to share this. We aim to keep waits short and to listen carefully to everyone, and we are sorry if that was not your experience. Please contact our practice manager on [number] so we can look into it properly." Your check: no condition, no clinician name, no visit detail, no defence of the practice.

A quick sum for the whole set

Using the illustrative figures above: reminders recover about 35 kept appointments a month, worth roughly $2,600 at $75 each. A recall group of 486 returning about 248 visits at $75 is $18,600 of booked care, although that is spread across several months and is only partly extra revenue, since some of those patients would have returned anyway. Review requests produce about 24 reviews a month, which is hard to turn into a figure and should not be forced into one. Against that, the costs are message fees, perhaps four to six hours a month of staff time and an hour a month of review. Replace every figure with your own once two months of data exist.

A mistake that shows up in week two

Illustrative failure: a practice sets up recall messages and forgets to exclude patients who booked by phone after the list was run. For two weeks, patients who already have an appointment get a "you are due a review" text, a few reply angrily and reception spends a morning apologising. The fix is a rule: run the list fresh on the day of each send, filtered by "no future appointment", and never reuse an exported spreadsheet. A second failure is time of day: reminders sent at 6 a.m. produce complaints, so set a send window, say 9 a.m. to 6 p.m. on weekdays.

Check each flow worked

Pick one number per flow and read it monthly:

• Reminders: no-show rate for reminded appointments, against the baseline taken before launch.
• Recalls: share of contacted patients who book within 30 days, by touch.
• Reviews: new reviews a month, rating trend, reply time.
• All three: opt-out rate and complaints about messages. If opt-outs climb above what you expect, shorten the sequence before you change anything else.

Give each flow an owner on the team and a monthly slot of 20 minutes. Ask one question in the review: "did anyone receive something they should not have?" That keeps the automation honest.

If you would like help working out which flow your practice software can run today and what is worth paying extra for, that is the sort of thing my AI implementation consultation covers, and you can start a conversation here.

Questions about automated patient messages

Do appointment reminders need the same consent as marketing texts?

Often not, because reminders are usually treated as part of care, while promotions need explicit opt-in. The line varies by place and by channel, so ask a qualified adviser, record how each patient agreed to be contacted, and keep recall wording clinical rather than promotional.

Can AI read patient replies and act on them?

Keep that inside your practice software or a tool your privacy rules approve. A reply such as a cancellation is fine for a rules-based flow. A reply that describes symptoms must reach a person, so route anything that is not a simple yes, no or reschedule to the front desk.

How often should a recall message repeat?

Start with three touches over about six weeks: a first message, a reminder and a final one by a different channel. Stop the moment the patient books, declines or opts out. More than that tends to produce complaints and opt-outs rather than bookings.

Is it acceptable to ask every patient for a review?

No. Skip sensitive appointment types, patients with an open complaint and anyone who may be vulnerable. Ask a mixed sample fairly, never only the happy ones, and never offer a reward, because Google's policy forbids incentives and selective solicitation.

Further reads

• AI for Healthcare Clinics: Referral Tracking and Chase Letters — Chase outstanding referrals with the same rule-based approach.
• GP Practice DNA Policy: Wording, Reminders and Letters With AI — Wording for missed-appointment letters that stays fair.
• Best AI for Paediatricians: Vaccine Recalls and School Forms — A recall flow for children's vaccinations and forms.
• Patient Satisfaction Survey With AI: Questions and Comment Summaries — Collect private feedback separately from public reviews.
• ChatGPT for Optometrists: Recall Letters, Reviews and Front-Desk Replies — How recall letters and reviews work in an optical practice.
• How Dental Practices Use AI for Recalls and Appointment Reminders — How a dental practice runs its recall cycle.
• How to Use AI to Respond to Google Reviews — Tone for replies to positive and negative reviews.
• AI Chatbot for a Doctor's Website: Scope, Disclosure, Handover — A practice owner's set-up guide: scope sheet, disclosure line, hand-over rules, logging and named chatbot tools with costs.
• Cliniko for Doctors: Fit, Cost and the AI Tools That Sit Around It — Cliniko is built for allied health. What that means for a doctor's practice, what it costs, and the AI scribes and phone tools that connect to it.
• AI Marketing for Chiropractors: Claim Limits and Review Replies — Use AI for chiropractic posts and review replies without outcome promises: ASA and FTC limits, a claims register, safe reply wording and prompts.
• AI Marketing for Dentists: A 30-Day Content Plan With Claim Checks — A 30-day AI-assisted content calendar for a dental practice, with a claim check on every post, photo rules and confidentiality-safe review replies.
• AI Marketing for Medical Practices Without Making Health Claims — Service pages, review replies, referrer updates and local search for a medical practice, drafted with AI and kept free of health claims.
• AI for Naturopaths: Intake Summaries and Follow-Up Emails — A naturopath's guide to AI intake summaries and follow-up emails: what to redact, what to keep out, and prompts to copy.
• AI for Psych Providers: Telehealth Consent Forms and Reminders — Draft a telehealth consent form with AI as a first draft, then run it through named clinician and legal review steps before use.
• Childhood Immunisation Reminders: AI-Drafted Letters and Texts — AI-drafted immunisation reminder letters and texts for a practice, plus an overdue-tracking sheet and opt-out handling that plain templates skip.
• Best AI for Medical Practices: Eight Tool Types and Who Each Suits — Eight types of AI tool for a medical practice, what each does, the practice it suits, and the checks to run before buying.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Google Maps user-generated content policy (support.google.com), Cliniko help pages on reminders, your own practice-software documentation and messaging provider terms.