An optometry practice can use ChatGPT for recall letters, review replies, FAQ answers and website text, provided no patient name, date of birth, test result or prescription is ever typed in. Draft the wording with placeholders, have someone approve it once, and merge real names inside your practice software, not the chat.
That rule covers most of the useful work. The jobs that need patient data, such as reading a recall list or summarising a clinical record, belong in tools built for that and bound by an agreement, which a general chat plan is not. What follows is a practical set of safe uses, with wording you can adapt and the checks that keep each one honest.
ChatGPT for optometrists: where it earns its place at the front desk
Look at one ordinary week in a small practice. The same eight messages get typed again and again: recall for a routine eye test, a gentle second reminder, a children's recall, a contact lens aftercare invitation, a thank-you for a review, a reply to a mixed review, an answer about opening hours or frames, and a polite note about a missed appointment. Each takes two to five minutes when written fresh, and every one is written by someone who also has patients at the desk.
The aim is a library of about a dozen approved messages, written once, owned by the practice manager, and filled in with real details by the software. That is a day's work to build and saves a few hours a month afterwards. The tool's job is the first draft. It is not in the loop each time a message goes out.
Recall letters and texts that say enough and nothing more
A recall message has three jobs: tell the person it is time, make booking easy, and respect that other people may read it. The safest wording describes the kind of visit and nothing about the patient's eyes.
Write three versions of a recall text and one letter for an
independent optometry practice. Placeholders: [FIRST NAME],
[PRACTICE NAME], [BOOKING LINK], [PHONE]. The message invites
the person to book their next routine eye test. Rules: 160
characters or fewer for texts; friendly, not pushy; no mention
of any condition, result or prescription; include a way to opt
out of reminders; no urgency claims; no discounts. For the
letter, 120 words, with one clear next step.An illustrative text from that prompt: "Hi [FIRST NAME], it's time to book your next eye test at [PRACTICE NAME]. Book online: [BOOKING LINK] or call [PHONE]. Reply STOP to opt out." It is 141 characters with the placeholders in place, and a long practice name or booking link can push the merged text past 160, so test the merge with your longest real values. The letter version runs to the same message with a warmer opening, a line about how long the appointment takes, and the phone number. Notice what is missing: any hint of why the person is being recalled.
The common error is helpful specificity. A draft proposes "your contact lens check is overdue and your prescription may have expired", which tells a partner who opens the envelope far more than they should know and may be wrong. Ask for the generic form each time, then add clinical specifics only in a conversation with the patient.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowMain example: rebuilding a recall sequence for a three-chair practice
Take an illustrative three-chair practice with about 6,000 patients on its books. The manager believes recalls go out late and inconsistently, because the wording lives in one person's head. Over an afternoon she builds a sequence in the practice's existing software, using ChatGPT for the wording only.
- Message 1, sent 8 weeks before the due date: a friendly invitation with a booking link.
- Message 2, 4 weeks before: a short nudge with the phone number.
- Message 3, at the due date: a letter for people without a mobile number, with the practice's opening hours.
- Message 4, 6 weeks after the due date: a last, low-key note that says "we'll leave it with you" and offers a way to stop reminders.
Suppose most patients are recalled every two years, so about 3,000 recalls fall due each year. If hand-typing took 2 minutes each and the sequence is now automatic, the saving on messages alone is around 100 hours a year, though this ignores set-up time and is a ceiling, not a promise. The number that matters is the proportion of recalls that lead to a booked test, which she tracks for a quarter before and after. The tool wrote the words; the practice software sends them, merges names and holds the opt-outs.
For a deeper look at running recalls and bookings in an independent practice, see how independent opticians use AI for recalls and bookings, and for contact lens patients specifically, aftercare and reorder messages.
Using ChatGPT for review replies without confirming who anyone is
Review replies are public, which makes the privacy rule stricter, not looser. A reply that thanks someone for "their recent visit for a sight test" has confirmed that the reviewer is a patient. Many reviewers use their real names, and some do not, but the principle is the same either way.
A safe pattern has four parts: a thanks, a general statement about the standard you aim for, a route to talk offline, and nothing about the visit. Draft it from a prompt such as this.
Draft a public reply to a one-star review of an optometry
practice. Do not confirm the reviewer is a patient, do not
mention any visit, appointment, product or clinical matter, do not
admit or deny fault, do not use their name. Thank them for the
feedback, say we take it seriously, and invite them to contact the
practice manager at [PHONE] so we can look into it. Under 70 words.
Calm tone.An illustrative output reads: "Thank you for taking the time to share this. We take all feedback seriously and want to understand what happened. Please ring our practice manager on [PHONE] so that we can look into it properly." The manager adjusts one thing: she removes "what happened", which implies an event took place, and writes "understand your experience". That edit matters because it avoids implying that the reviewer visited.
For positive reviews, ask for three variations that thank the reviewer without referring to their purchase or test. Rotate them so the replies do not look robotic. Do not let a tool auto-post replies; a person clicks publish. The tutorials on responding to Google reviews with AI and on patient reviews and confidentiality explain the wider rules.
Front-desk FAQ answers a receptionist can trust
Patients ask the same practical questions: how long does an eye test take, do I need to bring my glasses, can I bring a child, what happens if I'm late, do you take my insurer, can I use a voucher. ChatGPT can turn your answers into clear, friendly wording and a consistent order. Give it your facts, and never ask it to supply them.
Here are our practice's facts. Turn them into 8 short FAQ answers
for the website, each under 50 words, plain English, no clinical
claims. Facts: eye tests take about 30 minutes; bring current
glasses and a list of medicines; children are welcome and
appointments for under-16s run 40 minutes; if you are more than
10 minutes late we may need to rebook; we accept [INSURERS LIST];
contact lens fittings need a separate appointment.A good response sticks to the facts you supplied and says "call us to check" where a fact was not given. A weak one invents an answer, for example "most insurers cover a test every year", which is untrue for many plans. Delete anything that was not in your input. Then put the final text on the website and into a one-page reference for the front desk, so staff and site say the same thing.
Three smaller jobs worth a prompt
A missed-appointment message
Ask for a kind, short text that says "we missed you today" and offers a booking link, without any guilt and without saying what the appointment was for. Keep it to a neutral line about "your appointment".
A post about frame styles or seasonal opening
Social posts about new frames or holiday hours contain no patient data, so draft freely. The check is accuracy: dates, brand names and opening times come from you.
A script for a difficult phone call
Ask for a three-line calm script for a patient who is upset about a wait. Practise it with the team. The script never includes a real person's details, so it can sit in a shared folder.
A children's recall, which needs its own wording
Children's recalls are written to a parent or carer, not the child, and they often prompt a school-holiday booking rush. Ask for a version that addresses "the parent or carer of [FIRST NAME]", says that a routine eye test for children is a good thing to book during the holidays, and offers the earliest slot types without promising dates. The illustrative draft should avoid any line such as "your child's eyes may be changing", which sounds clinical and may worry a family for no reason. Add a sentence about bringing current glasses, because the child will not remember. A parent who reads three short lines and finds a booking link will act; a parent who reads a paragraph about eye health will put the letter on the fridge.
A mixed review, handled in two stages
The hardest reviews are three-star ones: friendly about the staff, unhappy about the wait. The first stage is internal. Ask ChatGPT to turn the review into a bulleted list of what is being praised and what is being criticised, with no judgement. Pasting a public review is fine, because it is already public, though remove any name. The second stage is the reply. Ask for a response that thanks them for the kind words about the team, acknowledges that waiting is frustrating without saying that it happened to them, and invites a call. An illustrative version: "Thank you for the kind words about our team. We know waiting is never welcome and we are always looking at how we run appointments. If you would like to talk it through, please ring [PHONE]." The practice manager then decides separately whether the complaint points at a real scheduling problem, and the AI has no say in that.
Large-print, plain-language and translated versions
Many patients are older, partially sighted or read a second language, and an optometry practice should hold itself to a higher standard than most on readable letters. Ask for a version of each template in large-print layout with short sentences and no abbreviations, and one at a reading age of about 10. ChatGPT can also draft versions in other languages used by your community. Treat these with caution: have a fluent colleague or a professional translator check any translation that goes to patients, because a mistranslated opening time or phone number is worse than no translation. Keep the English source as the master, and tag each translation with the check date.
A 30-minute briefing for the team
Do not rely on a policy document alone. Run one short session with the front desk and optometrists, in this order. Spend five minutes on the rule: nothing from the patient system goes into a chat. Spend ten minutes live-drafting one recall text together from placeholders, so people see how little the tool needs. Spend ten minutes on a mock review reply that tries to be helpful and breaks the rule, so the team can find the problem, such as confirming the visit. Finish with five minutes on what to do after a slip: tell the manager the same day, delete the chat, and write down what was entered. A short, practical session beats a long document, and the live drafting shows the team that the safe way is also the quick way.
What ChatGPT cannot know about your practice
Drafts fail in predictable places because the tool has no knowledge of your local facts. It does not know your opening hours, your insurers, whether you offer same-day contact lens collection, how long your tests run or what your cancellation notice is. It does not know how much a pair of glasses costs in your shop. It will fill those gaps with plausible-sounding text if you let it. Put a block of verified practice facts at the top of every prompt, or keep it in a Project's instructions, and add the line "if a fact is not in this block, write [CHECK] instead of guessing". Then search the draft for [CHECK] before approving. That small habit removes most of the embarrassing errors.
A quick sum for choosing which message to automate first
If you can only build one thing this month, pick by volume and pain. Say the practice sends about 250 routine recalls a month, 30 contact lens aftercare invitations, 12 review replies and 20 answers to the same five FAQs. At roughly 2 minutes for a recall, 3 for an aftercare note, 5 for a review reply and 2 for an FAQ answer, the monthly typing time is about 500, 90, 60 and 40 minutes. Recalls dominate by a wide margin, so they come first. Review replies are low volume but high risk, so they come second for a different reason: a standard, approved structure prevents a public mistake. The figures are illustrative; time your own front desk for one week before deciding.
Consent and tone checks before wording goes live
Two checks apply to every message in the library. The first is consent. Reminders about care a patient already receives are generally treated differently from promotions, and the rules differ by channel and place. Keep a record of each patient's preferred contact method, separate reminders from marketing in your system, include a clear opt-out, and have your adviser confirm what applies to you. If a message includes an offer, a new service or a frame promotion, treat it as marketing.
The second is tone. Read each message as if you were a nervous patient. Does it sound like a demand? Does the missed-appointment text sound like a telling-off? Would the text make sense if someone else read it over a shoulder? Trim exclamation marks and anything that sounds like a sales message. A recall that sounds like a reminder from someone who cares converts better than one that sounds like a campaign, and you can check this with your own booking numbers.
| Use | Safe with ChatGPT? | The condition |
|---|---|---|
| Recall and reminder wording | Yes | Placeholders only; merge names in your own software |
| Review replies | Yes | No confirmation of who the reviewer is or what they had done |
| FAQ answers | Yes | Facts supplied by you; no invented policy |
| Social posts and website text | Yes | A human checks dates, prices and claims |
| Reading a recall list or patient file | No | Needs a tool covered by an agreement and your own policy |
| Explaining a specific patient's result | No | Clinician-written only |
A few things that go wrong in practice
Three illustrative slips show the pattern. A receptionist pastes a screenshot of the recall list into the chat to "tidy the formatting". It contains names, dates and mobile numbers. The fix is the same as for any practice: nothing copied from the patient system goes into a chat, and the format change is done in the spreadsheet.
A draft promises "free follow-up checks" because the model liked the phrase. It went out in two texts before anyone noticed, and the practice honoured them. The fix is to read every message for offers and claims that you did not write.
A review reply lists the practice's cancellation policy in detail, because the model was trying to be helpful, and now the thread reads as an argument in public. Keep replies short; take the debate offline.
Storing the approved library and keeping it fresh
Keep the final wording in one shared document with a version date, the name of the person who approved it, and a tag for when it is used. Review the document every quarter, and whenever hours, prices or policies change. Use ChatGPT Projects, or the equivalent in other tools, to hold your tone rules and facts so each new draft starts from the same base. Ask each team member to use their own login rather than a shared one, and avoid connecting mailbox or drive apps (formerly connectors) to the workspace used for drafting.
Measure results after one quarter. Compare the proportion of recalled patients who book, the time the front desk spends typing messages, and the number of complaints about wording. If bookings rise and complaints stay flat, extend the approach to contact lens aftercare and to your waiting-list messages. If you would like a hand to map your own set of messages, that is what my AI implementation consultation covers, and you can begin at the contact page. A fuller week-by-week plan is in the first-month plan for an optometry practice.
Questions optometry owners ask about ChatGPT
Can ChatGPT read my patient recall list?
Not safely on a general plan. A recall list has names, contact details and test dates. Use ChatGPT to write the wording, with placeholders, then merge names and dates inside your practice software or mail tool, which is the system already holding that data.
Should recall messages mention the patient's condition?
No. Texts and letters can be seen by a partner, a parent or a colleague at the same address. Keep recalls to the type of visit and a booking route, such as time for your next eye test, and leave results, prescriptions and conditions for the appointment.
Can I reply to a negative review with details of what happened?
No. Confirming that someone is a patient, or describing their visit, is itself a disclosure. A safe reply thanks them, says you take feedback seriously, gives a way to contact the practice and says nothing about whether or what they were seen for.
Do I need consent before sending recall messages?
Reminders about care a patient already receives are treated differently from promotions, and rules vary by place and channel. Record how each patient prefers to be contacted, keep marketing separate from reminders, and have your adviser confirm the rule that applies to you.
Further reads
- Explaining Lens Options and Eye Test Results With AI Help — Explaining lens options and test results in plain words.
- AI Receptionist for Optometrists: Eye Test Bookings and Recalls — When a voice assistant should take eye test bookings.
- Optometry Insurance Verification: A Front-Desk Workflow With AI — A front-desk workflow for insurance checks.
- Optometry Referral Letter Template: Draft With AI, Then Check — Draft a referral letter template, then check it.
- AI Marketing Automation for Doctors: Recall, Reminder and Review Flows — Recall, reminder and review flows from another clinic type.
- How Dental Practices Use AI for Recalls and Appointment Reminders — How a dental practice runs recalls, for comparison.
- GDPR in an Optical Practice: Which AI Tools May Touch Records — A scored 12-point checklist for vetting any AI tool before it touches eye-test records, with four tools worked through and the paperwork behind each.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI help centre articles on projects, data controls and business associate agreements, and OpenAI's enterprise privacy page (checked 4 November 2026).