Expect roughly $79 to $600 a month, depending on whether the tool bills per caller or per call and how many calls it takes. It should answer opening hours, book routine eye tests, take messages about frames and repairs, and hand every clinical question, complaint or prescription request to a person.
The cost is the easy part. What most pages sold by vendors skip is the sorting: an optical practice gets a mix of calls, and a few of them are about someone's eyesight, where a wrong answer is not an inconvenience but a risk. So this tutorial starts with the call types, decides who takes each, and then prices the plans against a realistic volume.
Sorting the calls before you sort the vendors
Put a tally sheet by the phone for two weeks, or export the call log if your phone system keeps one. Tag each call with one of the categories below. These are illustrative shares for a two-chair practice, not an industry statistic; yours will differ, and the point is to find your own.
| Call type | Example | Illustrative share | Who takes it |
|---|---|---|---|
| Opening hours, location, parking | "Are you open Saturday?" | 14% | AI |
| Eye test booking, new patient | "I need an eye test next week." | 16% | AI books or takes details |
| Rebooking or cancelling | "I need to move Thursday's test." | 10% | AI if linked to diary; otherwise message |
| Frame and lens questions | "Do you stock this brand?" | 12% | AI for ranges and hours; staff for stock and price |
| Repairs and adjustments | "My arm has snapped." | 9% | AI books a drop-in or takes a message |
| Collection: "are my glasses ready?" | "Name is [surname]." | 15% | Message to dispensing team |
| Contact lens orders and aftercare | "Reorder my usual lenses." | 8% | Message, then staff |
| Clinical questions | "My eye is sore and red." | 6% | Person, with a safety script |
| Complaints, prescription requests, funding queries | "I want a copy of my prescription." | 5% | Person |
| Sales calls, wrong numbers, silent calls | 5% | AI filters |
On that mix the AI can fully resolve roughly 45 per cent of calls (hours, new bookings, rebookings, filtered junk), take a clean message on another 44 per cent, and must pass about 11 per cent to a person. That last slice is where the quality of the setup shows. The sheet also tells you what to test: the booking rules and the clinical script matter more than the frame chat.
Booking rules the AI needs in writing
A receptionist who has worked in a practice for a year has a head full of rules; an AI has only what you write down. Put these in a plain list before you start:
- Appointment types and lengths: a routine eye test is, say, 30 minutes with the optometrist; a contact lens check is a separate slot. List each name as it appears in your diary.
- New or existing patient: the AI should ask, because new patients may need a longer slot or a form first.
- Children and over-65s: if either group has different slot lengths, funding checks or a guardian rule, say so, and tell the AI to pass on anything it is not sure about.
- Who can see whom: if only one optometrist does children's tests or domiciliary visits, name them.
- Lead time: the AI offers only slots from tomorrow onward, so you never get a same-day booking made by a machine at 5 pm.
- Cancellation wording: a fixed sentence about the notice period, with no discretion to waive a fee.
If your booking software has an API or a standard calendar, the tool may read it; if it does not, the AI takes details and reception confirms. Both are acceptable. A guess is not: the AI must never say "you're booked in" unless it can see the diary. This is the same discipline described in writing call scripts and escalation rules.
The clinical line: what always goes to a person
This is the part to get right, and it is a matter for your clinicians and your professional rules, not for a software vendor. A safe starting point, to be approved by your lead optometrist, is a short script and a hard rule:
If the caller describes sudden loss or change of vision, flashes or a shower of floaters,
eye pain, a recent eye injury, or something in the eye, do not ask further questions.
Say: "I can't assess that. Please contact an urgent eye service or emergency services now."
Then transfer to [duty phone]. If no one answers, take a name and number
and mark the message URGENT for the next available clinician.The AI does not decide what is serious. It spots a trigger phrase and gets out of the way. Test it with ten phrasings, such as "it's gone blurry on one side", "I got bleach in my eye" and "my contact lens won't come out", and check the right ones transfer. Where a call is ambiguous, the safe bias is to transfer; accept a few unneeded transfers as the price of that.
Also write down what it must not do: interpret a prescription, advise on whether glasses are needed, comment on a test result, or give an opinion on a child's vision. Anything that sounds like clinical advice goes to staff. Because patients may share health details unprompted, check how the vendor stores and uses recordings; the questions are in what to check in an AI tool's privacy policy and terms.
What it costs at a realistic call volume
An AI receptionist rarely takes every call in an optician's. It covers overflow when the front desk is with a patient, lunchtimes, evenings and weekends. Take the same two-chair practice with 45 inbound calls on a working day, 25 working days a month. If the AI covers about 30 per cent, that is roughly 340 calls a month, from perhaps 250 different phone numbers. Here is what the main list prices give at that volume (USD, from each vendor's pricing page on 27 October 2026):
| Tool and plan | List price and counting | Cost at about 340 calls from 250 numbers |
|---|---|---|
| Goodcall Growth | $129 a month; 500 unique customers; extra at 26 cents | $129 |
| Goodcall Starter | $79 a month; 100 unique customers; extra at 79 cents | About $197 (150 over) |
| Quo Starter with Sona | $28 per user; 10 calls included, then $1.00 a call | About $358 for one user |
| Smith.ai Pro, 300-call tier | About $500 for 300 calls; extra calls about $2.17 | About $587 |
The gap is almost entirely about what each vendor counts. Goodcall counts a phone number once a month, which favours a practice whose patients ring several times about one order. Smith.ai counts every answered call, with a spam filter and an allowance to remove up to 10 per cent of calls from the bill. Neither is wrong; the cheaper one for you depends on how many repeat callers you have. Count unique numbers in your log before you compare. A more detailed price walk-through for a different clinical practice is in what an AI receptionist costs a dental practice, and the general ranges are in the cost of an AI receptionist for a small business.
The main worked example: setting it against a part-time receptionist
The comparison most owners really make is with a part-time hire. Take an illustrative practice that is thinking of adding a receptionist for 20 hours a week to cover lunchtimes and the busy first hour of the day. At a fully loaded $18 an hour that is 20 x 52 / 12 x $18 = about $1,560 a month. The AI plan from the table, at $129 to $587, is a fraction of that. But the AI does not do what the hire does: greet patients at the desk, handle the walk-in with a snapped arm, chase a payment or take a frame out of the cabinet.
The right question is which calls the hire would have answered. If it is mainly lunchtime and overflow calls, the AI is a good match. If the hire would also run the dispensing desk, the AI is the wrong comparison, and the better move is to use it for the phone and keep your hours for the floor. A practical result: use the AI on the unanswered calls and measure how many extra eye tests are booked. If it books ten extra tests a month at an illustrative fee of $60 each, that is $600 of revenue against a $129 plan; at ten tests the sum works even on the dearest option above.
Frame queries: where the AI helps and where it guesses
Frame calls are the part owners worry about, because stock and prices change weekly. A safe design: the AI knows the brands and the price bands you list and nothing finer. Give it three sentences:
- "We stock several ranges from around $X to $Y; staff will confirm what's in stock."
- "We can fit most frames with the patient's own prescription; staff will confirm for each pair."
- "Repairs and adjustments are done at the practice; walk in during opening hours or leave a message."
If a caller asks for a specific model, the AI should take the model name and number, say the team will check stock and ring back, and log it. That one habit turns a frustrating "I'd have to ask someone" into a tracked lead. The ways AI can help on the shop floor, and whether virtual try-on is worth paying for, are covered in AI frame advice and virtual try-on for opticians.
Five test calls to run before going live
- A new-patient booking: "I haven't been before, I need an eye test." Pass: asks new or existing, offers a valid slot or takes details, repeats the date and time back.
- A snapped frame: "My arm came off this morning." Pass: offers a drop-in time or a message, does not promise a same-day fix.
- A sore eye: "My eye is red and hurts." Pass: gives the approved safety line and transfers; fail: asks for the symptoms in detail.
- A glasses-ready check: "Are my glasses ready?" Pass: takes name and number, promises a text back; fail: says yes or no with no data.
- A prescription request: "I'd like a copy of my prescription." Pass: takes the request and hands it to staff; fail: tries to answer or promises a time.
Score all five and repeat them after any change to the script. One failure on the clinical call means the tool is not ready, whatever it costs.
A sample call, and the message it should leave
An illustrative transcript for a booking call, with what you would fix in it:
AI: Thank you for calling the practice. I'm the automated assistant. I can book, take messages and answer opening-hours questions. How can I help?
Caller: I'd like an eye test for my daughter. She's nine.
AI: Happy to help. Is she an existing patient with us?
Caller: No, first time.
AI: Thanks. I'll pass this to the team to confirm a suitable time, as children's tests are booked by hand. Could I take your name, a phone number and two days that suit?
Caller: [first name], and any afternoon this week.
AI: Thank you. The team will text you within two working hours. Is there anything else?What is good here: it declines to book a child's test it has no rule for, it says who will do what, and it gives a time frame. What to fix: it never asked whether the child wears glasses already or whether there is a concern, which matters to the clinician, but asking is a script change, and it must be phrased as "any concerns you'd like the optometrist to know about?" and never as a question the AI interprets. The message left for reception should look like this:
NEW PATIENT, CHILD (9). Parent: [first name], [number].
Wants: eye test, any afternoon this week.
Not booked. Please text within 2 working hours.
Concern noted by caller: none stated.A clear template like that saves reception a call back to ask the basics. Ask the vendor to send messages in a fixed format, and read the first twenty to see whether they are complete.
A realistic mistake: "your glasses are ready"
Illustrative case: a practice configures its AI with a friendly line, "your glasses should be ready within a week". A patient takes this as a promise, travels in on day five, and finds the lenses are still being edged. Nobody is malicious; the sentence was in the script as general reassurance, and the AI repeats it as fact. The fix is to remove any sentence that sounds like a commitment: say "the team will confirm" and nothing more. Search your script for "should", "will be", "usually" and "normally" and delete each one unless the practice would stand behind it every time.
Three practices, three setups
- A one-chair independent. One optometrist, one dispenser, and the owner doubles as receptionist. Calls go unanswered mostly while the owner is testing. A $79 plan on the unanswered-call forward is enough; use it for messages and hours, and keep booking manual.
- A three-chair practice with a diary link. Heavy phone use and a booking system with an open calendar. This is where direct booking pays, and where a mid-priced plan with an integration is justified. Ask for the integration to be demonstrated on your own diary before you commit.
- A two-site group. Callers mix up which branch they want. The AI must ask "which branch?" first and route accordingly. Test it with a caller who says "the high street one", and check the vendor supports separate hours and numbers per site.
Opening hours, bank holidays and the exceptions list
The simplest calls go wrong most often, because hours change. Give the AI a normal week, then a dated exceptions list: Christmas closures, a training day, a half-day for a staff event, the week the optometrist is away and tests are not offered. Ask the AI to read the exceptions first on any call that mentions a date, and put a reminder in your calendar to refresh the list on the first of each month. Test it by ringing and asking, "Are you open on [the next exception date]?" A pass answers with the exception, and a fail gives the normal hours.
The same applies to anything seasonal, such as a promotion on second pairs or a late-opening evening. Remove the promotion from the script the day it ends, because an AI that cheerfully offers an expired deal creates exactly the sort of dispute that reception then has to settle in person. Keep a short change log, with the date and who updated it, so you can see which version of the script was live when a complaint arrives.
Keeping staff on side
Reception teams often fear the tool will replace them, and they are the people who will tell you where it fails. Ask them to mark every message that arrives with a gap in the first month, such as a missing number or an unclear request, and fix the script each Friday. Tell them in advance what the AI will and will not do, and that its job is to catch the calls nobody can answer. A team that sees fewer interruptions at the desk will usually support it; a team that finds out by surprise will not.
Recalls reduce the load you are paying to answer
A quiet way to lower the call count is to stop patients ringing in the first place. A recall reminder that includes a booking link answers the "when am I due" call before it starts. If you have not set that up, do it before choosing a receptionist, because it changes the volume you are pricing. The method is in AI recalls and bookings for independent opticians. Do not count on the AI to replace this; it answers the calls that come, and recall messages reduce them.
Judging it after a month
At the end of four weeks, tally again. How many calls did the AI answer? How many became eye-test bookings? How many were passed to staff, and were they the right ones? Read every transcript marked clinical, twice. A good month looks like this: fewer unanswered calls at lunchtime, a handful of extra bookings, no safety incidents and a message list the dispensing team can clear in ten minutes each morning. A bad month is a stack of messages nobody owns, or a patient who was told something the practice would not have said.
Then decide with numbers. If it books at least as many extra tests as the plan costs, keep it. If not, narrow its hours to the gaps, such as lunchtime and after 5 pm, and look again. For more jobs where AI earns its place in a practice, see eight practical AI uses for an independent optician.
Questions opticians ask about AI on the phone
Can the AI book eye tests directly into our diary?
Only if the vendor connects to your practice-management or booking software, and many general tools do not. Without a link, the AI takes the patient's details and preferred time, and reception confirms. Ask for a named optical customer on your system before you rely on direct booking.
What should happen when a caller describes a sudden change in vision?
The AI should not assess it. A clinician-approved script should tell the caller to seek urgent eye care or emergency services at once, and transfer the call to a person if one is available. Have your lead optometrist write and sign off that wording.
Can it answer 'are my glasses ready'?
Not unless it can read your dispensing records. Otherwise it takes a name and phone number and sends a message to the dispensing team. A wrong 'ready' costs the patient a wasted trip, so the safer answer is a prompt text back.
Is patient information safe with an AI receptionist?
That depends on the vendor and the rules where your patients live. Ask where call recordings are stored, how long they are kept, whether the vendor trains on them, and whether it offers a data-processing agreement. Collect the minimum on the call, such as name, number and reason, and not clinical detail.
How do I tell patients it is an AI?
Say so in the first sentence: 'This is the practice's automated assistant; I can book, take messages and answer questions about opening hours.' People accept it more readily when the greeting is honest and tells them how to reach a person.
Further reads
- AI Receptionist vs Front Desk Hire: A Dental Practice Comparison — AI versus hiring for a front desk, worked through.
- Gym AI Receptionist: Monthly Cost, Tours, Trials and Class Bookings — Four AI receptionist plans costed for one small gym, plus what they can book for tours, trial passes and classes, and when a person takes over.
- Optician Website: AI-Drafted Eye-Test Pages, Frame Ranges and FAQs — Prompts and a sign-off routine for drafting an optician's eye-test page, frame ranges and FAQs with AI, with booking wording and no clinical claims.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Goodcall pricing page; Smith.ai AI Receptionist pricing page; Quo pricing page (all checked 27 October 2026).