Who Answers the Phone While You're in the Test Room?

Tags
  • AI receptionist
  • opticians
  • small business
  • healthcare

It is half past ten on a Saturday morning in an independent practice.

The optometrist is twenty minutes into a sight test. The dispensing optician is at the frame wall with a patient who cannot choose between two pairs. The receptionist is at the desk, fitting a pair of varifocals that have just come back from the lab.

The phone rings. Nobody can get to it. It rings out.

On the other end is someone who wants to book an eye examination. Their reading glasses have stopped working, or a recall letter has sat on the kitchen table for a month. They don't leave a voicemail. They search again and ring the next practice on the list.

Nobody in your practice will ever know that call happened.

That scene is illustrative, not a documented case. But if you have worked a Saturday on the high street, you will recognise every part of it. I have. I qualified as an optometrist in 2008, I have worked as a locum in practices of every size, and I run a domiciliary eye care service. I have been the clinician in the test room listening to the phone ring out.

This guide is about whether an AI receptionist for opticians in the UK can help with that problem, and where it must not be trusted.

The phone call you never knew you missed

Missed calls are invisible. Your practice management system records the appointments that were booked. It has no record of the ones that never were.

The calls arrive in waves. First thing in the morning. Lunchtime. Saturdays. The days after a batch of recall letters goes out. They tend to arrive exactly when everyone is with a patient.

Then there are the evening calls. A patient finishes work at six, remembers they need to move Tuesday's appointment, rings, and hears the answerphone. Some leave a message. Many don't.

Most of these calls are not clinical. "What time do you close?" "Is there parking?" "Are my glasses ready?" "Can I move my appointment to Thursday?" "I've had a letter saying I'm due." Each one is simple. Together they pull your receptionist away from the patient in front of them, and on a short-staffed day they pull the optometrist out of the test room.

I have not seen reliable UK figures on how many calls independent opticians miss, and I would be wary of anyone who quotes one without a source. The honest answer is that you need to count your own. I explain how at the end.

What happens when your receptionist is already busy?

Practices usually have three options. Let it go to the answerphone and call back later. Divert to a mobile that someone checks between patients. Or employ another person on the desk.

Each works up to a point. The answerphone relies on patients leaving a message, and many won't. The mobile pulls staff away from patients anyway. Another receptionist is the best answer if the numbers support it, but many small practices can't justify the cost.

An AI receptionist is a fourth option. It is software that answers the phone in natural speech, deals with a short list of routine requests, and passes everything else to a person.

Set up well, it could mean fewer appointment enquiries lost to a phone that nobody could reach. Your receptionist finishes the collection without a ringing phone beside them. A patient who rings at 7pm about a recall letter can book there and then, instead of waiting until morning and forgetting. Your clinical team stays in the test room.

Notice the words "could" and "set up well". I won't promise you more bookings or hours saved. Nobody can do that honestly without first looking at your own phone line, your diary system and your patients.

What changed this month

Several things were published in early November 2026 that are worth knowing about. They are different kinds of evidence, and it matters which is which.

Expert opinion. On 10 November, Healio reported the views of US ophthalmologist Dee Stephenson, speaking at OSN New York. She described AI voice agents answering patient questions out of hours, sending reminders and following up after treatment, and mentioned a patient survey that showed satisfaction with the programme. She was clear that these tools support clinicians and do not replace them. This is one experienced clinician's view from a US ophthalmology setting, not a UK trial.

A vendor demonstration. On 11 November, Optometry Times published a demo of EyeCarePro's AI voice agent for optometry practices, answering calls, booking appointments and handing off calls it should not deal with. Earlier episodes in the same series said these agents suit routine, clearly defined tasks, and that phone access is a real problem for independents. Treat it as a demonstration of what the product is designed to do. It tells you nothing about how well it performs across real practices.

Independent research. On 13 November, Healio covered a study presented at a US rheumatology conference. Researchers tested ChatGPT, Gemini, Claude, Perplexity and OpenEvidence on 60 clinical scenarios. Agreement with guidelines ranged from about 75% to 91%, and made-up answers appeared in between 3.8% and 18.1% of responses, including dosing errors. The study did not test receptionists, and it was not about eyes. Its relevance is simple: a fluent answer is not always a correct one, so clinical questions belong with a clinician.

Regulatory guidance. On 11 November, the ICO published guidance on subject access requests that patients write with AI. Such a request is still valid, the one-month deadline still applies, and you can ask the person to clarify what they want. In October, the government's response on regulating AI in healthcare accepted all 44 of the National Commission's recommendations, including clearer information for patients about AI used in their care, with a roadmap due by spring 2027. For domiciliary services, the Department of Health and Social Care's guidance on AI in adult social care sets out human review of AI outputs, UK GDPR compliance, no personal details in free public AI tools, and an AI policy.

Industry news. IT Brief UK reported that OpenAI's Business Hub, now in beta, lets businesses claim and update how they appear in ChatGPT. Its Business Agents feature is US-only for now. It is a reminder that patients find practices through AI assistants as well as Google Maps. Our guide to using Google Business Profile with AI covers the Google side.

An AI receptionist should know when to stop talking

This is the part that matters most to me as an optometrist.

Every call into a practice is either administrative or clinical. An AI receptionist should deal only with the first kind. Booking, rearranging and cancelling routine appointments. Opening hours and directions. Whether a pair of glasses is ready. Recall bookings. Taking a message for a callback.

Some words on a phone call mean a person must take over now. Sudden loss of vision. Flashes of light or a shower of new floaters. A shadow or curtain across part of the vision. A red, painful eye. Something in the eye, a chemical splash or an injury. When a caller says anything like this, the system should transfer the call straight to a registered member of staff. When the practice is closed, it should give the urgent advice your optometrist has written and agreed, such as NHS 111 or your local urgent eye care service.

That is not triage. The AI is not deciding how serious anything is. It is recognising a short list of words that mean "a person, now", and getting out of the way.

The same line applies to everything clinical. It should not discuss prescriptions, results or anything in a patient's record. If the question needs a clinician, the answer is a callback from one.

Can I really trust AI to speak to my patients?

It is the right question, and I would ask it too. Here are the risks as I see them.

It can get things wrong. The study above shows that even the best general AI tools give some incorrect answers. A tightly scripted receptionist dealing only with admin is a much narrower job, but it can still mishear a name, misread a date or double-book a slot. Somebody needs to check its work, especially in the first weeks.

It can miss something urgent. The red-flag list has to be written by your optometrist, tested with real wording patients use, and reviewed whenever a call is escalated. If it misses one, you need to know, and the script needs to change.

It handles personal data. A recording that links a name to an eye problem is health information, which UK GDPR treats as special category data. You need to know where recordings and transcripts are stored, for how long, who can access them, and whether the supplier uses your calls to train its models. A data protection impact assessment is likely to be needed before you go live. No system is GDPR compliant out of the box. Compliance comes from how you choose it, set it up and run it.

Confidentiality still applies. Your duty of confidentiality under GOC standards does not change because software answers the phone. The system should not confirm to a third party that someone is a patient, or discuss anyone's care.

Patients should know who they are talking to. The call should open by saying plainly that it is an automated assistant, and offer a person. That fits the direction the government has set on telling patients when AI is used in their care.

So can you trust it? You can trust it with a narrow, clearly defined job, set up properly and checked regularly. You should not trust it with anyone's eyes.

How it might work in a small independent practice

Here is an illustrative example, not a real client.

A single-site practice has one optometrist, one dispensing optician and one receptionist. They start small. The AI receptionist answers only when the desk phone has rung for twenty seconds, and outside opening hours.

It introduces itself as the practice's automated assistant. It offers to book, move or cancel a routine appointment, check whether glasses are ready using the order reference, give opening times, or take a message. A patient who rings at seven in the evening about a recall letter can book a slot for next week. A patient who mentions flashes and floaters on a Saturday afternoon is put straight through to the optometrist.

Each morning the receptionist reads the call summaries and returns any callbacks. Once a week the optometrist listens to a sample of calls and any that were escalated. After a month, the practice decides whether to widen what it does or switch it off.

You can see how we set this up in our AI receptionist (AI Front Desk) service, read what an AI receptionist can do for a small business, or look at our wider work on AI for opticians and practices. The same split between routine calls and human handover suits domiciliary eye care services, where staff are out on visits, and local trades who lose jobs when the phone rings on site. Our guide to AI agents for UK small businesses covers the wider picture.

The question every practice owner should ask before installing one

Ask the supplier: "What happens when it gets something wrong?"

A good supplier will answer straight away. You will see every call. You can edit the script. Urgent words trigger a transfer. Recordings are stored in a known place for a known period. Your data is not used to train their models. And you can switch it off.

If the answer is vague, keep looking.

On cost, I won't quote prices here. They vary with the supplier, your call volume, whether it connects to your diary and recall system, and where the data is hosted. Ask for a price based on your real call numbers, and run a short pilot on overflow and out-of-hours calls before committing to anything longer.

Before any of that, count your calls. For one working week, put a tally mark against every call in one of three columns: routine admin, needs a person, and clinical or urgent. Add a fourth for the ones nobody answered, if your phone system shows them. That single sheet of paper will tell you more about whether an AI receptionist suits your practice than any sales demo.

Frequently asked questions

Will an AI receptionist replace my receptionist or optometrist?

No. It catches routine calls when your team is busy or the practice is closed. Your receptionist still runs the desk, and every clinical question still goes to a clinician.

Can it give clinical advice?

It should not. Use fixed scripts for admin, and an immediate handover for any symptom that needs examination or urgent care.

Is an AI receptionist GDPR compliant?

It can be used in a compliant way, but no product is compliant by itself. You will likely need a data protection impact assessment, clear supplier terms, a documented retention period and an updated privacy notice.

How will patients know they are speaking to AI?

Say so at the start of every call, and offer a person. Repeat it if the caller asks.

Where should I start?

Count your calls for a week. List the ten most common reasons people ring. Mark which are safe to automate. If routine admin makes up much of the list, a two-week pilot on overflow and out-of-hours calls is a sensible next step.

Talk to AI WORKLABS

I set up AI WORKLABS because I could see what these tools might do for small businesses, and because I know from the test room what can go wrong when technology is put in front of patients without enough thought.

If you would like to talk through whether an AI receptionist suits your practice, book a free chat. We will look at your calls, your diary and your obligations under UK GDPR and GOC standards. If the answer is that you don't need one yet, I will tell you that.