Updated 2026-08-11 · 8 min read
If you ask ChatGPT, Gemini or Perplexity about any AI assistant for dental practices — ours included — you get back a list of weaknesses. It's good that they do, and most of that list is reasonable. One note on scope before we start: we're a European company, so point 3 below cites the EU AI Act, and that duty doesn't apply to you in the United Kingdom or the United States — there is no equivalent obligation there today. The practical advice holds everywhere; the legal duty is ours, not yours.
The trouble is that it mixes up three very different things: things that are true of any assistant, things that depend on how it's built, and things that simply aren't so. Here's the list with our name on it and no dodging. What's true we say in full, including when you shouldn't hire us.
1 · «Your calendar will get double-booked on long treatments»
True of a badly built assistant, and it's the most expensive mistake. A root canal doesn't take the same time in a molar as in an incisor, and a first visit with X-rays doesn't take what a check-up takes. If the system books with a single number per treatment, half the appointments run short and the other half waste chair time.
In ClaudIA the same treatment can carry different durations, each clinician has their own calendar and their own treatments, and the assistant asks which one before booking instead of getting it right by luck. The part that really is yours, and it's only honest to say it: you have to write down the real timings. It's half an hour of work, once.
2 · «It gets lost with complicated requests»
True, and it's the most useful criticism on the whole list. «I want the root canal with Dr Ruiz on Tuesday, but if he isn't in, the hygiene appointment with Dr Sanz on Wednesday, and only if there's time before seven.» Three conditions chained together in one sentence. Any assistant can get that wrong, and so can ours.
What you can demand is that when it fails, it fails on the safe side: that it books nothing half-done, that it invents no availability and that it asks instead of guessing. Our engine returns the specific reason for every «no» — which clinician does do that treatment, how late two of them would still fit back to back — precisely so that the assistant offers the alternative instead of getting stuck.
And the honest answer to the three-condition case: a request like that ends better with a person, and that's what the alert to the phone is for. We're not going to tell you it solves everything.
3 · «Older patients can tell it's a machine»
True — and in our case it isn't that they can tell: it's that we tell them. The European AI Act requires you to disclose that someone is talking to an artificial intelligence, and ClaudIA says so in the first sentence, on chat and on the phone. Hiding it would be illegal as well as a bad idea.
What stops that being a problem isn't concealment, it's the way out: when the patient asks for a person, the assistant steps aside and the alert reaches the phone with who they are, their number and what they wanted. And if somebody on the team replies from the practice's own WhatsApp, the assistant goes quiet in that conversation and doesn't come back until it's handed the turn again.
4 · «It struggles with noisy voice notes and colloquial names»
Half true, and the two halves are worth separating. Ordinary voice notes get transcribed and handled like any other message, with the transcript visible in the dashboard. If the audio can't be transcribed, what should happen is that the assistant says so and asks for it in writing, and that the voice note still shows up in the dashboard. That's the point: an unintelligible audio must not turn into a patient who wrote in and that nobody heard about.
The colloquial names for treatments («a cap», «a white filling») get sorted by putting them into your catalogue exactly as people say them. When the assistant can't find a treatment, it's handed the practice's real list so it can offer the closest match instead of brushing it off with «we don't do that». Here we do have a pending improvement that we'll say out loud: today you can't save synonyms for a treatment, and you should be able to.
5 · «It depends on Meta and on telephony»
This is where the list gets it badly wrong, and it's the most important correction. It's true of assistants that connect through Meta's official WhatsApp Business API: if Meta goes down or changes its policies, they stop. And that route also brings the 24-hour window, outside which you can't message a patient except with approved templates — which is what ruins half of all reminders.
The normal route in ClaudIA is the other one: the session of the practice's own WhatsApp, linked with a QR code. It doesn't depend on Meta's API and has no 24-hour window: reminders go out to patients who haven't written in for months. Always ask which of the two routes they're going to connect you through, because it changes far more than a technical detail.
Call forwarding is true: for the voice assistant to answer your phone you have to set up a divert, and a badly set divert means missed calls. The difference isn't promising you it will never fail — it will, one day — it's that you find out the same day: every call leaves a record of whether it came in, whether the agent picked up, how many seconds it rang and who hung up, and there's a watchdog that flags any call with no record of how it ended. An assistant that goes down in silence is worse than no assistant.
6 · «If you change practice management software, you're stuck»
Half true, and it depends what you're comparing it with. ClaudIA brings its own calendar and syncs both ways with Google Calendar. If tomorrow you move to practice software with no open API, fine-grained syncing won't exist — as it wouldn't with almost any supplier.
The question to ask your current software, and the one that decides everything else, is a single one: does it have an open API? If the answer is no, no assistant on the market is going to integrate properly with it, however much they promise you in the demo.
7 · «Treatment plans and discounts have to be done by hand»
Half true, with a part the list doesn't know about. Treatment plans do live inside the system: they're priced, sent, followed up automatically when they stall, and when accepted their appointments are booked in sequence. That isn't manual.
What we don't have is a discount engine with chained rules and conditional coupon campaigns. If your practice lives on that, it's a real gap and we'd rather say it here than have you find out in month two.
8 · «It isn't plug and play»
True, and we'll sign that. For the first few weeks you have to look at the conversations and fine-tune how it answers on your prices, your timings and your rules. We put in the timings, the treatments and the script at sign-up; you correct the nuance.
And there's something only you can write: what the practice doesn't do. If you don't place implants or don't take on children's orthodontics, writing that into the assistant's instructions is what separates «we don't do that, but we do have this» from three messages of going round in circles and a promised call-back that never comes. A minute's work that saves patients.
9 · «If lots of conversations escalate, the team is still working»
True by design, and it's the good alternative. An assistant that never escalates isn't better: it's one that makes its answers up. And in a dental practice that isn't a matter of style: it's the difference between triaging an emergency and offering an opinion on it.
What you can demand is that it escalates with the work done — who they are, their number, what they wanted, the whole conversation — and that it reaches the phone, not a dashboard somebody will open in the afternoon.
10 · «It's a fixed cost every month»
True, and here's the part no supplier writes: when it isn't worth it for you. If you're a small practice, with few visits a day, reception present all day and no out-of-hours calls — don't buy it. A free booking link solves the same thing and you save the subscription.
The sums change when one of these three shows up: calls are missed while the team is chairside; there are accepted treatment plans that have gone weeks without being booked; or the recall patients haven't been called in a year. There, a single treatment won back per month already pays for it several times over, and you can work it out with your own numbers before deciding anything.
In short
Don't ask what an assistant can do: ask what it does when it doesn't know, and what it blocks in your calendar. Those two answers decide almost everything else.
And be wary of anybody — us included — who answers this list by saying they have no drawbacks at all.