AI in dental marketing, described honestly

Six forty on a Tuesday. A man is sitting in his truck outside a strip mall in Fishers with a cracked lower molar, calling the third practice on his list. Your line rings out. By the time your front desk plays the voicemail on Wednesday morning, he has been seen somewhere else.

That gap is real. It is also the gap every AI vendor in dentistry is currently pointing at while they sell you something much larger than the gap requires.

So here is our attempt at an honest accounting. What artificial intelligence genuinely does well inside a dental practice today, where it fails in ways that cost you patients, and the specific things a practice owner should refuse to buy — including a few things we sell.

What it is actually good at

Strip away the category names and there are four jobs AI does well right now. Each one is narrow. That is why they work.

Answering when nobody can.

The most valuable thing AI does for a practice is refuse to let a call die. Not sell. Not diagnose. Answer. A caller who reaches something — a voice that says the office is closed, takes their name, tells them when a human will call back — behaves very differently from a caller who reaches a beep. The same is true of a text that lands thirty seconds after a missed call.* The technology here is unglamorous. It is a machine picking up. That is most of the value.

Drafting the thing you keep not writing.

Your post-op instructions email. The paragraph on your website explaining what sedation actually feels like. The reply to a two-star review that has been sitting there for eleven days because you cannot find the right tone at nine at night. AI produces a decent first draft of all of these in under a minute. It is a fast, tireless, mediocre writer — which is exactly what you want standing between a blank page and a human editor.

Triaging what comes in.

A month of form fills, voicemails, and text threads, and a handful of them are an emergency, and a larger handful are insurance questions that never needed you. Sorting is a genuinely good machine job: reading intent, tagging it, routing it, and putting the urgent one at the top of your office manager's list. Sorting, not deciding.

Showing instead of describing.

Cosmetic consultations have always suffered from the same problem — the patient cannot picture it. A visual simulation, which is what our Smilix tool does, lets someone see a version of themselves and react to it. It is a conversation opener and nothing more. It is not a treatment plan, it does not predict a clinical result, and it belongs in the hands of a dentist who says so out loud when they show it.

Where it fails, specifically

The failure modes are not mysterious. They are the same four every time.

It is confidently wrong about your practice. An AI that has not been given your fee policy, your hours, your parking situation, and your list of what you refer out will invent something plausible. A patient told the wrong thing by a machine wearing your name is worse off than a patient told nothing at all.

It writes fluently and says nothing. Unedited AI copy has a texture — balanced, smooth, faintly enthusiastic, empty of any fact only you would know. Patients feel it even when they cannot name it. So, increasingly, do search engines.

It does not know when to stop. Someone calling about swelling and a fever needs a human immediately. Any AI touching your phones needs a hard escalation path, and you should test it yourself by calling in and describing something urgent.

It quietly erodes the thing you actually sell. Dentistry is a trust purchase. Every interaction a patient later realizes was automated spends a little of that trust. Sometimes the trade is worth it. Sometimes you have automated away the reason they chose you.

The part of our own product line we will talk you out of

We build AI receptionists. We also think most practices asking for one should not buy one yet.

Here is the order that actually works. First, find out what is happening to your calls today — how many go unanswered, at what hours, and what happens to a voicemail once it exists. Second, fix the cheap things: a missed-call text-back, a phone tree that does not bury the option to reach a person, someone assigned to the voicemail box by name. Third, and only then, consider whether an AI voice agent earns its place.

Most of the time, steps one and two recover a good deal of what was leaking, and step three becomes a smaller, calmer decision. An AI receptionist bolted onto a broken intake process just answers the phone faster and disappoints people sooner.

When a practice does run one, we hold two lines. It is optional — no other part of our work depends on it. And it always tells the caller it is an AI, in the first breath, without being asked. A patient who discovers mid-call that they have been talking to software feels tricked, and they are right to.

We will argue against volume too. An agency offering forty AI-written blog posts a month is selling you the appearance of activity. Four pages that answer the questions your front desk actually gets asked, written with things only your practice knows, will do more for you than forty that could belong to any office in Indiana.

What to refuse to buy

Any tool claiming to verify insurance automatically and reliably. Verification is a mess of payer portals and phone calls. Anyone promising it is clean has not shown you the exceptions.
"Integrates with any practice management software." No one integrates with all of them. Ask which specific system, which version, and what happens when the connection breaks. Whether your system connects is a question with a real answer, and you should be given it before you sign anything.
AI that predicts a clinical outcome or interprets an image for a patient. That is not a marketing decision, and it is not yours to outsource.
AI-generated reviews, or "review assistance" that writes the patient's words for them. It is against every platform's terms, it is obvious in aggregate, and it puts your profile at risk. We will not do it, and neither should anyone else.
Any promise of a ranking, a number of new patients, or a revenue figure attached to an AI feature. Nobody controls those. What can honestly be promised is the work.
A system you cannot switch off in an afternoon. If turning it off means losing your patient data, your call recordings, or your website, you no longer own your practice's front door.

What to check on your own site this week

Twenty minutes, no vendor required. Do it from your phone, on cellular, not the office wifi.

Call your own main line at seven in the evening. Write down exactly what a stranger hears, word for word.
Text your own main line. See whether anything comes back, and how long it takes.
Open your website on your phone and count how many taps it takes to call the office. If it is more than one, that is the fix.
Read your three newest website pages out loud. Anything you would not say to a patient chair-side came from a machine and needs rewriting.
Check whether your site explains what actually happens at a first visit, in plain language. It is one of the most-read things on a practice site and one of the most commonly missing.
Search your practice name and compare the hours on your Google profile against your real hours, including the day you closed last month.
If you already run AI on your phones, call in and describe an emergency. Confirm you reach a human, and time it.

Every item on that list is something you can act on without buying anything. That is deliberate.

Where this leaves you

AI is a good answering machine, a fast first-draft writer, a competent sorter, and a useful mirror to hold up during a cosmetic consult. It is not a strategy, it is not judgment, and it is not a substitute for a practice that knows what it is good at. The practices getting real value from it are unromantic about it. They automated four boring things and left the rest alone.

If you want a plain read on which of those are worth wiring up in your practice, and which to leave alone for now, the free Practice Growth Blueprint covers it, and no sales call is required.

* Missed-call text-back works with compatible phone setups. Whether yours qualifies is confirmed on your discovery call.