17 AI receptionists launched in 2025. Your front desk just got nervous.

U.S. dental offices spend roughly $3.56 billion a year on people to answer the phone. That figure comes straight out of the federal wage data: 83,950 receptionists and information clerks work in offices of dentists at an annual mean wage of $42,460, and it is the number every AI phone vendor has circled on a whiteboard. Count the AI receptionist products pitched at dental practices in 2025 and you will lose the thread somewhere in the teens. By the time you finish reading this, add two more.

Our view: the front desk is not the first dental job AI takes, and buying an AI receptionist to cut a salary is the worst possible version of a decent idea. We would buy one this year. We would not fire anyone. Those two sentences are the whole argument, and plenty of owners will disagree with the second one, because the pitch deck they just sat through did the subtraction for them.

The Market Already Voted, and It Voted for Insurance Verification

Dentists are not AI holdouts. The ADA's Health Policy Institute found that 43.3% of dentists are currently using AI for at least one task in the practice. Adoption is here. But look at where it landed. Only 10.1% use AI for reception and front-desk check-in, while 13.6% use it for insurance verification.

That gap is small in percentage points and enormous in meaning. Given the choice between pointing software at the phone and pointing it at the payer, more practices picked the payer. They picked the task that is rules-based, repetitive, invisible to the patient, and absolutely miserable to staff. Nobody has ever written a Google review about how warm the eligibility check felt.

Reception is the opposite. It is the task with the most judgment per minute, the most upside when done well, and the most catastrophic downside when done badly. A verification bot that gets it wrong costs you a write-off. A reception bot that gets it wrong costs you the patient, the referral, and the review.

The $44,001 That Isn't There

Here is the arithmetic the vendor wants you to run, and we will run it honestly so you can watch it fall apart.

Take one front desk employee at the industry mean of $42,460. Load it with payroll taxes and benefits at a conservative 22% and the true cost is $51,801. Now price an AI receptionist at $650 a month, which is roughly the middle of what is being quoted: $7,800 a year. Subtract, and you have saved $44,001.

That number is fiction, for one structural reason. Across the whole industry there are 83,950 dental receptionists supporting 205,088 professionally active dentists, which works out to about 0.41 front desk people per dentist. The average practice in this country is not running a redundant front desk. It is running a thin one.

Now look at what that thin person actually does. Phones are a slice of the job. Insurance verification, treatment plan presentation, financial arrangements, recall, reactivation, collections on the aging report, the clinical handoff, the no-show call-back, the two-hour fight with a payer over a downgraded crown: none of that goes away when the phone starts answering itself. If phones are 14 of a 40-hour week, the AI is addressing 35% of one person's time, worth $18,130 in loaded cost. Net of the subscription, that is $10,330 of theoretical capacity returned.

Theoretical, because you cannot lay off 35% of a person. What you can do is stop paying for the overflow: the overtime, the Saturday catch-up, the temp agency week, the vacancy. Those are real dollars. They are just a tenth of what the deck promised.

Six and a Half New Patients, and Then You're Printing

The honest case for an AI receptionist has nothing to do with payroll. It is a revenue case, and it is a good one.

The subscription costs $7,800 a year. If a new patient is worth $1,200 in first-year production in your practice, and you should pull that number from your own ledger rather than from anyone's marketing page, then the tool breaks even at 6.5 new patients. That is one new patient every eight weeks, captured from calls that currently ring into a voicemail box at 6:40 p.m. on a Tuesday.

Capture 24 instead of 6.5 and you have booked $28,800 of first-year production against a $7,800 cost. That is a 3.7x return. It is also entirely dependent on something no vendor can tell you and most owners have never measured: how many calls are going unanswered right now, and what those callers wanted.

Notice the difference between the two frames. The payroll frame caps your upside at one salary and puts your reputation at risk to get it. The capture frame has no cap, costs nobody their job, and fails cheaply if it fails. Same product. Two completely different bets.

What Most Owners Get Wrong: They Buy Before They Baseline

The specific mistake, and we see it most in two-to-four location groups with an office manager stretched across all of them, is signing a 12-month agreement without a single week of before-data.

No call volume log. No answer rate. No count of after-hours calls. No record of how many callers hung up before anyone picked up, and no idea what share of them were new patients rather than a supply rep or a mother asking about a form. The owner buys on the strength of a demo where the AI handled a scripted call beautifully, then spends the next year unable to answer the only question that matters: did this book appointments that we would otherwise have lost?

Without a baseline, two bad outcomes are guaranteed. The first is that the tool works, nobody can prove it, and it gets cut in a cost review nine months later. The second is worse: the tool quietly mishandles calls, the front desk team notices and says nothing because leadership is visibly in love with it, and the practice renews. We would not sign anything until we had four weeks of call data in hand. Four weeks costs nothing. A year of unmeasured subscription costs $7,800 and your ability to make the next decision on evidence.

Hiring Got Easier, Which Is Bad News for the Vendors

The pitch rides on a labor panic that is real but aimed at the wrong seat. ADA reporting puts staffing at the top of the pile: about 62% of dentists named staffing shortages as the top challenge facing their practice in 2025. But read the next line. Difficulty hiring dental assistants was rated very or extremely challenging by roughly 70% of actively recruiting dentists in late 2024, down from about 87% in late 2022.

Two things follow. The hardest seats to fill are clinical, not administrative, and the market is loosening rather than tightening. The workforce numbers say the same thing: 222,740 hygienists and 387,790 dental assistants support those 205,088 dentists, and that clinical base is where your production actually comes from.

A dark hygiene column is the expensive problem. If one column produces $900 a day and you lose 200 days to a vacancy, that is $180,000 in production you never billed. Set that against the $44,001 of imaginary front desk savings and the priority is not close. No AI receptionist has ever scaled a quadrant.

The labor market for the front desk, meanwhile, is softening on its own. Federal projections have receptionist employment declining 2% from 2025 to 2035, with roughly 105,100 openings a year driven entirely by replacement, at a median wage of $38,010. Dental practices pay above that general median, at $20.34 an hour against $18.27 nationally. The seat is not disappearing from your office. It is churning, which is a different problem with a different fix.

Three Questions That Kill Nine Vendors

Ask these on the demo call, before the pricing slide, and watch how fast the room narrows.

• What happens on the call it cannot handle? You want a named escalation path to a human in your office, with the context carried over, not a promise that the model is getting better. "It will take a message" is a voicemail box with a subscription fee.
• Does it write to my practice management software, or do we retype everything? A bot that books into a separate calendar someone reconciles every morning has added a job, not removed one. Ask for the write permission, in your system, named by version.
• Will you quote insurance benefits to a patient, and who pays when the quote is wrong? This is where the whole category gets quiet. If the tool will read coverage off a verification feed and say a number out loud, you need to know whose write-off that is. We would not sign with a vendor that will not put an answer in writing.

Pair that with the pricing structure. A flat monthly fee pays the vendor whether or not the thing books anything. Per-booked-appointment pricing puts their incentive next to yours. Very few are offering it. Ask anyway, and note who flinches.

Do This Before Your Next Demo

Pull four weeks of call logs this week and get three numbers: total inbound calls, calls that went unanswered, and calls that arrived outside business hours. Then sit with whoever runs your front desk and list every task on that person's plate, marking which ones a phone bot touches. If phones turn out to be under a third of the list, you are shopping for revenue capture, not labor savings, and you should price and measure it that way. Budget $7,800 as a marketing test with a 90-day kill switch, not as a payroll line.

We are tracking one thing above all in the next ADA usage cycle: whether reception AI closes the gap on insurance verification, or falls further behind. If reception crosses 20% adoption while verification stalls, we were wrong about which task goes first and we will say so. What would change our mind faster is a vendor that prices per booked appointment and indemnifies its own benefit quotes. Nobody has done both. Your front desk is not nervous because software can answer a phone. It is nervous because for twenty years nobody bothered to measure what the phone was worth.

Before you price any of this, find out what your front desk already costs you as a share of collections. Most owners carry administrative payroll at a percentage they have never actually calculated, which is exactly how a $7,800 subscription gets approved to fix a problem that was never quantified. Pull your overhead breakdown and run your own numbers before the next demo lands in your inbox.