AI tools that flag caries on your radiographs are everywhere now. Their pitch: more reliable detection, an eye that never gets tired, fewer missed lesions.

All true. And yet, that is not really what they are selling.

Here is what you should absolutely consider before spending your money, and how to actually make a better use of it:

Look at the marketing closely. The number that comes up again and again is not sensitivity or specificity. It is case acceptance. The patient sees the lesion outlined in red on the screen, understands it, and says yes to the treatment plan. Vendors in this space report case acceptance lifts in the 20 to 25 percent range.

And that makes sense. Let's be honest: most dentists do not need an AI to spot a carious lesion on a bitewing. It is a nice safety net, a second set of eyes at the end of a long day. But it is not a clinical necessity. What these tools really deliver is a way to show the patient what you already see with your trained eye.

The value is not in the detection. It is in the illustration.

Here is where it gets interesting. If the real product is illustration, do you need a $250 to $500 per month subscription, roughly $3,000 to $6,000 a year, to get it?

Because the underlying technology is not locked in a vault. Caries detection models built on open computer vision frameworks already exist in the research world, trained on public radiograph datasets, with code freely available. In peer reviewed studies, some of these models perform close to expert level. What they lack is everything around the model:

So imagine a stripped down version. A simple tool, maybe $30 a month, maybe free. It makes a rough first pass on the radiograph, then you correct it: you delete the false positives, you add what it missed, you highlight exactly what you want the patient to see. Less polished, less accurate on its own, but in the end the annotated image looks the same on the screen.

Honest question: if that tool existed on your desktop tomorrow, would you use it?

Notice what your answer reveals. If you would, then you agree that the diagnosis was never the product. You remain the diagnostician, the tool is a laser pointer. And since you correct everything anyway, the raw accuracy of the model stops being the deciding factor.

So, is the cheap version enough? Honest answer: it depends. On two things that have nothing to do with the software: who you are, and who is sitting in your chair.

If you are already a strong communicator, if your treatment plans land, if you can walk a patient through a radiograph with confidence, you need these tools less. A good closer closes, with or without red overlays. For you, the AI is a bonus: it looks modern, it adds polish, but it is not moving your numbers much.

If communication is not your strong suit, the equation flips. These tools do part of the talking for you. You show the image, the image speaks. You are essentially delegating the "explain and convince" step to the machine. Paradoxically, the weaker the presenter, the stronger the return on investment.

The second variable, often forgotten: the person in the chair.

There is no universal answer. There is a judgment call, patient by patient. Which is exactly your job.

One more nuance, and it is the one nobody in this industry has an incentive to point out.

Say you try one of these tools and the effect on your case acceptance is dramatic. Great news for production. But it is also data. It means that before the overlay, a meaningful share of your patients were hearing your diagnosis and not acting on it.

Sometimes the explanation is completely benign. Some patients simply process images better than words. Complex cases genuinely deserve a picture. Nothing to fix there, the tool is doing exactly what it should.

But sometimes the tool is a bandage. If patients only say yes once a machine confirms what you just told them, the real issue may sit upstream: how do your patients find you, and how much do they trust you before the radiograph ever appears on screen? A patient referred by a friend, who has been in your chair for ten years, rarely needs a red box to believe you. A patient who found you through an ad while comparing three quotes might.

Neither situation is a fault.

But they call for different remedies. One is genuinely solved by a $400 subscription. The other is solved by working on the relationship, the acquisition channel, the trust built before the exam even starts. And here is the trap: because the tool works, it can hide that difference from you indefinitely.

So treat the before and after as a diagnostic of your own practice, not just a sales metric. A modest lift means your communication was already doing the job. A spectacular one is worth a harder look in the mirror.

Whatever tool you use, expensive or cheap, never present your diagnosis and the AI's at the same time. Order matters:

The difference is subtle but huge. Presented together, the two diagnoses suggest you copied the machine. Presented in this order, the AI becomes a confirmation of your expertise. You stay the clinician, it stays the tool. The patient walks out with double confidence: in you, and in your practice.

There is a second life these tools have that a cheap alternative cannot replicate, at least in the US: claims. Annotated radiographs double as documentation that supports insurance submissions and reduces denials, and some of these platforms sit on both sides of the table, selling the same AI to payers who use it to review the very claims you submit. That changes the buying logic entirely, and it deserves its own email. Coming soon.

AI does not replace your eye. Used well, it makes your eye visible.

See you next week,

Salim, Founder at DentAI