The scanners, printers and AI software are already in your dentist’s office. The research says appointments get shorter and more consistent. Almost nobody has checked whether your teeth end up better.
More than two in five American dentists now use AI for something in their practice, and around one in five runs it on X-rays. That comes from the American Dental Association’s own survey panel, published in July 2026.
So the interesting question isn’t whether digital dentistry has arrived — it’s what it’s doing for the person lying in the chair. Dig into the evidence and you find something lopsided. There’s a great deal of research on whether these machines are precise. There’s almost none on whether patients end up healthier.
Key takeaways
- Pooled across roughly 29,000 image tests, AI software correctly flagged about 85 out of every 100 decayed tooth surfaces and correctly cleared about 90 out of every 100 healthy ones — broadly comparable to a careful visual exam by a dentist.
- Computer-guided implant surgery places the implant around a millimetre from where it was planned, based on 67 clinical studies covering 5,673 implants in real patients.
- In a crossover trial where 20 denture wearers tried both kinds, people rated conventional dentures higher than 3D-printed ones for comfort, stability, speech and cleaning — but the printed sets took fewer appointments, and three participants chose to keep wearing them.
- The endpoint in nearly all of this research is accuracy, measured in millimetres or percentages. Pain, cost, chewing ability and how long the work lasts are mostly not measured at all.
- Fewer than one in 20 US dentists currently uses AI to decide on treatment, and roughly four in five say they don’t plan to.
The software reading over your dentist’s shoulder
The most common way a patient meets AI in dentistry is without noticing. You get bitewing X-rays, and before your dentist looks at them, software has already outlined anything it thinks is decay.
In August 2025, a team led by Sarah Arzani at Isfahan University of Medical Sciences, with Paul Dummer, emeritus professor of restorative dentistry at Cardiff University, pulled together 14 systematic reviews of that software and pooled the usable numbers — roughly 29,000 individual image tests drawn from 20 original studies.
The software caught most of the decay that was there and correctly left most healthy surfaces alone. Put in terms a patient might care about: if it flags a surface, there’s about a four-in-five chance decay is really there. If it clears one, there’s roughly a one-in-twenty chance something was missed.
That’s about where a careful visual examination by a dentist lands, according to Cochrane’s own review of visual caries detection. Different studies, different reference standards, so it isn’t a head-to-head — but it puts the claim in proportion. This software isn’t a superhuman detector. It’s a second pair of eyes that doesn’t get tired at four in the afternoon.
Arzani’s team is direct about the ceiling. AI can assist in diagnosing decay, they write, but
“it should not be a substitute for human judgment.”
Arzani et al., PLOS ONE, 2025 doi.org/10.1371/journal.pone.0329986
They also flag something awkward about the training data. Decay showed up in about a quarter of the images these systems were tested on, which is lower than what dentists encounter in the real world. Train and test on an unusually healthy set of images and the numbers look better than clinic performance would.
The implant that lands where it was planned
Guided implant surgery is the most mature piece of digital dentistry, and the evidence base is genuinely large. A 2024 meta-analysis in BMC Oral Health, led by Angkoon Khaohoen at Chiang Mai University, pooled 67 clinical studies covering 5,673 implants placed in actual patients.
The pattern holds across methods. Whether the surgeon works from a printed drill guide, a real-time navigation screen or a robotic arm, the implant ends up roughly a millimetre from the planned spot at the gumline, a little further out at the tip, and tilted by around three and a half degrees. Fully guided protocols — where the guide controls every step from the first drill to final placement — beat the cheaper pilot-guided approach, which only positions the initial hole.
Robots looked best of all. They also appeared in just two studies covering 44 implants between them, which is why the authors say the robotic advantage should be treated with caution until more research exists.
The more useful observation sits near the end of their discussion. Nearly all of this work has chased accuracy above everything, rather than what patients report — measures the team argues
“should constitute the virtual endpoint of clinical trials.”
Khaohoen et al., BMC Oral Health, 2024 doi.org/10.1186/s12903-024-04033-y
Millimetres are easy to measure. Whether the implant hurt less, cost less or lasted longer is harder, and largely nobody has done it.
Dentures printed instead of packed
Here the news is more mixed, and more interesting for it.
Researchers at Kanagawa Dental University in Japan ran a crossover trial between 2017 and 2020. Twenty denture wearers each received both a conventional set and a 3D-printed set, in randomised order, and rated them. It’s the right design for this question, because every participant acts as their own comparison.
Conventional dentures won on comfort, stability, speech and ease of cleaning, and on overall satisfaction. The printed ones weren’t close on those measures. But they required meaningfully fewer trips to the clinic, and participants reported less social difficulty while wearing them. Asked at the end which set they wanted to keep, 12 chose conventional and three chose printed.
The team’s conclusion is carefully worded: satisfaction with printed dentures may be somewhat lower, yet one in five wearers used them daily by choice, so printed dentures
“may have comparable practicality and efficacy.”
Ohara et al., Journal of Prosthodontic Research, 2022 doi.org/10.2186/jpr.JPR_D_21_00048
Worth knowing before you read a clinic website claiming printed dentures are simply better. In the one trial that asked patients directly, most preferred the old way — while spending more days in the chair to get there.
What the three studies measured
| Area | Evidence base | Headline result | What was actually measured |
| AI decay detection on dental images | 20 original studies pooled from 14 systematic reviews; 29,423 diagnostic tests | Pooled sensitivity 0.85 (95% CI 0.83–0.93); specificity 0.90 (0.85–0.95); AUC 0.86; positive post-test probability 79%, negative 6% | Whether software labels a tooth surface correctly. A diagnostic-accuracy outcome, not a health outcome. Caries prevalence in the test images was 27.3%, below real-world rates. |
| Computer-guided implant placement | 67 clinical studies, 5,673 implants (19 randomised, 23 prospective, 21 retrospective, 4 case studies) | Mean deviation 1.11 mm at entry (95% CI 1.02–1.19), 1.40 mm at apex (1.31–1.49), 3.51° angular (3.27–3.75). Robotic subgroup 0.81 mm / 0.77 mm / 1.71° from 2 studies, 44 implants. | Distance between planned and achieved implant position. A surrogate for prosthetic fit and surgical safety, not implant survival or patient comfort. |
| 3D-printed vs conventional complete dentures | 1 randomised crossover trial, 20 participants enrolled, Kanagawa Dental University, 2017–2020 | Conventional rated superior for phonetics, ease of cleaning, stability, comfort and general satisfaction. Fewer clinic visits and less social disability with printed. 12 preferred conventional, 3 preferred printed. | Patient-rated satisfaction on a 100 mm visual analogue scale, quality of life, and number of appointments. A genuine patient-reported outcome — the exception in this set. |
Sources: Arzani et al. 2025 (PLOS ONE); Khaohoen et al. 2024 (BMC Oral Health); Ohara et al. 2022 (Journal of Prosthodontic Research).
Where this fits
The ADA has moved faster on governance than on endorsement. It published a white paper on AI in dentistry in December 2022, followed by a technical report and, in 2025, the first US national standard on the subject. Robert Faiella, DMD, who chairs the ADA’s working group on augmented intelligence, frames it as a professional obligation rather than a technology question — dentists can use these tools, he says, while
“upholding their ethical and professional obligations to provide safe, effective and equitable care.”
Robert Faiella, DMD, chair, ADA Standards Working Group 12.7, in ADA News, February 2025 adanews.ada.org/ada-news/2025/february/what-are-the-standards-for-ai-use-in-dentistry/
Dentists themselves have drawn a sharp line. The ADA’s Health Policy Institute found a substantial minority using AI for imaging and scheduling, and a hard stop at clinical judgement — fewer than one in 20 uses it for treatment recommendations, and about four in five say they never will. One respondent summed up the reservation in a sentence:
“It has its place but still must be monitored and verified by humans.”
Anonymous respondent, ADA Health Policy Institute dentist panel, July 2026 ada.org/resources/research/health-policy-institute/dental-practice-research/dentists-ai-usage-and-attitudes
That gap is the real story. These tools have cleared the accuracy bar and are being adopted for the tasks where accuracy is the whole job. They haven’t cleared the outcomes bar, and dentists are behaving accordingly.
What it means for you
If you’re a patient
If your dentist shows you an AI-marked X-ray, treat it as a prompt for a conversation rather than a verdict. Reasonable things to ask: is this a lesion you’d have watched anyway, and what changes if we review it in six months? Early flags can lead to fillings that a dentist monitoring the tooth might never have placed. For implants, guided surgery is well supported and generally means fewer surprises, though it usually adds a scan and a planning fee. For dentures, the printed route currently buys you fewer appointments rather than a better-fitting denture — a real benefit if travel or time off work is the hard part for you, less so otherwise. Any of this is worth raising at your next check-up rather than acting on alone.
If you’re a dentist
The honest chairside line is that the software helps you see and you decide, and patients respond well to being told the flag is a starting point. On guided surgery, the fully guided protocol is the one the pooled evidence supports; half-measures give up much of the advantage for a modest saving.
What nobody knows yet
Nobody has run a trial where patients are randomised to AI-assisted or unassisted examination and then followed to see who ends up with more fillings, fewer fillings, or healthier teeth in five years. Nobody has shown that an implant placed a millimetre more accurately survives longer or fails less often. Nobody has tested whether printed dentures improve with better materials and a longer settling-in period, which the Kanagawa team suspects matters. And the AI training sets still under-represent how much decay is in the general population, so real-world performance remains an open question rather than a settled one.
The bottom line
Digital dentistry has earned its accuracy claims. AI reads X-rays about as well as a dentist looking carefully, guided surgery puts implants within roughly a millimetre of plan, and printed dentures cut the number of visits. What none of it has yet earned is the claim that patients end up with better teeth, because that question has barely been asked.
Frequently asked questions
Is AI better than my dentist at spotting cavities?
Roughly comparable, not better. Pooled results across about 29,000 image tests put AI at around 85% detection of decayed surfaces, similar to a careful visual exam. It works as a second check on X-rays. No study has shown that adding AI leads to healthier teeth, and the dentist remains responsible for the diagnosis.
Should I pay extra for guided implant surgery?
Guided placement is well supported: across 67 studies and 5,673 implants, implants landed around a millimetre from the planned position, and the fully guided version performs best. What’s unproven is whether that accuracy means implants last longer or fail less often, so weigh the added cost against your own case with your dentist.
Are 3D-printed dentures as good as conventional ones?
Not yet, on comfort. In a trial where 20 people wore both, conventional dentures scored higher for comfort, stability, speech and cleaning. Printed dentures needed fewer appointments, and three of the 20 preferred them. If cutting clinic visits matters most to you they’re reasonable; if fit is the priority, conventional still leads.
Can AI replace a dentist?
No, and dentists are saying so most firmly. In ADA survey data, fewer than one in 20 US dentists uses AI for treatment recommendations and about four in five don’t plan to. The ADA’s standards work treats AI as augmented intelligence — a tool assisting the clinician, who stays responsible for diagnosis and care.
Does an AI-flagged spot on my X-ray mean I need a filling?
Not necessarily. A flag means the software detected something resembling decay, and roughly four in five flags correspond to real decay. Early lesions can often be monitored or treated without drilling. Ask how deep it is, whether it’s progressing, and what happens if you wait and review it next visit.
Sources
- Arzani S, Karimi A, Iranmanesh P, Yazdi M, Sabeti MA, Nekoofar MH, Kolahi J, Bang H, Dummer PMH. Examining the diagnostic accuracy of artificial intelligence for detecting dental caries across a range of imaging modalities: An umbrella review with meta-analysis. PLOS ONE 2025;20(8):e0329986. Open access, CC BY. https://doi.org/10.1371/journal.pone.0329986
- Khaohoen A, Powcharoen W, Sornsuwan T, Chaijareenont P, Rungsiyakull C, Rungsiyakull P. Accuracy of implant placement with computer-aided static, dynamic, and robot-assisted surgery: a systematic review and meta-analysis of clinical trials. BMC Oral Health 2024;24:359. Open access, CC BY. https://doi.org/10.1186/s12903-024-04033-y
- Ohara K, Isshiki Y, Hoshi N, Ohno A, Kawanishi N, Nagashima S, Inoue M, Kubo D, Yamaya K, Inoue E, Kimoto K. Patient satisfaction with conventional dentures vs. digital dentures fabricated using 3D-printing: A randomized crossover trial. Journal of Prosthodontic Research 2022;66(4):623–629. Open access, CC BY-NC. https://doi.org/10.2186/jpr.JPR_D_21_00048
- American Dental Association Health Policy Institute. Dentists Use AI to Make Appointments More Efficient, but Draw the Line at Clinical Decision Making. HPI Perspectives, July 2026. ada.org/resources/research/health-policy-institute/dental-practice-research/dentists-ai-usage-and-attitudes
- American Dental Association. What are the standards for AI use in dentistry? ADA News, 3 February 2025. adanews.ada.org/ada-news/2025/february/what-are-the-standards-for-ai-use-in-dentistry/
Leave a Reply