The 5 questions every dental practice owner should ask before buying any AI tool

The dental AI market is producing a new vendor every few months. Each one has a polished demo, a case study from a practice you have never heard of, and a sales rep who will tell you their tool does everything. Most will not follow up honestly when you push on the detail.

This is not a product roundup. The tools worth evaluating change too quickly for any list to stay current, and recommending specific vendors is not something we do — we have no commercial relationships with any of them. What does not change is the framework for evaluating any vendor before you sign. These five questions work regardless of which tool is being pitched, which workflow it claims to fix, or what price point it is positioned at.

Run through them in order. The questions that kill bad deals do so quickly.

Question 1: "Does this tool require a signed Business Associate Agreement, and can you provide one before we share any patient data?"

This is the first question because it is the only one with a binary pass/fail answer. Any AI tool that touches protected health information (PHI) — patient names, appointment records, clinical notes, treatment data, insurance details — requires HIPAA compliance. The legal mechanism that establishes that compliance between your practice and the vendor is a Business Associate Agreement (BAA).

A vendor who cannot produce a BAA, who delays signing it, or who suggests you can start using the tool and sort the paperwork later is not a vendor you should be working with. Full stop. The HHS Office for Civil Rights has issued enforcement actions against practices that used non-BAA-covered tools with patient data, including tools the practice owner believed were compliant based on vendor marketing. The liability sits with you, not the vendor.

If a vendor says their tool does not handle PHI and therefore does not require a BAA, ask them exactly which data points their system receives, processes, or stores. If the answer includes anything linkable to an identified patient, the tool handles PHI. Verify this with your compliance advisor before proceeding.

The direct recommendation: Do not provide patient data to any AI vendor, even for a demo or trial, until a signed BAA is in place. This is not negotiable, and a good vendor will not ask you to treat it as such.

Question 2: "Does this integrate natively with [your PMS], or will staff need to enter data in two places?"

The most common reason dental AI tools get abandoned within six months of purchase is not that they fail to deliver the promised outcome. It is that the workflow required to use them creates more work than it saves. The typical failure mode: the AI tool and the practice management software (PMS) do not talk to each other, so someone has to bridge the gap manually. That someone is usually the receptionist or treatment coordinator who is already stretched thin.

The dominant PMS platforms in US dental practice — Dentrix, Eaglesoft, Carestream Dental, Open Dental, Curve Dental — each have different API architectures and integration ecosystems. Not every AI vendor has built integrations with all of them. Some have built integrations that work in one direction only (data goes in but does not come back). Some advertise integrations that are technically possible but require configuration your IT support cannot handle.

During the demo, do not let the vendor show you the tool in isolation. Ask them to demonstrate it running live alongside your PMS in a test environment. If they cannot do that, ask them to show you a recorded walkthrough of the integration with your specific PMS. If they cannot do that either, assume the integration does not exist in a usable form.

The direct recommendation: The integration question should come before the features question. A tool with strong features and no PMS integration will cost you more in staff time than it saves. A tool with adequate features and seamless integration will actually be used.

Question 3: "What is the documented time saving or outcome improvement for a dental practice at our volume, and what is the source?"

Every AI vendor claims their tool saves time. The claims range from plausible (20 minutes per day on clinical notes) to improbable (50% reduction in treatment planning time). The difference between a real claim and a marketing number is evidence: was the saving measured in a real practice, over what time period, with what patient volume, and has it been published anywhere independent of the vendor?

Vendor-funded case studies are not worthless, but they are the weakest form of evidence. A case study from a single practice selected by the vendor tells you what is possible in optimal conditions, not what is typical. Peer-reviewed research is stronger. Independent benchmarking — such as ADA survey data on practice efficiency, or MGMA benchmarks adapted for dental — is stronger still.

For clinical note AI in dentistry specifically, published research on ambient documentation tools has shown time savings of 15 to 22 minutes per clinician per day in general practice settings (Journal of the American Dental Association and similar publications through 2025). If a vendor is claiming significantly above that range, ask them to show you the measurement methodology. If they are claiming below it, ask why.

For appointment scheduling AI, look for no-show reduction data. The ADA's 2025 Dental Practice Health Report found that the average dental practice loses 8 to 12% of appointment revenue annually to no-shows and late cancellations. A credible AI reminder tool should be able to show you a percentage reduction figure from a comparable practice, with the measurement period stated.

The direct recommendation: Ask for the evidence before the demo. If the vendor sends you a case study PDF featuring a practice you cannot verify exists, treat it with appropriate skepticism. Ask for contact details of a reference practice in your state and specialty. A confident vendor will provide them.

Question 4: "Where is patient data stored, who can access it, and is it used to train your AI models?"

This question matters more than most dental practice owners realize, and it matters more than most AI vendors want to discuss in a sales call. There are three distinct issues bundled into it.

Storage location is a HIPAA requirement. Patient data processed or stored by a US dental practice must, with narrow exceptions, reside on US-based servers. Some AI vendors — particularly those with development operations outside the US — route data through non-US infrastructure. This creates compliance exposure regardless of what the marketing materials say. Ask for the data processing addendum and check where servers are physically located.

Access controls determine who, within the vendor's organization, can see patient data associated with your practice. Reputable vendors operate with strict role-based access controls and audit logging. Ask whether their staff can access patient data and under what circumstances. If the answer is vague, probe until it is specific.

Model training is the most significant issue for 2026. Many AI platforms use customer data to improve their models by default. For general consumer software this is normal. For a dental AI tool processing patient records, it means your patients' data may be used to train a system that serves your competitors. Some vendors allow you to opt out of model training; others make it a contractual default that requires explicit negotiation to remove. Ask directly whether your data is used for model training, and if so, what the opt-out mechanism is. Get the answer in writing before signing.

The direct recommendation: Request the vendor's data processing agreement and privacy policy before the final vendor review, not after signing. Have your compliance advisor or healthcare attorney review the model training clauses specifically. This is a 30-minute review that can prevent a significant compliance problem.

Question 5: "Can we run a full-functionality trial for 30 days before committing to an annual contract?"

The final question is also the simplest, and vendor responses to it are reliably revealing. A vendor confident in their product's performance in real dental practice workflows will offer a trial. A vendor whose product performs well in demos but struggles in day-to-day practice will find reasons to avoid one.

A meaningful trial has three characteristics. First, it runs at full functionality — not a feature-limited demo environment, not supervised sessions with vendor support on hand. The tool should run as it would on day 201, when your trial period is long over and vendor attention has moved to the next prospect. Second, it runs long enough to encounter normal operational variation: busy weeks, quiet weeks, staff off sick, patient complaint workflows, end-of-month reconciliation. Thirty days covers this; fourteen days may not. Third, it is measured against the specific outcome the tool was purchased to improve. If it is a clinical note tool, track note completion time before and after. If it is a recall system, track reappointment rate for overdue patients during the trial period versus the prior 30-day baseline.

Practices that skip the proper trial and sign annual contracts on the back of a demo account for a significant proportion of abandoned AI implementations. The tool does not fail because AI does not work in dentistry. It fails because no one tested whether this tool works in this practice before committing $5,000 to $15,000 to a year of it.

The direct recommendation: Make the 30-day full-functionality trial a contractual condition of purchase. If the vendor will not agree to it, that is your answer. Move to the next vendor on your list.

What to do with the answers

Run these five questions across every vendor you are evaluating and score them honestly. A vendor who clears all five — provides a BAA, demonstrates real PMS integration, shows published evidence, gives you a clear data governance answer, and offers a proper trial — is worth proceeding with. One who stumbles on two or more is telling you something important about either the maturity of their product or the honesty of their sales process.

The dental AI market will consolidate over the next two to three years. The tools that survive will be the ones that hold up to this kind of scrutiny. Evaluating them now with real rigour protects your practice from becoming an early casualty of a market that is still maturing.

If you are evaluating multiple tools simultaneously and want an independent assessment against the CARE Framework — scored against your specific workflows, patient volume, and practice management setup — that is exactly what the AI Opportunity and Growth Assessment produces. It takes the vendor evaluation off your desk entirely.

Related: Why 73% of AI tools go unused in healthcare practices · From $0 to $150: the real cost of AI for an independent practice · The AI Opportunity and Growth Assessment