The AI tech sales are everywhere! You’ve likely seen an AI system sales rep put together a package for your practice at some point this year, maybe had a presentation or two, and the price looked like it would pay for itself by spring. The demo went perfectly. Then someone in the room checks the output and finds errors. Now no one knows how to reply.
That moment is happening in a lot of budget meetings right now. AI tools are showing up for prior authorizations, notes, phones, scheduling, and even patient education. Some of this technology will save your practice real money, while others can quietly create risk. A good demo shouldn’t be the whole buying decision.
The best way to evaluate any ophthalmology AI vendor is to ask the same 10 questions every time: what job it does, what it replaces, what it learned from, who checks it, how it protects patient data, whether it identifies itself, how patients reach a person, whether it’s optional, what it measures, and what it hands back to your team.
Why does a standard set of AI questions matter for practice owners?
If you own or run a practice, you’re probably fielding AI pitches every week, or working to leverage your own LLM. Each vendor shows its best features and frames the risks its own way. Comparing them is hard when every conversation starts from a different place. A fixed list of questions gives you something to bring to partners, investors, and your compliance team, so the decision doesn’t rest on whoever saw the best demo.
In a 2026 iScience study of four AI models answering cataract patient questions, inaccurate answers ranged from 1.85% to 12.96% depending on the model, and the authors concluded human oversight is still required. And in the University of Michigan National Poll on Healthy Aging, 92% of adults 50 and older said they want to know whether information comes from a person or from AI.
What should you ask before you sign an AI contract?
Group the 10 questions into four steps. Use them for any AI tool, back office or patient-facing.
Step 1: Scope the job
What specific job does this tool do? “Improves patient engagement” isn’t a job. “Drafts prior authorization letters for surgeon review” is. Ask for the narrowest honest description.
What does it replace, in hours or headcount? Ask the vendor to show the math, then check it against your own staff time. Who is going to oversee your new system going forward? If the savings depend on a role you can’t actually reduce, the ROI case is weaker than it looks.
Step 2: Check the safety
What was it trained on? A general-purpose chatbot learned from the open internet. A tool built for cataract care should be able to tell you what clinical or patient data shaped it, and how that data relates to your patients.
Who checks its work, and how often? Every AI tool makes mistakes. Even in a randomized trial of ambient AI scribes in NEJM AI, the authors noted that occasional inaccuracies require ongoing vigilance. Ask who reviews the output, and what happens when they find an error. Utilizing staff to babysit inadequate AI tools costs more in the long run.
Will you sign a business associate agreement? If the tool touches patient information, HHS guidance says your practice needs written assurances, in the form of a business associate agreement, before sharing it. Ask where data is stored, who can see it, and whether it’s used to train the vendor’s models. Adequate compliance coverage typically comes at a price, be aware of all the costs before you sign.
Step 3: Protect the patient experience
These questions matter most for any AI that talks to patients.
Does it say it’s “AI,” every time? Patients deserve to know who they’re talking to. If the tool records calls, it should say that too.
Can a patient reach a trained person at any point? “Press zero for the front desk” doesn’t count. The handoff should go to someone trained to continue the same conversation.
Is it optional for your practice and for each patient? Some patients like texting with AI at midnight, while others want a human voice. A good tool lets them choose.
Step 4: Demand proof
What outcomes does it measure? Engagement and message counts are the numbers most vendors lead with. Ask about outcomes your partners care about: patients reached, appointments kept, evaluations that move to surgery, and ATIOL adoption by surgeon.
What does it hand back to your surgeons? If a patient shares fears or vision goals with an AI tool, that information should reach the chart before the evaluation. Otherwise your surgeon starts from zero.
Here’s how Navigate answers them for patient education. With NavigateOS, Navi is an optional, AI-driven Virtual Education Assistant for cataract surgery. She learned from thousands of cataract patient conversations with Navigate’s trained Patient Navigators.
Navi doesn’t work alone. She says who she is every time and says the line is recorded. When a patient asks for a person, she hands off to a trained human Navigator, not to your front desk. She’s optional for your practice and for your patients. Our integrated AI system also coaches every human Navigator, or your staff, and every call is reviewed for quality afterward.
What happens if you skip the questions?
The risk usually doesn’t show up in the first month. It shows up later as a wrong answer a patient repeats in the exam lane, a data question from compliance, or an ATIOL adoption rate that slips while message counts climb. Older patients who feel brushed off by a bot may also stop answering. Then you’re paying for a tool and still making one phone call to patients who need four.
What does a confident AI decision look like?
You walk into the budget meeting with the same 10 answers for every vendor. Your back office tools have clear owners and review steps. Your patient-facing tools say what they are and hand off to real people. Your surgeons see each patient’s goals in the chart before the visit. When partners or investors ask about your AI plan, you can show them the questions, the answers, and the results.


