This article is based on insights shared during a Provet-sponsored Level Up Summit panel featuring Dr. Karen Bolten, Adam Wysocki, Jack Peploe, and Dr. Lior Kerner.
AI is already in your practice. It’s been running in some of your lab work, your cytology, your urinalysis, and parts of oncology for years. Dr. Karen Bolten, founder of The Business Vet, made the point on a recent Provet-sponsored Level Up Summit panel, "There's been things in practices for several years now that have been AI that I think a lot of people don't know."
And then there is the loud kind. Adam Wysocki, founder and principal consultant of VetSoftwareHub, said "Probably 50% of the doctors that I talk to are either using scribes, have played with scribes, or are interested in scribing."
So the question is no longer whether AI belongs in your practice. It is how you evaluate it properly - before a team member makes that decision for you by uploading a lab report into a personal ChatGPT account.
This guide pulls together a practical framework from the panel - Dr. Karen Bolten, Adam Wysocki, Jack Peploe of Veterinary IT Services, and Dr. Lior Kerner, Clinical AI Lead at Provet - on how to effectively evaluate AI tools for your practice.
Start with the mindset, not the tool
The biggest mistake is not picking the wrong AI product. It is assuming AI is like every other tool you have bought - vetted, regulated, and safe by default.
Jack Peploe was direct about it on the panel: clinics "massively underestimate in many cases the data exposure risk with these tools, potential hallucination risks, and the impacts of workflow disruption."
Dr. Bolten frames the gap clearly, "A lot of us have existed in practice with the assumption of, hey, we have this medication, this tool, whatever. It's been run through the wringer to make sure it works, that it's safe. And that's not true right now for AI."
Her advice? "You should be treating it like any other tool, any other education. You want to see the research."
That mindset is the prerequisite for everything below.
What questions should you ask an AI vendor before buying?
1. Where does your practice's data actually go?
Adam Wysocki advises that you should always start your questioning here: "It's the practice's data, right? They own it. They own the medical records. They own the conversations. Where is that data going? Is it going to a public model? Is it going to a private model? Can you get it out?"
The follow-up matters just as much. If a client comes back in six months and asks you to remove their conversation from the system, can you actually do that? "Is it possible to remove a single conversation out of that data if a patient or a client comes back to you in the future and says, hey, I'm not comfortable with this at all. I want to opt out?" questioned Adam Wysocki.
If the vendor cannot answer those questions cleanly, that is your answer.
2. Does the privacy policy pass the read-it-out-loud test?
Dr. Bolten describes herself as "personally obsessed with reading privacy policies and terms and conditions" because that is where the green flags and red flags actually live.
Her rule on training data? "You do not have to allow them to use your data to train."
What to look for in the policy:
- Opt-in by default, not opt-out
- A clear, simple way to opt out - ideally a few clicks, not a buried support ticket
- Time limits on data retention - never "forever"
- Named sub-processors so you know who else touches the data
3. Are the contract clauses fair?
Adam Wysocki advised to "Look at those indemnification clauses in contracts because if the company is saying, "Hey, we're not responsible for anything that happens with our system,” you should consider that before rolling it out in your practice."
A vendor confident in its product will accept some level of accountability. A vendor that wants zero responsibility is telling you something.
4. Is there a clear human in the loop?
"AI must assist clinical judgment and never obscure it," Jack Peploe remarked.
Dr. Kerner makes the same point about implementation: "There really needs to be a clear human in the loop." Any feature that pushes AI output straight into a clinical note, a discharge instruction, or a client message without review deserves scrutiny.
5. Does it fit your actual workflow?
A tool that adds another tab, another login, or another copy-paste step has to clear a high bar to be worth it.
Dr. Bolten's word for the killer of adoption is friction: "The more friction there is in your day, the less likely you are to use it… if one of your tools doesn't integrate with your PIMS, it's more friction."
Who is responsible if AI gets something wrong?
Three stakeholders, one shared risk.
Dr. Bolten on the clinical side argues that "If something goes wrong with the patient, it's your problem." That sits with the veterinarian operating the tool.
Jack Peploe widens the lens and adds that "Operationally, it's going to be the practice owner that's going to be responsible for the tool set. And if we look reputationally, it will be your practice brand that will be the one that suffers."
And as Dr. Kerner points out, this is not just a clinical problem. "If you send a discharge instruction and you're mentioning that your dog is a cat… that's reputational damage for you. And it doesn't even go through the clinical route."
This is why due diligence is a practice-wide responsibility, not a clinical one.
Bolt-on AI vs. integrated AI: how to choose
This is one of the most important decisions you will make, and the panel went deep on it.
The short version: bolt-ons are a useful way to test a single use case. Integrated AI is the lower-risk long-term play - because AI tools are only as good as the data and context they can see.
Dr. Kerner argued that "Context is really, really important. As these tools become more advanced, you would want to do more things with your data to save you more time. The context of a full patient's history, for example, or a history of invoices, reminders, appointments... where your data lives at the moment, usually on the PIMS, whether the PIMS will make its own features that work very well, or at the very least offer integrations that can take these third party tools and use them with your workflows."
His warning about the alternative: "Any tool that requires you to constantly keep another open tab to copy paste data from one window into another… that's a really, really strong point of friction. And as you know, we're going to go into the age of real agents… everywhere doing all kinds of actions for all kinds of things. You need to have a system that allows these agents to kind of walk in and out and do as you want them to do."
Jack Peploe's warning on bolt-on sprawl reinforces the same point from the operational side, "You could wake up in 18 months with like seven AI tools and no form of cohesion. You're going to have situations where staff might go, which system do I use for this? You know, duplicate documentation, you get the integration and risk of shadow IT, and then there's the rising support burden."
Bolt-ons still have a legitimate role, especially when you want to prove a concept fast before committing to a long-term direction. Here is the trade-off in one view.
The strongest position for most practices is a core platform with full data context, open enough to integrate the bolt-ons you genuinely need, and disciplined enough not to add the ones you don't. Coherence is what gives AI room to work.
What should different veterinary roles look for from AI?
.jpg)
What should a veterinarian look for in an AI scribe?
Four things matter most.
- Does it actually reduce consultation time?
Dr. Bolten ran a baseline measurement on her first scribe before switching it on, then compared. The time difference was "statistically very significant" but only because she had the control data to prove it.
- Does it let you be present with the client?
Dr. Kerner makes the point that veterinarians used to divide their attention between the client, the patient, and the keyboard. A scribe that works in the background lets you focus on the first two.
- Does it adapt to how you consult?
"It does require a bit of a change of mindset into how you do your consultations… the more that you say, the more you're giving the scribe the chance to capture more data," says Dr. Kerner. A good scribe rewards verbal, descriptive consulting.
- Does your setup capture clean audio?
The microphone is easy to overlook, but it matters. The cleaner the audio going in, the better the summary coming out. As Dr. Kerner puts it, "one small mistake is disregarding hardware." A decent microphone is usually inexpensive and can noticeably improve what the scribe captures.
What should a practice manager check before approving AI tools?
You are the one who carries the operational risk. Five checks:
- Data governance: where the data goes, who can access it, and how to get it back
- Contract terms: indemnification, opt-out rights, time limits on data
- Team adoption plan: who is the product champion, how is training handled
- Shadow IT audit: what is your team already using on personal accounts
- Pilot results: run one workflow for a month before rolling out
Dr. Bolten's reminder: "It's the same as you would do in a regular research project. You only want to change one variable at a time to know if it's working or not."
What should a receptionist know about AI client communication tools?
AI reception tools are spreading fast. Wysocki, relaying what practice teams tell him: "It has freed up so much time for us to be patient with the client standing at the desk… the people at the front desk tell me they don't have to worry about the phone tree blowing up."
What to push for as a receptionist:
- The AI handles routine calls and confirmations, not grieving clients or emergencies
- Clear escalation rules - when does the AI hand off to a person
- Visibility into what the AI said - you need to know what a client was told before they walk in
What AI mistakes do veterinary teams make?
Five patterns come up repeatedly.
- Skipping the pilot. Going straight from demo to rollout because the demo looked good.
- Not involving the team. No product champion inside the practice means resistance later.
- No internal AI policy. Without one, you have no answer to "can I paste this lab report into ChatGPT?"
- Treating scribes like perfect transcribers. Adam Wysocki: "AI scribes are not perfect transcribers… you still have to review what's captured."
- Ignoring shadow IT. This is the one most practices miss entirely.
Jack Peploe warned about the dangers of shadow IT, "Even if you haven't formally adopted AI, there's a very high chance that your team have, and they'll be using it in some way. That might be copy and pasting console notes into ChatGPT using their own personal AI accounts… It might be uploading lab reports."
"The biggest data risk for me and the biggest thing that I see isn't necessarily the adoption - it's the unmanaged adoption," Peploe concludes.
What does good AI adoption look like in a veterinary practice?
.jpg)
Once the guardrails are in place, the upside is real - and often shows up where you didn't expect it.
Dr. Bolten's scribe experience was statistically significant on time but the bigger surprise was qualitative: "I felt so much better after starting to use the scribe… I was less testy with the technicians… I was actually like able to get out of work when it was light out for once."
Dr. Kerner highlights another quiet gain from clinical AI - the scribe trains the clinician, not just the other way around. Over two or three months, his users were certain the product had been improved, when in fact "they were the ones that were improving the way that they conduct their consultations. They are saying more things out loud."
The pattern is consistent: when AI is chosen with discipline, it gives time back - to patients, to clients, and to the people doing the work.
Jack Peploe's closing line on the panel is the one to leave with: "The future of veterinary care isn't about AI or humans. It's how well the two work together."
Getting that balance right is not about enthusiasm. It is about doing the work upfront by asking where your data goes, reading the contract, running the pilot, and bringing your team along.
Staying ahead with AI
If you enjoyed this blog, check out the AI in Veterinary Practice Summit where you can earn three hours of CE credit, and get deeper insights from experts who know the clinical reality. Watch here >
Frequently asked questions
Is it safe to share my client and patient data with ChatGPT?
Not on a personal account. A personal ChatGPT account may not provide the data governance, contractual protections, or administrative controls your practice requires. Dr. Bolten's rule applies: "You do not have to allow them to use your data to train." If your team is already doing this, that is shadow IT and it is a data exposure risk.
Who is responsible if AI tools make a mistake in a clinical note?
Clinical accountability remains with the veterinarian using the tool. Dr. Bolten on the panel: "It's always the vet. This is no different than anything else in vet med. If something goes wrong with the patient, it's your problem." The practice owner carries operational responsibility, and the practice brand carries the reputational hit, but the clinical accountability sits with the clinician.
Can AI replace a veterinary receptionist?
No, but it is changing the role. AI reception tools are handling phone trees, after-hours appointment requests, and refill calls. Consulting on practice workflows, Adam Wysocki sees this as freeing reception staff to be more present with the client at the desk, not replacing them. The empathy work - grieving clients, emergency triage, end-of-life conversations - still belongs to a person.
How long does it take to see ROI from a veterinary AI scribe?
It depends on whether you measured before you started. Wysocki's recommendation: "Pick one workflow and run a pilot in your practice. Understand how long it took to do that workflow prior to AI and then time it." Dr. Bolten's first month with a scribe was rough during adoption, then settled into something she could measure and feel. She called the result "statistically very significant," and added: "I felt so much better after starting to use the scribe… I was actually like able to get out of work when it was light out for once."
What is shadow IT and why does it matter in a veterinary practice?
Shadow IT is any technology your team is using that the practice has not formally approved - typically free AI tools on personal accounts. Peploe considers it the single biggest AI risk to most practices: more dangerous than any tool you actively buy, because there is no oversight, no data agreement, and no audit trail. The fix is a simple internal AI policy.
What is the environmental impact of using AI in a veterinary practice?
It is real, but worth keeping in proportion. Training large AI models uses energy and water for datacenter cooling. The everyday use inside your practice, like a scribe transcribing a consult or a summary being drafted, carries a much smaller cost per task, though it still adds up across a busy clinic. Exact figures vary a lot between providers and are still debated, so treat any single "per query" number you see with caution and check it against the provider's own reporting.
There are a few practical things you can do. Ask where a vendor runs its datacenters and whether they are powered by renewable energy. And favor integrated tools over using lots of bolt-ons, since fewer overlapping systems mean less duplicated processing. It is the same instinct as data governance: ask the vendor to tell you their impact, and what they are doing to reduce it.
