AI in practice: five questions vets are asking

Few topics divide and excite vets quite like artificial intelligence (AI). That burning question, “is AI going to replace us?”, continues to make its way into public debate.

With veterinary services under relentless pressure from every direction, it is no surprise that new technologies are often met with raised eyebrows.

After speaking with our own clinical team, whose careers span every corner of practice, this question rarely stands alone. It usually opens the door to a wider set of concerns and curiosities about where AI fits, and what it means for our everyday work. It’s time to delve into the five questions I hear the most, and I suspect they could be on the minds of vets well beyond our own team, too.

1. How accurate can AI be?

Having been a vet for 13 years, I have seen owners stumble across misinformation, leaving them more stressed than when they first began their search for help. However, we simply cannot ignore the fact that more than 60% of owners browse problems online before seeing a vet, calling for the need to guide them towards safe and clinically informed resources. Another way that owners seek advice is through “large language models”, or LLMs for short. This is a type of artificial intelligence programme that is trained to understand, generate and respond to human language so it can answer questions, write text or have conversations. Think ChatGPT and Google Gemini with instant and detailed answers.

To explore the impact of veterinary-led training, we conducted blind testing of Vet-AI’s large language model. Independent veterinarians reviewed anonymised cases and assessed triage recommendations. This model correctly identified the appropriate endpoint in 81% of cases, compared to two widely used models that scored 75% and 56% accuracy when tested in the same conditions.

However, the work does not stop here. Veterinarians remain actively involved beyond the initial development of a tool like this by manually assessing the outputs against a defined set of clinical and communication criteria. One of the veterinary surgeons involved in this process, Samantha Walker, reported feeling a boost in confidence after seeing how an LLM can blend empathy, clinical information and clear instructions. From what we have seen, AI performs far better when it is shaped by those who live and breathe this area of work.

2. Is there a way to prevent AI from making the wrong call?

While AI can enhance veterinary care by offering rapid insights and supporting diagnostic decisions, we cannot underestimate the critical role of human oversight.

AI systems, no matter how advanced, operate within the limits of their training. At its core, AI relies on patterns learned from data that we provide. These systems then operate within the boundaries of that information. Cases can fall outside those boundaries, due to rare presentations or complex and multifactorial conditions. This is where clinical expertise becomes even more critical, to guide decisions where data alone may fall short.

One example of how these systems are being refined is through the work of Chantal Serra, senior veterinary developer at Vet-AI, with nearly two decades of clinical experience. Alongside a team of vets, she is leading a project focused on improving the reliability and responsiveness of AI systems in urgent care settings.

She explained: “If a dog has ingested a raisin, a situation that requires immediate, in-person treatment, then the triage tool would advise the owner to contact their local clinic straight away.

“For cases where the danger is unclear, we are introducing a system to flag these queries as urgent, ensuring they are prioritised by the veterinary team.”

AI is, therefore, not static; it evolves over time as models are updated, retrained or improved with new data, knowledge and user feedback. Just like all of us, these systems are always learning, and we need to work together to find the blind spots.

Owners often turn to online resources out of care and concern for their pets, but with so much conflicting advice circulating on social media in particular, the margin for error can be dangerously thin.

We need to keep establishing robust frameworks that ensure these systems either make the right call independently or, when uncertainty arises, escalate the case promptly, so it lands in the hands of a qualified clinician.

3. Will AI change how I consult with clients?

Many vets worry that AI tools could depersonalise care or add more to their workload. AI can guide, support and inform, but it can’t palpate a mass, administer vaccinations or comfort a distressed pet owner.

AI has the potential to reshape consultations, but not by taking on the role as the vet. One example is our AI-generated summary feature, which collects information from owners before they join an online consultation. This means that owner observations are ready to go before the vet connects and key questions have been asked in advance, speeding up the consultation process on both ends. The feature was rolled out to approximately 100,000 UK pet owners, and feedback was collected to understand the impact on vets’ daily workflows.

Feedback has been highly encouraging to date. Veterinary surgeon Amy Greenhill found that the feature made things “easier, smoother and less stressful” during a full day that left little time for manual transcription.

Veterinary surgeon Michelle Markram found it similarly useful, noting that she jumped straight into an appointment without needing to go back and forth gathering the pet’s history. At no point does the AI tool take over the consultation; rather, it acts as a virtual assistant to our busy vets, giving them time and space to deliver care.

The key here is to keep speaking directly with vets and maintaining an open dialogue. What works, what isn’t working, and what do we need to consider? That will very much steer what happens next, so that changes are solving problems, not adding new ones.

4. Could online technology displace in-person practices?

A large number of the vets at Joii have adopted a hybrid working arrangement, splitting their time between in-person practices and remote consulting from home.

I have been asked whether online tools, including those driven by AI, will pull pet owners away from their in-person clinics – especially as rising costs continue to bite. A 2024 report from the RSPCA found that 36% of pet owners were taking steps to reduce the cost of care, with a worrying 4% not going to the vet and 3% considering rehoming. One case that has stayed with me is Julie Barker, the owner of a young spaniel named Sheila. Their story offers a timely reminder that technology can support, and not supplant, in-person care. Julie used telemedicine to discuss Sheila’s symptoms at a time when her local practice was closed, including lethargy and incontinence. She was becoming increasingly reactive, to the point where Julie considered the possibility of rehoming.

After checking Sheila’s gums and running through vitals, our remote vet advised the need for immediate, out-of-hours care. Ultimately, Sheila urgently needed IV steroids for Addison’s disease and was treated at The Kettering Vets. Julie was advised that if she had waited until morning, as was her original plan before using Joii, her dog likely would not have survived the ordeal.

Addison’s disease is notoriously difficult to diagnose early; around 30% of dogs with the disease are diagnosed at the time of a crisis. This highlights the importance of having online tools that help owners navigate their pet care from home.

What this also shows us is that if tools are created in a safe and regimented way, then features such as chatbots, telemedicine consults and imaging tools can help to ease the load on our colleagues, while nudging owners into clinics when they need it most. In the UK, we now have 28% of UK adults owning a dog and 24% owning a cat, so extra pairs of hands are not just welcome, but needed, now more than ever.

5. Should I even be using AI?

It seems too easy to imagine daunting rows of computers and machines tapping away, doing all the work for us, slowly pushing us to the sidelines. But the reality is much more human.

AI is not one single system running in the background; it is a collection of specialised tools designed to solve specific problems. From burnout to backlash, there is no shortage of these to address. One common theme that runs through them all is lack of time.

A team of more than 100 people still cannot escape the pressure of time. But we focus our efforts on the work we believe should remain in capable human hands: vets speak directly with owners, customer care supports pet parents, clinical teams collaborate on research, and trial teams oversee participant progress. In everything we do, it is not about a strict requirement to use AI at every turn and corner, but a genuine interest in making life that little bit easier for such a skilled workforce.

It is also worth reminding ourselves that there may be instances where we are using AI, without even realising it. If you are listening to a recommended Spotify podcast, using Google Maps to travel to a call-out appointment, relying on Google Docs to correct a document or hoping your next social media post resonates with clients, then AI is already part of your everyday life.

More than 50% or more of vets are reporting moderate to high levels of burnout. Meanwhile, according to BVA research, one in five vets are now using AI technology in their work. With all of this setting the scene, we understand why some vets are wary. Trust, ethics, job satisfaction, time and professional identity all matter.

We have only scratched the surface here, but as AI becomes more embedded in human health care, we expect its role in veterinary care to grow too. But that growth must be veterinarian led, transparent, inclusive and focused on positive outcomes for both animals and those who care for them.

If technology could lighten the load, even by a margin, isn’t it worth asking not why we would use it, but why we wouldn’t?

• This article appeared in Vet Times (2025), Volume 55, Issue 38, Pages 16-18.