Pet owners preferred ChatGPT answers to a veterinarian’s but the study did not test clinical accuracy

In a study from the University of Edinburgh, 68% of participants preferred answers generated by ChatGPT to responses written by a veterinarian when presented with two pet-health questions. AI responses were rated more highly for structure, perceived quality and empathy, and were also trusted slightly more often. But the study involved only one veterinarian, two clinical scenarios and GPT-4o, and did not independently assess whether the medical advice was correct.

The study explored a question that is becoming increasingly relevant to veterinary practice: how do pet owners and veterinary professionals perceive AI-generated medical communication when they do not know who wrote it?

The researchers selected two pet-owner questions from Reddit’s AskVet community. A veterinarian and ChatGPT using GPT-4o, May 2024 answered both questions under comparable conditions. The responses were anonymised and randomised before being presented in an online survey. Participants scored them for structure, quality and empathy, indicated whether they trusted the information and selected the answer they preferred.

Two-thirds preferred the AI response

A total of 201 people completed the survey, including 43 veterinarians, 70 veterinary students, 10 veterinary nurses, five receptionists, 15 researchers or medical professionals and 58 members of the public.

Across the two scenarios, 68.1% preferred the ChatGPT-generated answer, compared with around one third who preferred the veterinarian’s response. Trust followed a similar pattern: 88% said they trusted the ChatGPT answers, compared with 80% for the veterinarian-generated responses.

The difference was also visible in the detailed ratings. ChatGPT received an average structure score of 4.52 out of 5, compared with 4.03 for the veterinarian. Its perceived information quality was rated 4.28 versus 4.03.

The authors also found higher scores for empathy, suggesting that participants responded positively not only to the organisation of the information but also to the tone used by the AI.

Communication, rather than clinical expertise

The study does not show that ChatGPT provided better veterinary medicine. Importantly, the clinical accuracy of the answers was not independently assessed. The “quality” score measured participants’ perception of the information rather than whether the advice was medically correct. This distinction is central to interpreting the findings. A response can appear clearer, more reassuring or more complete while still containing errors or inappropriate advice.

The authors themselves argue that much of ChatGPT’s advantage may have resulted from communication style: clear organisation, explicit acknowledgement of the owner’s concerns and reassuring language. They describe these as skills that veterinarians can also be taught rather than advantages unique to AI.

Important limitations

The experiment was deliberately small. Only one veterinarian provided the comparison responses, and the analysis involved just two Reddit-derived scenarios. A different veterinarian, different questions or another prompting strategy could have produced different results. The responses were also subject to word limits, which may have influenced how detailed or polished they appeared. The authors therefore caution against generalising the results to veterinary consultations as a whole.

The study also used GPT-4o as available in May 2024. It does not assess newer models or how AI would perform when faced with more complex cases, incomplete histories, diagnostic uncertainty or misleading information supplied by an owner.

A role as a communication assistant

The most immediate application identified by the researchers is not autonomous diagnosis, but assistance with client communication, written advice and clinical documentation. In a profession where workload is a persistent concern, AI could help veterinary teams draft clearer explanations or structure follow-up information. But the authors stress that professional responsibility must remain with the veterinarian and that AI-generated content requires critical review.

The study therefore highlights an interesting contrast: participants often preferred the AI answer, but the experiment tested how convincing and well communicated the response appeared, not whether AI was better at veterinary diagnosis.

For veterinary practices, that may be the more useful finding. The competitive advantage of generative AI may currently lie less in replacing clinical expertise than in showing how strongly owners value clear, structured and empathetic written communication.