Pet owners ask AI first: what a veterinary knowledge benchmark shows
Across 18,827 expert-answered question pairs, how reliable a model is on pet clinical questions depends on the material it reads.
Key numbers
- Benchmark scale: 10,076 text-only plus 8,751 multimodal question-answer pairs drawn from real questions about dogs and cats, paired with answers from expert veterinarians.
- Eighteen models were evaluated under three settings — zero-shot inference, retrieval-augmented generation and supervised fine-tuning — using ROUGE, BERTScore and LLM-as-a-judge metrics for factuality and helpfulness.
- The benchmark annotates question types and clinical conditions, and ships translated test versions in five languages for cross-language comparison.
- The authors' conclusion: current models have uneven strengths on veterinary clinical queries, and clinically reliable systems still need better adaptation methods.
Method and sample
arXiv paper “PetQA: Benchmarking Veterinary Knowledge and Clinical Reasoning” (2026-09-04): a Korean long-form QA benchmark built from real dog and cat questions with expert veterinary answers (10,076 text, 8,751 multimodal pairs), evaluating 18 models across zero-shot, RAG and fine-tuning settings, with five translated language versions.
What it means for brands
Pet owners increasingly ask AI before choosing a clinic. What the model says depends on what it can read: whether hospital credentials, vet specialisms, treatments and pricing are structured and present in the sources it trusts decides how you are described.
How SourceSight handles it
Pet care is a local-services project for SourceSight: prompts come from the questions locals ask (nearby, directions, opening hours, emergency care), the off-site list prioritises maps, reviews and local directories, and the content factory adds credentials and treatment scope.
Limits
The benchmark centres on Korean pet questions with five translated versions and is not representative of Chinese query distributions; it measures model knowledge reliability, not clinic acquisition outcomes.
References
Run it on your own project
What the same prompts return across 10 engines is only knowable by running them.
Related research
Cross-category recommendations: +36% add-to-cart on item pages, +15% on cart pages
Recommending general merchandise to grocery shoppers — a milk frother with the milk — has long been hard; LLM retrieval plus real-time cart context moved production A/B metrics.
Industries · 2026-10-07Home services: 22% of homeowners already use AI to find a provider, and 80% of businesses don't know how to be found
The local services funnel is moving upstream: before a homeowner opens a website or calls, AI has already synthesised reviews, reputation and service descriptions into one answer.