Can AI read pain in your cat's face? What the research shows — and where chatbots fail

tailstory-app.com · 2026-10-03 · by Long Nguyen ·generic
A smartphone model spots pain in cat photos with 95.5% accuracy. Four popular chatbots, tested on the same task, scored pain too low. How the Feline Grimace Scale works, and how to use it at home.

Can AI read pain in your cat's face? What the research shows — and where chatbots fail

A tabby kitten with coral landmark dots on its face next to a score card with five rows of coloured bars

Cats are good at hiding pain, and owners are often the last to know. But a cat in pain does change its face — in five specific, measurable ways. Veterinary researchers turned those changes into a scoring tool, then trained AI to read it from a photo. In 2025 a team tested whether everyday chatbots could do the same job. The answer is a useful lesson for anyone who has ever uploaded a pet photo and asked "is something wrong?"

The five things a cat's face does when it hurts

The Feline Grimace Scale was developed at the Université de Montréal and published in 2019. It scores five "action units" from 0 to 2:

| Action unit | Relaxed (0) | Obvious pain (2) | | ----------- | ----------------------- | ------------------------------------------------------------------- | | Ears | Facing forward | Tips pulled apart and rotated outwards | | Eyes | Open | Narrowed to less than half their width, or squeezed shut | | Muzzle | Round and relaxed | Flattened and stretched into an oval | | Whiskers | Loose and curved | Straight and pushed forward, away from the face | | Head | Above the shoulder line | Below the shoulders, or tilted down with the chin towards the chest |

A score of 1 means the change is moderate or you are not sure. Add the five scores for a total out of 10. In the original study, 4 or more separated cats that needed pain relief from those that did not, with a sensitivity of 90.7% and a specificity of 86.6%.

What AI added: 95.5% on a phone-sized model

In 2023 the same research group trained neural networks on 3,447 photos of cat faces, each marked with 37 landmarks — the corners of the eyes, the base of the ears, the edges of the muzzle. One model finds the landmarks; a second turns their positions into a grimace score. The best combination told painful from pain-free cats with 95.5% accuracy, and the models were chosen to be small and fast enough to run on a smartphone. It is a narrow tool, built for one job, trained on photos that experts had scored by hand.

The 2025 twist: chatbots are not pain scorers

A study published in Scientific Reports in 2025 asked a different question: what if an owner simply uploads a photo to a general chatbot? The researchers gave 50 cat face photos to ChatGPT, Gemini, Claude and Perplexity, asked each to score the Feline Grimace Scale, and compared the answers with a veterinary anaesthesiologist. They ran the test twice, two months apart. The average gap, converted to points out of 10:

| Chatbot (version tested) | Gap, first test | Gap, second test | What it means | | ------------------------ | ------------------ | ------------------ | ------------------------------------------------------------------------- | | ChatGPT (GPT-4o) | 2.2 points too low | 2.1 points too low | Consistent, but consistently underscored pain | | Claude (3.5 Sonnet) | 1.1 points too low | 0.5 points too low | Closest on average, but single scores scattered across the pain threshold | | Gemini (1.5 Pro) | 2.2 points too low | 2.5 points too low | Too low and too scattered | | Perplexity | 2.1 points too low | 1.8 points too low | Too low; scattered across the threshold on retest |

Every chatbot scored the cats as less painful than the expert did. The authors concluded that most showed poor agreement and could have led to painful cats going without pain relief, or pain-free cats receiving it. They also note the obvious caveat: chatbots change with every update, so this is a snapshot, not a final verdict. The deeper point holds anyway. A general model that sounds sure is not the same as a tool validated for one task — the reason we keep a model away from triage decisions in TailStory.

How to score your own cat at home

You do not need AI for this. In a 2023 survey, 1,262 cat owners and caregivers from 66 countries scored cat photos with the Feline Grimace Scale, and their total scores were reliable regardless of who was scoring. Whiskers and head position were the hardest to judge, so lean on ears, eyes and muzzle when you are unsure.

1. Pick a calm moment. Do not score a cat that is asleep, eating, grooming or meowing — wait until it is awake and still. 2. Watch for 30 seconds without touching or calling the cat. 3. Score each of the five areas 0, 1 or 2 using the table above. If unsure, give a 1. 4. Add them up. A total of 4 or more out of 10 suggests pain that a vet should know about. 5. Take a photo of the face, front-on, at the cat's eye level, and note the time. 6. Score again later — after a vet visit, a new medicine or a bad night — so you can compare like with like.

What a face score cannot tell you

  • It is built for acute pain — after surgery, an injury or a sudden illness. It is not reliable for long-term pain such as arthritis, where changes in jumping, grooming and litter-tray use say more.
  • It is for cats only. There is no validated grimace scale for dogs yet; vets use behaviour-based pain scales for them instead.
  • It is less accurate in the first 30 minutes after sedation or anaesthesia, when a groggy face can look like a painful one.
  • A low score does not rule pain out. If your cat is hiding, has stopped eating, is straining in the litter tray or breathing fast, call the vet whatever the score says.

The most exciting part of this research is not the AI. It is that a cat's pain, once invisible, now has a number that owners, vets and software can all agree on. Use it, write it down, and let the vet make the call.

ShareWhatsAppXLinkedIn Copy link

This article is general information for pet owners, not veterinary advice. If your animal is unwell, contact a veterinarian.

Sources

1. Evangelista et al.: Facial expressions of pain in cats — the development and validation of a Feline Grimace Scale (2019) — Scientific Reports 2. Steagall et al.: Fully automated deep learning models with smartphone applicability for prediction of pain using the Feline Grimace Scale (2023) — Scientific Reports 3. Ngai et al.: Agreement of Feline Grimace Scale scores between chatbots and an expert rater (2025) — Scientific Reports 4. Monteiro, Lee & Steagall: Can cat caregivers reliably assess acute pain in cats using the Feline Grimace Scale? A large bilingual global survey (2023) — Journal of Feline Medicine and Surgery 5. Mota-Rojas et al.: How facial expressions reveal acute pain in domestic animals with facial pain scales as a diagnostic tool (2025) — Frontiers in Veterinary Science 6. Feline Grimace Scale — acute pain assessment in cats — Université de Montréal 7. How to assess pain in cats using the Feline Grimace Scale — Improve Veterinary Education

Questions this article answers

Can AI tell if my cat is in pain from a photo?

A purpose-built model trained on 3,447 expert-scored cat photos reached 95.5% accuracy at telling painful from pain-free cats. General chatbots are a different matter: in a 2025 study, ChatGPT, Gemini, Claude and Perplexity all scored pain lower than a veterinary expert, and most showed poor agreement.

What is the Feline Grimace Scale?

A validated tool that scores five facial features — ears, eyes, muzzle, whiskers and head position — from 0 to 2 each. A total of 4 or more out of 10 suggests the cat may need pain relief. It is designed for acute pain, not long-term conditions such as arthritis.

Can cat owners use the Feline Grimace Scale?

Yes. A 2023 survey of 1,262 caregivers from 66 countries found their total scores were reliable. Score a calm, awake cat after watching it for 30 seconds, and contact a vet if the total is 4 or more, or if other signs of illness are present.

Keep reading

HealthAutumn hazards for dogs and cats: acorns, conkers, mushrooms, antifreeze and rat baitAutumn walks bring a handful of seasonal poisons. What acorns, conkers, wild mushrooms, antifreeze and rodent bait do to dogs and cats, which signs to watch for, and when it is a same-hour call to the vet.25 September 2026·7 min readHealthHow to count your dog's or cat's sleeping breathing rate — and when the number mattersThe sleeping breathing rate is the one heart and lung check you can do at home with a phone timer. How to count it, what is normal for dogs and cats, and which numbers mean a call to the vet.25 September 2026·6 min readHealthCat vomiting: when it's just a hairball and when to call the vetCats vomit more than most pets, and owners get used to it. How to tell a hairball from vomiting that needs a vet, how often is too often, which signs cannot wait, and what to note before the appointment.25 September 2026·7 min read

→ Back to the blog

One photo. The whole record.

Fifteen days free with one pet — the full record, logging, health score, weight, timeline, the EU passport and TailStory Doctor. After that it is TailStory Plus.

Download on theApp StoreGet it onGoogle PlayComing soonScan to install on iPhone

No forms, no card, and your data stays in Europe.

tailstory-app.com · 8 images · 34 links

Comments (0)

No comments yet