PetSignal.ai does not try to translate what your pet is "saying." We surface signals you can observe — body posture, facial tension, movement and context. Below is exactly how an analysis runs, what it can't do, and the four peer-reviewed and lab-based research traditions that shaped how we describe behavior.
One boundary, stated plainly: PetSignal.ai currently analyzes visible cues in photos and short video. The vocal-prosody and bioacoustics work below is scientific background that informs how we describe behavior — audio is not a current product capability.
How an analysis actually works
No separate engines and no model trained on your pet — one AI read, checked against a signal reference.
One AI vision read
Your photo — or, for a video, up to 8 key frames taken about every 2 seconds — goes to a general-purpose AI vision model in a single analysis. If the main AI channel is unavailable, a backup model takes over. We did not train the model ourselves.
Checked against 20 documented signals
The model is given a reference of 20 body-language signals — 10 for dogs and 10 for cats, including two relaxed ones, the play bow and the slow blink — each with a risk level. We compiled it from publicly available behavior research, and the model is told to name only signals it actually sees.
Uncertainty is part of the output
Confidence is set by how clear the image is and how much of the animal is visible, and the full report lists what could not be assessed. A blurry or partial photo should read as less certain.
Clear boundaries
The model is instructed never to present itself as a veterinarian, never to name a medical condition or recommend medication, and to say plainly when an in-person vet exam is warranted.
What it does not do
We publish no accuracy percentage, and the reads have not been independently validated.
Each analysis focuses on one pet. Audio is not analyzed.
The notes you type on the analysis form are not part of that first read; they are only used for the extra guidance in an unlocked report.
A single photo usually takes about a minute; several photos or a video take longer.
Tradition 1 · Body Signals
Purdue Body Signal Framework
Purdue University · Canine Welfare Science
Dogs communicate through their whole body — ears, eyes, mouth, tail, posture, muscle tension, and weight distribution. Purdue's Canine Welfare Science framework breaks this down into observable parts so we can describe what a dog is doing, not what we guess it 'means.'
Shapes the body regions every report describes: ears, eyes, mouth, tail, posture, and muscle tension
Our signal reference applies the same approach to cats, with cues such as whisker position, the slow blink, and the puffed tail
Informs the risk level every report gives: low, moderate, or high risk to people
Alexandra Horowitz · Barnard College Dog Cognition Lab · Inside of a Dog
Dogs do not experience the world the way humans do. Their primary channel is smell, not vision. Horowitz's umwelt framework reminds us to ask 'what is this animal sensing from its own perspective' before drawing any conclusion — and to reject anthropomorphic shortcuts like 'the guilty look.'
Shapes how we write our guides: umwelt first, and no 'guilty look' readings
Reports are structured as observed signals and possible causes, not 'your dog says…' translations
Lund University · MEOWSIC (Melody in Human–Cat Communication)
Cat vocalizations are not a phrasebook. The MEOWSIC project at Lund University studies the prosody of cat-human communication: pitch, melody contour, duration, intensity, rhythm, and voice type. PetSignal does not analyze audio — this research is background for our cat guides and for how carefully we talk about what cats 'mean.'
Studies prosodic features of cat vocalizations: F0 pitch, melody contour, duration, intensity, rhythm, and voice type
Treats meows, trills, purrs, hisses, and growls as context-dependent rather than fixed translations — the stance our cat guides take
Earth Species Project · NatureLM-audio · multimodal animal communication research
Earth Species Project's research is a reminder of how careful AI claims about animal communication need to be: probabilistic outputs instead of confident translations, and attention to the individual animal. PetSignal only reads visible body language, but we borrow that discipline.
Body cues are read together — ears, eyes, mouth, tail, and posture — rather than from one signal in isolation
The model is asked to report confidence and what it could not assess, rather than false certainty
Principles we hold ourselves to — written into the instructions the AI follows where we can, and into how we write our guides.
We surface signals, not translations
Reports are built around what is observable and plausible causes, hedged — not 'your dog is saying I am hungry.'
We refuse the 'guilty look' myth
Lowered head, averted gaze, pinned-back ears are stress and appeasement signals — not moral guilt. Decades of cognition research back this up.
We always offer an out to a professional
When signals suggest pain, fear escalation, or risk to humans, reports point you to a licensed veterinarian or a qualified behavior professional. Software never replaces a clinical exam.
We show our uncertainty
Confidence depends on image clarity and how much of the animal is visible. We don't publish accuracy percentages, and a single photo can only show so much.