Tailo reads your dog's behaviour and describes what it sees. Everything about how it does that comes from published research on how dogs show stress, fear and ease. We read more than 200 studies to build it, and the ones that do the heavy lifting are listed on this page in full.
A short video is a snapshot, not a diagnosis. Tailo is a way to notice things and keep a record, not a replacement for your vet or a qualified behaviourist.
There is no one signal that reliably means "stressed". When researchers measured it directly, flattened ears were present in most worried dogs but also in plenty of untroubled ones, and a nose lick pointed the other way. So we never hang a reading on one behaviour. We describe what the dog did and read it as part of the whole picture.
Sources: Bremhorst et al. 2022; Riggio et al. 2026
Across every stressor one team tested, the response that showed up every time was a dog no longer lying down and resting. A dog settled with a loose body is, at the least, not wound up. A dog that cannot settle, moves away or hides is the clearest thing a short clip can show. So rest is where our reading starts.
Sources: Beerda et al. 1998; Hekman et al. 2012; Part et al. 2014
Lip licking, yawning and a lifted paw track whether a person is present as much as whether anything is wrong. In one study a dog licked its lips more when a person who had just frightened it invited it over than when it was shouted at. We treat these as weak on their own, and we say so.
Sources: Firnkes et al. 2017; Riggio et al. 2026
An excited dog and a worried dog can look the same on camera. The science is blunt about this: what you can see reflects how aroused a dog is, not whether that is good or bad. So we tell you how settled your dog looked, and we are careful not to put words in its mouth about how it feels.
Sources: Mendl, Burman & Paul 2010
Asked to rate the same dogs, trained staff could not agree closely, and on a fine scale around half of behaviour experts missed the reference answer. Reading a dog to a number out of 100 claims a precision nobody has. So we read in three plain steps rather than a score.
Sources: Diesel et al. 2008; Gatehouse et al. 2025
Whether a person is present, and whether the dog is alone, change the reading. Whining and going to the door matter most when a dog is on its own. We ask about the setting, and we tell you plainly what a single clip could not let us judge.
Sources: Riggio et al. 2026; Palestrini et al. 2010; Scaglia et al. 2013
We describe, we do not score. You get a plain account of what your dog did, in what setting, and a reading of how settled it looked in three steps. No number out of 100, because the research does not support one.
We weight by the evidence. Each behaviour we note is marked by how well studies support reading it from video, so "did not settle" counts for more than a single lip lick, and signals with no good evidence are flagged as such.
We say what we could not judge. If the ears were out of frame, the tail was docked, or we could not see what happened before filming, we tell you, instead of guessing.
A behaviour change is a vet matter first. If something looks new, sudden, or like it could be pain, we say to see your vet before anything else. We never name a product or medicine as the answer.
Tailo has not been tested against a panel of behaviourists scoring the same clips, so we do not claim it is clinically accurate, and we are building that evaluation now. People read dogs from video unevenly, and so can software. A clip catches one moment, and a dog left alone behaves in cycles that a short video only samples once. None of this makes a reading useless. It means a reading is a prompt to look closer, keep a record, and ask someone qualified if anything has changed.
The studies that most shape how Tailo reads behaviour. Each link opens the paper's record at its publisher.
tailo