It's 6:12 on a Tuesday morning and the app has already made up its mind about your dog.
Resting heart rate overnight, 58. Sleep quality, fair. A small amber dot next to something called respiratory trend. Meanwhile the dog himself is stretched out in a patch of sun on the kitchen floor, living his best life and looking like he hasn't got a care in the world.
Maybe you bought the collar after a diagnosis. Maybe nothing is wrong yet, he's just getting on a bit, and you wanted something keeping watch while you sleep. Whatever brought you here, you're looking at a number and wondering where it came from.
There's no wire and no needle. There's a plastic box sitting on top of a very thick coat.
So what is that box reading?
What's actually inside the collar?
A 2025 review in ACS Sensors describes pet wearables as mostly physical sensors, built into collars and vests, relying on radar and acoustic technologies.
No chemistry. Nothing sampling his blood. Movement, sound and radio waves, and a great deal of software.
Three sensors, and the software that reads them:
- A movement sensor. Usually a six-axis unit: three axes for acceleration, three for rotation. Every collar has this, and it carries the strongest evidence.
- A thermometer, resting against the skin rather than going anywhere unpleasant (anyone who's had to take an animal's temp manually before can relate).
- Something reading his neck. Either a sensor listening for the heart and lungs, or a tiny radar bouncing a signal off his skin.
- An algorithm deciding what all of that adds up to. On most of these devices the software does more work than the hardware.
This isn't everything, but between that review and the sensing patents, it's what turns up in the collars people buy.
Why doesn't it just do what my smartwatch does?
Because the thing your watch does stops working at the hairline.
A smartwatch shines green light into your wrist and reads what bounces back, because blood absorbs light differently as it pulses. It needs bare skin, firm contact, and a wearer who mostly holds still.
Now put that on a cat.
A Turkish veterinary team tested two well-known smartwatches on cats and got good heart-rate agreement against an ECG.
Read the method, though.
They clipped the fur off a patch on each cat's shin and strapped the watch tightly over the bare skin, and the cat was under general anaesthetic throughout.
The same team ran the equivalent test on 15 dogs, watches strapped to the shin, dogs under anaesthetic.
Even then, the blood oxygen reading kept failing. Each watch was given 140 attempts. The better of the two came back with nothing on 94 of those attempts, while the clinical pulse oximeter attached to the cat's tongue returned a value every single time.
So optical sensing wants a shaved patch, tight contact, and a still animal. A collar has none of the three.
And it isn't only cats. One of the companies building dog collars says the same thing in its own validation paper: because of their fur, optical and electrical sensors aren't feasible on dogs. That's the stated reason the product exists.
So how does it find a heartbeat through all that?
By feeling his neck instead of looking through it.
Every heartbeat is a small mechanical shove, and it reaches the surface as a movement far too slight for you to see. Breathing does the same thing, slower and bigger. That signature on the skin of the neck is what the collar is chasing.
Two ways of reading that neck show up in the products. Some collars listen, using acoustic sensors. To get the sound past the coat, little bumps are moulded into the inside of the band, shaped to push down through the fur and hold the collar steady when a dog swings his head. Others use a miniature radar that reads how the skin moves. Radio passes straight through hair, which is the one property optical sensing lacks.
Here's something you can check on your own dog tonight. A vitals collar doesn't need to be strap-tight, because none of this needs skin contact the way a watch does.
If the instructions said to fit it snugly and leave a finger's width, that isn't the maker hedging. It's how the sensing works.
How much can you trust the number?
Trust the trend. Don't trust the moment.
The cleanest test we have was run by cardiologists at Michigan State, working outside the companies that make these collars.
Twelve healthy dogs wore a monitoring collar and a Holter monitor at once, and the researchers compared the two. Averaged across 24 hours the collar held up well, within about 2 beats per minute of the clinical gold standard.
Ask it what his heart rate was over any two minutes while he was up and moving, and the agreement fell apart.
The manufacturer's own research lands in the same place from the opposite direction, which is what makes it convincing. Every recording in that validation study was taken while the dog was asleep, in sessions averaging five minutes. Sessions were dropped when the dog moved too much. A filter removed 16.6% of what remained before the accuracy figures were worked out. On what was left, the collar tracked an ECG very closely.
A real result. Not the same claim as "99.6% accurate," which is how it tends to reach you.
And in fairness to the technology, that same paper found performance held up across every coat it tested, from thin and short to thick and heavy. Fur is not what defeats these collars. Movement is.
Which makes these instruments of the quiet hours: he's asleep on the rug at 3am and you are not awake to watch him. That's the job.
We go further into the numbers in our piece on collar heart-rate accuracy.
What about the activity and sleep tracking?
The strongest evidence for any of this comes from an external validation at the University of Surrey. It put a collar accelerometer on 51 dogs of 36 breeds and checked its output against video of what each dog was really doing.
Over 95% of records came back correctly classified for walking, trotting, running and eating. Head-shaking too, which matters more than it sounds, because a rise in head-shaking is how an ear problem can reveal itself.
Sleep is where it slips. In almost half the moments the device recorded as sleep, the video showed a dog lying down with his head up, awake and very still.
If your app says he slept right through, some of that may be a dog lying in the dark not sleeping at all.
One more limit: those algorithms were built for dogs of 10 kg and up. Small-dog owners sit further from the evidence than the app suggests.
In fairness the other way, when the researchers slowed the footage down they often found the device right and the human wrong. It caught three paces of walking inside a stretch of trotting that nobody could see at normal speed.
Small things, repeated all day, are where these devices have the edge over monitoring your animal yourself manually.
Our gait analysis guide covers what that tells you about a sore leg.
And the temperature reading?
That one is more of an estimate than a measurement, and the difference matters.
The patents describe a skin temperature sensor, a second sensor reading the room, and the movement sensor. The software works out what his skin ought to read given how warm the room is and how much he's moved, then flags anything outside that range.
Sensible engineering. Also not a thermometer actually taking his temp.
Colorado State fitted 19 hospitalised dogs with a monitoring harness and checked it against manual measurements. Heart rate and breathing rate followed the same trends as the nurses' own numbers, but not closely enough to act on without confirming them.
Temperature was the least reliable of the three, and it got worse the longer the coat, because this time the fur is sitting between a skin thermometer and the skin.
A temperature alert doesn't necessarily mean a fever, but it is a reason to go check your dog yourself.
Does any of this hold true for cats?
The dog studies above run to dozens of animals each, and they sit on a much larger body of work. The feline equivalent is thinner.
The strongest cat validation we could find put accelerometers on 12 healthy domestic shorthairs and reached accuracies of around 0.70 and up. A real finding, and a good deal short of the dog numbers. An earlier and often-cited cat study used three cats.
This doesn't make the collar on your cat useless. It means the numbers rest on a smaller pile of evidence, so take them with a grain of salt.
Cats being the species most determined to hide illness, this is one area that definitely needs more research.
There's also an option the collar aisle won't point you towards. For a cat, the strongest monitoring often isn't worn at all. A litter tray that weighs him every time he uses it doesn't care about fur, doesn't need his cooperation, and can't be shaken off under the bed.
We cover that in monitoring a cat without a collar.
And before anything goes round a cat's neck, there's a safety question that doesn't apply to dogs in the same way. A monitoring collar has to stay on to collect anything, while cat collars are built to come off under pressure for good reasons.
We work through that trade in are collars safe for cats.
So when does one of these earn its place?
Start from what costs nothing, because the free measurement is also the strongest one.
Your best tool is still a count you take by hand. Watching your sleeping dog's ribs for thirty seconds gives you a number with more clinical evidence behind it than anything a collar produces.
Our guide to resting respiratory rate shows you how, and our free vitals tracker keeps the log and draws the trend line.
Unfortunately, you can't do that at 3am or from work. That's the gap a device fills. If your vet wants a number you can't collect reliably by hand, or there's a stretch of the day nobody is watching him, a monitor stops being a gadget and becomes a useful tool.
And if it's mobility rather than his heart that worries you, movement tracking is the part with the deepest evidence behind it, so a simpler activity monitor may serve you better than a vitals collar.
Match the sensor to the question and these devices do well. Ask one for a number it was never built to produce and you still get a number, which is what catches people out.
We work through what that costs in the real cost of pet health tech.
What can the sensors not see?
All a collar can measure is movement, sound and heat. That leaves out a great deal.
Gum colour, which goes grey or white when he isn't moving oxygen properly. A belly gone tight and hard. The difference between breathing fast and breathing with effort, which your eyes catch in a second and no accelerometer can.
These are the signs that can't wait for tomorrow's data, and not one of them reaches a collar.
The full list of signs that mean stop watching and start phoning is in our guide to monitoring a pet with a chronic condition.
If the app is reassuring while your gut is not, believe your gut. These devices notice slow drifts over weeks. You are the one who can see that something is wrong with him this morning.
Is there a sensor or a number in your app you'd like taken apart? Ask us in the comments and we'll get to it.
Part of our guides to how the technology works. If you're monitoring a specific condition, start with health monitoring for a dog or cat with a chronic condition. Wondering which features hold up under testing? See does pet health tech actually work, and for the apps that read a pet's face, AI pain scoring.