Our scientific foundation
What we know, what we think we know, and what's still open
Bruxism and jaw pain have never had much good research behind them, and that gap fills up with confident advice that contradicts itself. Real evidence does exist, though, and it's what we build on. Here it is, sorted by how sure we actually are — and where we're guessing, we say so.
The gap
Your sleep tracker sees the night. It doesn't see this.
Sleep trackers have made the night legible — stages, heart rate, how long you were restless. None of them look at what your jaw is doing while it happens, which is the one thing you cannot observe for yourself and cannot remember in the morning.
What we know
Well enough to build on
Replicated, large, or both — solid enough to plan around.
Bruxism is a behaviour, not a disorder
The international consensus that defines these terms was updated in 2025, and it is explicit on both counts. Bruxism is masticatory muscle activity during sleep, rhythmic or sustained: grinding, yes, but equally clenching, bracing and thrusting — and the quiet sustained kind is often the more taxing of the two. And it "is not a movement disorder or a sleep disorder". Nor is it unusual. Compare 82 people without a bruxism diagnosis against 33 diagnosed sleep bruxers and close to 60% of the undiagnosed group show the same activity overnight — less of it, at lower amplitude, but the same thing. It runs on a continuum rather than sorting people into those who have it and those who don't. Which doesn't mean you can decide to stop, since night-time activity runs while you're offline. It does mean you're not broken.
Half of it makes no sound at all
In a sleep-lab study of ten people with diagnosed sleep bruxism, scored against synchronised video, only about half of their bruxism episodes produced any grinding sound — 50.4% on average. The rest were silent: clenching, sustained holds, bracing. A partner who hears nothing is not evidence of nothing. The same study is a good picture of how busy a jaw is overnight. Of all the activity recorded on the chewing muscles, about 21% was bruxism, about 33% was other orofacial movement — swallowing, coughing, yawning, chewing, tongue and lip movement — and about 46% was head and body movement showing up on the same muscles. Take the video away and the bruxism count comes out roughly a quarter too high, with the sustained clenches least reliable of all: only 41% of those scored without video turned out to be real, because on a trace a swallow looks a great deal like a clench. It's a pilot study of ten nights and its own conclusion is reassuring — the scoring holds up well enough to be useful. But it maps exactly where the difficulty sits, and separating those activities is the actual work. It's also why this literature rates set-ups that read several muscles and can stage sleep above simple single-lead EMG, and why accuracy figures from one set-up say nothing about another.
Self-management reduces chronic orofacial pain — after months, not weeks
A meta-analysis of 12 randomised trials and 757 participants found self-management and education reduce chronic orofacial pain beyond three months, with a standardised mean difference of −0.32 (95% CI −0.47 to −0.17) and a GRADE rating of high. Below three months the same analysis is a null, so we won't promise you anything in month one. The effect is small by convention, and it comes from therapist-delivered trials — a stronger dose than an app. It is still the best evidence we own, and it is about the programmes rather than the wearable.
Counting episodes tells you nothing about how much someone hurts
The obvious story — fewer episodes, less pain — does not survive testing. In 220 people, the correlation between masticatory muscle activity and reported pain was r ≈ 0.02, with every ROC analysis performing at chance; the authors advise clinicians to stop using activity frequency for this altogether. The dissociation runs both ways in the trial literature: the studies with the biggest drops in muscle activity often showed no pain change, and the study with the best pain result showed no change in activity. Measuring the activity is still worth doing. Reading your pain off it isn't.
What we think we know
Probably true, not settled
Supported, but by small trials, few proper controls, or numbers that swing depending on how you measure.
Cueing during an episode reduces jaw muscle activity while it's running
Around fifteen studies point the same way: a device that notices jaw activity as it happens and answers it with a cue reduces that activity while it's in use. This is the principle our wearable is built on, and it isn't ours — it has been tested for the better part of two decades.
Reading that literature as a whole is harder than it sounds. How the activity is measured varies, and so does whether what's being counted is really bruxism. So does the kind of cue and how fast it arrives. So do the baseline levels people start from, who gets included, and how long the protocol runs. The trials are small — a median around fifteen participants — and only a handful have a genuine sham. Those are our own figures, from the twenty-three trials we hold, not a published tally. A fair share are run or funded by the maker of the device being tested, which is worth knowing and not something we can be sniffy about: when we run ours, we'll be in the same position.
Even the largest trial resists a headline. In 48 people, the number of activity events per hour did not fall significantly during the treatment weeks — in one arm it rose by 18% — but it did fall significantly afterwards, by around a quarter at eight weeks and a third at six months, alongside large sustained drops in pain and in grinding intensity.
Directionally we believe it. We wouldn't call it settled.
Several of those trials report real, lasting symptom relief
The pain results are better than the muscle-activity results. In that same 48-person trial, self-reported pain fell by about 71% and was still down six months later. A separate 74-person study of a different device reported roughly halved pain scores, also sustained. Those are numbers worth taking seriously. Most, though, are before-and-after or self-controlled rather than blinded — and the trial with the best pain result is the one that found no change in muscle activity at all. Whatever is helping people, it does not appear to travel through the episode count.
The full case for biofeedbackBruxism is common — but every prevalence number should be read twice
The best current meta-analysis puts sleep bruxism at about 21% of adults and awake bruxism at about 23%. The same analysis puts the sleep-lab-measured subgroup at 43%, with a confidence interval running from 17% to 75%. Numbers that wide aren't a fact about jaws; they're a fact about measurement. Anyone quoting you a single confident figure — us included — is rounding away a lot of uncertainty.
What we're still working out
Open questions, ours included
Questions the field hasn't answered — and a few that are ours to answer.
Why the things that help, help
The mechanisms usually offered for why any of this works are assumptions that have mostly not been tested directly, and the little work that has looked has not been kind to them. We claim the outcome and hold the explanation loosely.
What we should be measuring instead
If episode counts don't track how someone feels, something else must. The paper that made that case declines to propose a replacement, and so far nobody has. It's one of the more interesting open problems in the field, and we're sitting on a lot of data that bears on it.
Whether any of this survives outside a lab
Every trial in this literature ran under supervision, for a few nights to a few months. None of them had to survive somebody deciding it wasn't worth putting on. Adherence is what actually decides whether a device helps anyone, and it is the variable nobody has measured. We're gathering exactly that data in our own user testing now, and we'll share what it shows.
Whether the change outlasts the device
Symptom improvements hold up at six months in several studies. Whether the underlying behaviour is retrained is a different question, and nobody has shown it yet.
Two trials measured muscle activity after the cue stopped. In one, ten people using a vibrating appliance saw activity fall over four weeks — and climb significantly back within two nights of the cue being switched off. In the other, activity dropped while the cue was on and had returned to baseline by follow-up, while the improvements in muscle tenderness persisted. Both papers read as broadly positive overall, and both were independent of any manufacturer — which matters, because in this literature the studies reporting durability are largely the ones run or funded by the device maker.
People stayed better; the activity didn't stay down. Which of those two facts matters more to your life is exactly what the field hasn't settled, and it's the question we'd most like answered.
Why this field is so thin in the first place
Jaw problems are common, expensive and miserable, and the research base is a fraction of what problems this common usually attract. We don't have a good explanation for that. It's part of why we started, and part of why we publish what we don't know.
Our own work
How far our own evidence goes
Two things you can hold us to today. The algorithm that scores your nights was built and tested against polysomnography-scored recordings — the sleep-lab gold standard — and the wearable's signal quality is verified against reference recordings.
Early results from our own user testing are promising: we're seeing reductions in overnight jaw activity. That's a good start, and we'll publish more as it comes in.
What we have not run is a clinical trial of what this training does to health outcomes — TMD symptoms, headaches, migraine. And by our own argument two sections up, a fall in jaw activity does not license a claim about how anyone feels. So we aren't making one. It's the study we most want to see done.
If this is your field, we'd like to hear from you. Research with us
The data behind the smart sensor
3,000+
hours of real-world jaw-muscle signal, recorded by our own wearable
≈400 nights · our recordings
200,000+*
sleep-bruxism episodes analysed to train and test our algorithms
incl. ~10,000 nights of gold-standard lab data
30,000+
gentle biofeedback cues delivered mid-sleep
≈300 programme nights · counted by the system
* why so manyA large part of this library is curated gold-standard PSG lab data — polysomnography recordings selected precisely because they're episode-dense. Excellent teachers for an algorithm; not a picture of a typical night. Yours gets measured on its own.
Founded in 2022
4+ years developing the wearable & algorithms — still finetuning
Developed following ISO 13485
Patent granted in the Netherlands, pending in other regions
Outside view
Prof. Philippe Wilson
Professor of One Health, York St John University
“The health service has been grappling with these disorders, which are not only complex to diagnose but also lack licensed treatments tackling the root cause. This would have a profound and lasting impact on public health and well-being.”
Backed by & built with
Reviewing us, or want to see the working? Write to us
The people who check our workFrom the Blog
Research & evidence
Deep dives into the science of bruxism and TMD
The JawSense app
When you want to know what your own jaw is doing
The app is free and works from today — check-ins, exercises and routines, all there before you spend anything.
Free · iOS & Android · works without the device