What the sensors are actually doing
Fall detection runs on two sensors that are already in every phone and smartwatch: an accelerometer, which measures sudden changes in speed, and a gyroscope, which measures rotation and orientation. There is no camera, no microphone and nothing clever happening in the cloud. The whole thing is a pattern-matching test running on movement data.
The pattern it looks for has three parts in sequence: a brief period of near free fall, a sharp impact spike, then stillness. All three matter. The impact alone is not enough, because a phone dropped on a kitchen floor produces a bigger spike than most human falls. It is the free fall before it, and crucially the lack of normal movement afterwards, that separates a person going down from an object being put down hard.
That last part is why fall detection always pauses before it does anything. Typically it prompts you, vibrates or sounds an alarm, and gives you somewhere between 30 and 60 seconds to cancel. If you move normally or dismiss it, nothing happens. If you stay completely still, it escalates. The pause is not hesitation, it is the most useful signal in the whole system: a person who gets straight back up is usually fine, and a person who does not is usually not.
The falls it detects well
Independent testing and published research broadly agree on the shape of the answer: the harder and more sudden the fall, the better the detection. A trip at walking pace onto a hard floor, a fall down stairs, coming off a bike, a ladder fall, a faint that ends in a straight drop. These produce a textbook signature and are picked up reliably, often within a couple of seconds.
Falls where the person is knocked out or seriously hurt are, helpfully, the ones detected best of all, because the stillness afterwards is genuine and prolonged. The technology is most accurate in precisely the scenario where it matters most, which is not true of every safety feature.
Placement makes a large difference to all of this. A device worn on the body, a watch on the wrist or a device clipped at the waist or chest, moves with the person and sees the fall. A phone in a rucksack, a jacket pocket, a handbag or a van cup holder sees a version of the event filtered through whatever it is sitting in, and may not see it at all. If you take one practical thing from this guide, it is that fall detection is a wearing decision before it is a software decision.
The falls it misses, and why
There is one category that fall detection genuinely struggles with, and any honest supplier will tell you so: the slow descent. Someone who feels faint and slides down a wall, lowers themselves onto a sofa and cannot get back up, or gradually crumples from a standing position never produces the free fall or the impact spike. To the sensors, that looks like sitting down.
Three other cases are unreliable for the same underlying reason. Falls onto soft surfaces, a bed, a thick carpet, deep snow, damp ground, blunt the impact. Falls from a low height, out of a chair or in a shower, may not travel far enough to register. And a fall the person partly arrests with their hands or a bannister is spread out over enough time that the pattern breaks up.
None of that means the feature is not worth having. It means it should not be the only thing standing between someone and a long wait. The sensible design is layered: fall detection handles the sudden events, a no-movement or inactivity timer catches the slow ones, and a scheduled check-in or a simple "still going" prompt catches the situations where nothing dramatic happened but somebody stopped being able to answer. Any one of those alone leaves a gap you can describe in a sentence.
False alarms are the real failure mode
Ask people who have turned fall detection off and almost none of them will say it failed to detect a fall. They will say it kept going off. Dropping a phone onto a tiled floor, a heavy landing in a gym class, a rugby tackle, mountain biking over rough ground, a horse rider posting in the saddle, even slamming a car door with the phone on the seat. Anything that mimics impact-then-stillness can trigger it.
The important thing here is asymmetric. A false alarm you cancel in ten seconds costs you ten seconds. A feature you have disabled costs you everything it was there for. So the goal is not zero false alarms, it is few enough that nobody reaches for the off switch.
In practice that means: set detection sensitivity to match the activity rather than leaving it on one setting for everything, keep the cancel window long enough to actually reach the device (a wrist is quicker than a bag), make sure alerts escalate to a real person rather than only sounding locally, and if you are testing it, test it with a cushion and a deliberate drop rather than by falling over. If a system gives you no way to tune any of this, that is a fair reason to look at another one.
Detection is only half the system
A detected fall that nobody receives is not a rescue, it is a log entry. The question worth asking of any product is not "does it detect falls" but "what happens in the ninety seconds after it does", and the answers vary enormously between products that look identical in the marketing.
Three things decide it. Who the alert reaches, and what happens if that person does not answer: a single contact who might be in a meeting is a weak link, an ordered escalation list with a fallback is not. Whether location travels with the alert, precisely enough to send help to rather than to a general area. And what happens with no signal, because rural lanes, basements, lift shafts, stairwells and large parts of the UK countryside are exactly where being alone and hurt is worst. Store-and-forward, so the alert leaves the moment signal returns, is the minimum. A satellite path is the answer for genuinely off grid ground.
Automatic emergency calling is worth understanding too. Some wearables will dial the emergency services directly after prolonged stillness, which is the right behaviour for a lone individual and occasionally the wrong one, since a false alarm becomes a false 999 call. Systems built around an escalation chain usually put a human in that loop first. Neither approach is wrong, but you should know which one you have bought.
Getting it right in practice
For an older relative living independently, the combination that actually works is a worn device rather than a phone left on a worktop, fall detection on, a daily or twice daily quiet check-in so that a slow decline or a non-fall problem surfaces within hours, and a named list of people who get called in order. Discuss it with them rather than installing it around them: a device that gets left in a drawer because it felt like surveillance protects nobody. Our guide to keeping an elderly parent safe living independently goes through that conversation in more detail.
For solo sport, the same logic with different settings. Sensitivity turned down for the activity, so mountain biking does not trigger it constantly, live location shared with someone who would come and get you, and a plan for the stretch of the route with no coverage. There is more in our guide to staying safe running or cycling alone.
For anyone working alone, the honest summary is that fall detection is one control and a risk assessment needs several. Man down detection, check-in intervals sized to the actual risk and an escalation route that works at 3am are the set, not the pick-one-from. What a man down alarm is and who needs one covers the employer side.
How Vygard handles it
Vygard runs fall detection on iPhone, Android, Apple Watch and Wear OS, with a cancel window so a heavy landing does not become a false alarm, and sensitivity you can match to the activity rather than one setting for every situation. Wearing it on a watch rather than carrying it in a bag makes it noticeably more reliable, and we would rather say that plainly than sell a phone in a pocket as equivalent.
Because detection alone leaves the slow fall uncovered, it sits alongside no-movement detection, scheduled check-ins and one-tap SOS, with Howie, our AI watcher, raising the alarm when someone does not arrive, leaves a safe area or simply goes quiet. Alerts follow an ordered escalation chain with a fallback rather than a single contact, they carry location with them, and they queue and send when signal returns instead of being lost in a blackspot.
If you want to see how it behaves rather than read about it, everything is at vygard.com with a free trial and no card required: Vygard Sports for solo activity, Vygard Family for a household looking out for each other, and the employer packs for organisations whose people work alone.
Frequently asked questions
- Does fall detection work if my phone is in a bag or pocket?
- Less reliably than on the body. A phone in a trouser pocket usually detects a hard fall, but a phone in a rucksack, handbag or jacket pocket is cushioned and loosely coupled to your movement, which blurs exactly the signature the software is looking for. A watch on the wrist is the most reliable option because it moves with you and it is also the quickest thing to reach when you need to cancel a false alarm.
- Can fall detection spot a slow fall or a faint?
- Usually not, and this is the honest limitation of the technology. A slow slide down a wall or crumpling from standing produces no free fall and no impact spike, so it reads as sitting down. That gap is what no-movement detection and scheduled check-ins exist to cover: if someone stops moving for an unusual length of time, or misses a check-in, the alarm is raised without a fall ever having been detected.
- Will it automatically call 999?
- It depends on the product, so check before you rely on it. Some wearables dial the emergency services directly after prolonged stillness. Others, including most systems built for families and employers, escalate to a named list of people first, so a human can assess it and cancel a false alarm before an ambulance is dispatched. Both are reasonable designs, but you should know which one you have and who is expected to answer.
- How much battery does fall detection use?
- Less than people expect. The accelerometer and gyroscope are low-power sensors that most phones and watches already run continuously for step counting and screen rotation. The meaningful battery cost in any safety app is location: continuous high-accuracy GPS is what drains a device, which is why sensible apps sample location adaptively and only go to full accuracy when an alert is raised or a session is active.
Related articles
Last updated 2026-09-30