Can an app really measure your sleep without a watch?

Partly — and the honest version of "partly" is more specific than most sleep apps will tell you. A phone on your nightstand can give a reasonable estimate of when you slept and how broken the night was. It cannot tell deep sleep from REM, and the one thing it measures worst happens to be the one people most want to know.

Last updated September 2026 · about 8 minutes

Start with what the gold standard actually is

Sleep stages are defined electrically. Polysomnography reads brain waves, eye movements and muscle tone, and a trained scorer assigns each 30-second epoch to wake, light, deep or REM. That is what "sleep stages" means. It is not a vibe measured from across the room.

Everything consumer-grade is an inference from something easier to measure. A watch or ring combines movement with heart rate and heart-rate variability and predicts what the electrodes would have said. The good ones do this reasonably well. A phone on a nightstand has none of that: no contact with you, no pulse, no temperature. It has an accelerometer feeling the furniture and, if permitted, a microphone hearing the room.

What movement alone can tell you

The technique is called actigraphy, and it is older than smartphones. The standard scoring approach comes from Cole, Kripke, Gruen, Mullaney and Gillin's 1992 paper in Sleep [7]: take movement counts in short windows, weight them against neighbouring windows, and classify each minute as sleep or wake. Decades of clinical actigraphy — and, in a modified form, the motion half of most consumer trackers — descend from that algorithm.

It works because the relationship it exploits is real: sleeping people move much less than awake people. It is genuinely useful. But it has one famous asymmetry, and you should know the number.

That single asymmetry explains most of what is wrong with motion-based sleep tracking. It is a movement detector being used as a sleep detector, so it systematically flatters you: sleep is overestimated, wake is missed, and the quiet half hour you spent staring at the ceiling is quietly scored as sleep. This applies to research-grade wrist actigraphs. A phone on a nightstand is not in a better position.

What happens when you put the phone on the nightstand

Someone has tested this. Natale and colleagues, in Sleep and Biological Rhythms in 2012, had 13 healthy volunteers wear a research actigraph and put a smartphone near the pillow for at least four consecutive nights [9]. The result was mixed in a very informative way.

Agreement between phone and actigraph was reasonably satisfactory for total sleep time, wake after sleep onset and sleep efficiency — but not for sleep onset latency, which the phone tended to underestimate. The authors were explicit that this was a comparison against actigraphy, not polysomnography, and that a proper PSG study was still needed.

Two things follow, and the second is uncomfortable if you sell a sleep app.

• The broad shape of the night — roughly when you slept, roughly how much, roughly how fragmented — survives the move from wrist to nightstand.
• How long you took to fall asleep is exactly the measure that doesn't survive, and it is the measure that matters most if you're testing whether a sound helps you drop off faster. It is also the measure already weakest in actigraphy for the specificity reason above, because "lying still, not yet asleep" is precisely the state motion cannot see.

Why sleep onset is the hard one

Falling asleep looks, from the outside, almost exactly like being still. You have stopped moving, the phone has stopped registering anything, and nothing in the movement trace marks the transition. An algorithm watching motion has to guess where in that quiet stretch sleep began, and its best guess is systematically early — hence the underestimate Natale found.

There are partial answers. Whether the phone gets picked up is a strong signal: someone scrolling in bed is awake, and that is information pure stillness doesn't contain. Sound helps too — breathing changes as you fall asleep, and a microphone analysing the room can tell an occupied quiet bedroom from an empty one, catch snoring, and register that a conversation or a TV kept going. None of this turns an estimate into a measurement. It narrows the window.

What no phone can do

Print your sleep stages. Deep sleep and REM are defined by brain activity; watches and rings infer them from heart rate, heart-rate variability and movement — data a phone on the nightstand does not have in any form. There is no sensor on that table that could support a deep-versus-REM split.

Apps do it anyway. A four-colour hypnogram is the single most screenshot-friendly graphic in the category, and plenty of phone-only apps draw one. If you see one produced by a phone that never touched you, treat it as decoration.

Measure	Phone on nightstand	Watch or ring
Bedtime and wake time	Reasonable estimate	Good
Total sleep time	Reasonable estimate [9]	Good
Wake-ups / fragmentation	Reasonable estimate [9]	Good, still misses quiet wake [8]
Time to fall asleep	Weakest measure; underestimated [9]	Better, not perfect
Snoring, talking, room disturbances	Yes, with the microphone	Usually not
Deep sleep and REM	No	Inferred, reasonably

So is a phone estimate worth having?

Yes, for a narrower purpose than the marketing implies.

An estimate with a consistent bias is still useful for comparison. If your phone underestimates sleep onset by some amount every night, and it does so equally on rain nights and brown-noise nights, then the difference between those groups of nights still carries real information even though the absolute minutes are off. What you must not do is read a single night's "13 minutes" as a fact about your body. It is a reading from an instrument with known bias, and it means something only next to twenty other readings from the same instrument.

That is a real limitation, not a disclaimer. A watch or ring is better, and if you own one you should use it.

How SomniTune handles this

SomniTune takes both paths and is explicit about which one you're on.

If you sleep in an Apple Watch, Oura ring, Whoop band or Withings mat — anything writing to Apple Health — it reads the night from Health and uses your device's numbers, stages included. That is the accurate route.

Without a wearable it falls back to the iPhone estimate: how still the phone lay, whether it was picked up (which separates scrolling in bed from sleeping), and optionally on-device sound analysis that keeps only measurements — a breathing estimate, whether snoring occurred, whether talking or a TV was present. No audio is recorded, saved or uploaded. The settings screen states the limits in the app, not only on this page: steps are only counted when the phone is carried, so leaving it on the nightstand while you get up looks like stillness; under a loud rain or noise bed, quiet breathing is largely drowned out and the breathing figure may be missing; if the sound plays through a separate speaker the app can't cleanly separate its own audio from the room.

The reason the estimate is good enough is what it's being used for: comparing your own nights to each other, same instrument each time, over weeks — not handing you a clinical number. If you want to know whether brown noise beats rain for you, a consistent estimate across forty nights answers that better than an accurate measurement of one.

Sources