A white noise machine is one of the cheapest sleep interventions you can buy, which is part of why it is everywhere. It also has one of the strangest evidence bases in sleep science: dozens of studies, mostly small, mostly inconsistent, and a systematic review that concluded the quality of evidence was very low.
That does not mean it does not work. It means we are less sure than the marketing implies, and the situations where it clearly helps are narrower than "everyone should have one".
TL;DR: White noise works by masking, not by sedating. It raises the background sound level so sudden noises are less likely to register and wake you, which is why it helps most in noisy rooms and least in quiet ones. A 2021 systematic review of 38 studies rated the evidence for continuous noise improving sleep as very low quality, with findings ranging from benefit to harm. A 2025 meta-analysis of 12 randomised trials and 1,301 participants found meaningful improvements in subjective sleep quality scores in adults and older adults, larger gains in infants' total sleep time, and no clear effect on wake after sleep onset. Volume is the part people get wrong: WHO recommends under 30 dB in bedrooms at night, and infant sleep machines tested at maximum volume exceeded 50 dB at 30 centimetres, with three above 85 dB.
YouTube: Related video
What white noise actually is
"White noise" means sound with roughly equal energy across the audible frequency range, the way white light contains all visible wavelengths. It sounds like a hiss or a steady shhh, without pitch or pattern.
Most machines and apps give you something slightly different:
- White noise: equal energy per frequency, bright and hissy
- Pink noise: more energy at low frequencies, sounds softer, closer to rain or wind
- Brown noise: even more low-frequency emphasis, sounds like a deep rumble or a waterfall
- Fan and air conditioner sounds: similar masking profile to brown noise, minus the electronics
For sleep purposes, the subtype matters less than people think. The mechanism is the same for all of them, and the studies rarely separate the types cleanly. What matters far more is volume and whether you actually have a noise problem to solve.
The masking mechanism, and why it changes everything
Your auditory system did not evolve to ignore sound. It evolved to notice changes in sound, because a snap in the dark used to be information. That is why the thing that wakes you at 2 a.m. is not the steady hum of traffic, it is a door closing two floors up.
White noise works by raising the constant background level. When the noise floor is higher, a sudden sound has to be louder to stand out against it. The technical term is masking, and it is why the intervention is most useful in a room with unpredictable noise: a partner who gets up at 5 a.m., neighbours, a street, a hotel corridor.
Notice the logical consequence. If your bedroom is already quiet and your sleep problem is not noise, white noise has nothing to mask. It becomes a sound you are adding for its own sake, and the evidence for that is much thinner.
This is also why results in the literature swing so wildly. Studies run in intensive care units, where monitors beep all night, tend to show benefit. Studies run in quiet sleep labs often show nothing, or show harm, because the added sound is now the disruption.
What the evidence actually says
Two papers frame the current state of play, and they disagree in an instructive way.
The systematic review. Riedy and colleagues screened the literature in 2021 and included 38 articles on continuous white or broadband noise and sleep. Their conclusion was blunt: following GRADE criteria, the quality of evidence for continuous noise improving sleep was very low, which "contradicts its widespread use". They found substantial heterogeneity in the noise used, the measurement method, adherence, and what the control group did. The extremes across studies were that continuous noise improves sleep or disrupts it.
They also flagged two risks that rarely make it into product pages: potential negative effects on sleep and on hearing.
The meta-analysis. A 2025 meta-analysis in Sleep Medicine pooled 12 randomised controlled trials with 1,301 participants across infants, adults, older adults, and hospital settings. This one is more positive, and the subgroup detail is what makes it useful:
| Group | Outcome | Result |
|---|---|---|
| Adults 18-64 | PSQI sleep quality score | -3.70 points (95% CI -4.90 to -2.50) |
| Adults 65+ | PSQI sleep quality score | -2.71 points (95% CI -4.98 to -0.44) |
| ICU patients | PSQI sleep quality score | -4.04 points (95% CI -6.35 to -1.73) |
| Non-ICU patients | PSQI sleep quality score | -2.61 points (95% CI -3.84 to -1.38) |
| Infants 0-3 | 24-hour total sleep time | +137.5 minutes (95% CI 67.8 to 207.2) |
| Infants 0-3 | Nocturnal 12-hour total sleep time | No significant effect |
| Infants 0-3 | 12-hour sleep efficiency | +6.6% (95% CI 1.7 to 11.5) |
| Infants 0-3 | Nightly awakenings | -1.83 (95% CI -3.12 to -0.54) |
| Infants 0-3 | Wake after sleep onset | No significant effect |
A few things are worth pulling out.
First, a PSQI drop of 3.7 points in adults is not trivial. The Pittsburgh Sleep Quality Index runs from 0 to 21, with higher meaning worse, and a shift of that size is in the range of what behavioural sleep interventions produce.
Second, most of those PSQI outcomes are subjective. Participants knew whether the machine was on. That does not make the effect fake, because how you feel about your sleep matters, but it does mean we should be cautious about calling it an objective sleep improvement.
Third, the objective infant results are mixed in a specific way. Total 24-hour sleep time went up a lot, but night-time sleep time and wake after sleep onset did not change significantly. If a machine is mostly extending naps and not consolidating nights, the marketing story ("your baby will sleep through") is not what the data shows.
Fourth, and this is the big one: the authors themselves note substantial heterogeneity and limited methodological quality across the included trials. The direction of the pooled effect survived a leave-one-out analysis, which is reassuring, but this is not a body of evidence you should treat as settled.
The volume problem
If there is one thing to take from all of this, it is that the dose is the risk.
The World Health Organization recommends under 30 dB(A) in bedrooms at night for good sleep, and its night noise guidance identifies 40 dB Lnight outside as the level at which the first adverse effects on sleep appear in populations. A quiet bedroom is around 30 dB. A whisper is 30 dB. Normal conversation is 60.
Now consider what people actually do. Sound machines are usually placed within a metre of the bed, often closer, and often turned up until the noise "covers" whatever they are trying to mask.
The Pediatrics study on infant sleep machines measured 14 devices at maximum volume at 30, 100, and 200 centimetres. At 30 centimetres, every single device exceeded 50 dBA, which is the recommended noise limit for infants in hospital nurseries. Three of them produced output above 85 dBA, which exceeds occupational exposure limits for adults if played for more than eight hours.
The authors were clear about the implication: these devices are capable of producing levels that may damage infant hearing and auditory development. Their recommendations were practical, not alarmist: keep the machine further away, keep the volume lower, and do not run it at maximum all night.
The same logic applies to adults, with a bit more headroom. Put the machine across the room rather than on the nightstand. Set it to the lowest volume that actually masks the sound you are trying to cover. If you can hear it clearly over the shower, it is too loud.
A rough calibration trick: stand where you sleep, and set the volume so the machine is audible but you would have to strain slightly to follow a quiet conversation happening next to it. Then leave it there.
When white noise backfires
There are five situations where I would think twice.
1. Your room is already quiet. If you live somewhere silent and your sleep issue is falling asleep, not staying asleep, masking has nothing to do. You are adding a stimulus and hoping it relaxes you. That is a different claim with weaker evidence.
2. You have tinnitus. Steady background sound can genuinely help some people with tinnitus by reducing the contrast between the ringing and silence. It can also make it more noticeable for others, particularly if the noise has a lot of high-frequency content. Test carefully rather than assuming.
3. You need to hear things. A machine that masks the front door also masks a smoke alarm in some setups, and it definitely masks a child crying. If any of those matter in your household, use the lowest effective volume or skip it.
4. It becomes a dependency you cannot travel without. This is not a medical risk, but it is a practical one. If your sleep falls apart in a hotel without your machine, you have traded a small nightly benefit for a real logistical constraint. Earplugs are a more portable version of the same idea.
5. You are using it to mask a problem you could fix. A fan, a drafty door, a neighbour's dog, a partner's snoring: some of these have direct fixes. A white noise machine is the easiest solution, not always the best one.
On that last point, if you share a bed with someone whose snoring is the actual issue, the machine is treating the symptom. Snoring: what it actually means covers when it is worth investigating.
How to test it on yourself in two weeks
The evidence is muddy enough that your own data is more useful than the average effect size. Here is a clean way to run the experiment.
Week 1: baseline, no machine. Sleep as you normally do. Record what you already track, and add two subjective numbers each morning: how long you think it took to fall asleep, and how many times you remember waking.
Week 2: machine on, same settings every night. Same volume, same position, same bedtime. Do not change anything else: no new supplements, no new workout, no later dinners.
Then compare. What you are looking for:
- Sleep onset latency going down consistently, not just on the first night
- Fewer remembered awakenings, especially in the second half of the night
- Sleep efficiency (time asleep divided by time in bed) improving by more than a few percent
- Your own rating of how rested you feel, which is not nothing
What you are not looking for: a big jump on night one. Novelty affects sleep, and so does the expectation that a gadget will work. Give it a week before you decide.
If you use a watch, look at the overnight data rather than the single sleep score. Sleep fragmentation and frequent awakenings explains which of those numbers are worth trusting. A machine that reduces awakenings should show up as fewer brief wakings and a steadier heart rate trace. If your sleep score improves but your awakenings and overnight heart rate do not move, you are probably seeing expectation rather than physiology.
Century AI is useful here because it puts your sleep, HRV, and resting heart rate on one screen and shows the trend rather than the single night. Two weeks of comparison is exactly the kind of window where trends beat daily scores.
If masking is the goal, what else works?
In rough order of effectiveness for most people:
- Fix the source. A door seal, a draft excluder, heavier curtains, a chat with the neighbour. Unglamorous, permanent.
- Earplugs. Cheap, portable, and physically attenuating rather than perceptually masking. Foam plugs give roughly 25 to 30 dB of attenuation when inserted properly, which is more than most machines deliver at a safe volume.
- A fan. Moves air, which most people find pleasant, and produces broadband noise for free. The airflow is arguably the bigger benefit in a warm room.
- A machine at low volume, placed away from the bed. Fine, and the practical choice if you like the sound.
- Pink or brown noise if white noise feels harsh. Same mechanism, less high-frequency energy, often more tolerable at low volume.
What does not have good evidence: apps that claim to entrain your brainwaves with binaural beats, and "sleep frequency" playlists. Phase-locked acoustic stimulation of slow waves is a real research area with small, mixed results in lab settings, but it is not something a phone speaker delivers reliably. If you want the details on where that research actually stands, the sleep stages guide covers what is measurable and what is noise.
The short version
- White noise masks sudden sounds by raising the background level. It helps most when there is something to mask.
- The best systematic review rated the evidence very low quality. The 2025 meta-analysis is more positive, especially for subjective sleep quality in adults, but the trials are heterogeneous and small.
- Objective benefits are clearest for infant total sleep time, and unclear for night-time wake after sleep onset in anyone.
- Volume is the real risk. WHO recommends under 30 dB in bedrooms; infant machines at maximum exceeded 50 dB at 30 cm, and three topped 85 dB.
- Place the machine across the room, set the lowest effective volume, and do not run it at maximum all night.
- If your room is quiet or your problem is falling asleep rather than staying asleep, test it rather than assuming it works.
- Run a two-week comparison and judge it on awakenings and sleep efficiency, not on how much you like the sound.
A sound machine is a reasonable tool with a modest evidence base and a real dose-dependent risk. Used to mask an unpredictable environment at a sane volume, it is fine. Used at maximum volume, right next to your head, every night for years, it is doing something quite different from what the box promises.
Century AI helps you understand your body with a daily health score, recovery score, and sleep insights, using the watch you already wear.
