Attention and habit
Devices, light and sleep
The strongest evidence points away from light and towards timing: screens used in bed cost measurable sleep, while pre-bed screen use mostly does not.
The blue-light story is weaker than its reputation and the timing story is stronger. A Cochrane review of blue-light filtering lenses found no demonstrated benefit, and a polysomnography study of evening smartphone reading found no significant effect on how long people took to fall asleep, whether the screen was filtered or not. Meanwhile the one study that measured bedtime screen use objectively, with cameras and accelerometry rather than questionnaires, found that screen use in the two hours before bed had no association with most sleep outcomes while screen use in bed did, at a cost of several minutes of sleep for every ten minutes of interactive use. The mechanism that matters looks like arousal and displaced time, not wavelength. For a workplace-oriented comparison point on how digital activity is translated into metrics, see this overview.
The famous melatonin numbers come from an extreme protocol
The study behind almost every headline about phones and the body clock was a well-controlled inpatient experiment, and its conditions were nothing like an ordinary evening.
Four hours of evening reading on a light-emitting e-reader suppressed evening melatonin by 55.1%, delayed circadian phase by more than 1.5 hours, lengthened sleep onset latency from 15.8 to 25.7 minutes and cut REM sleep by about 12 minutes.
Direction: Decrease. Strength of evidence: Mixed.
Caveat Twelve participants, four hours at maximum screen brightness in a room lit to about 3 lux, against a print condition read at 0.91 lux — a near-maximal contrast by design and far more extreme than typical evening phone use.
The result is internally strong and externally narrow. Quoting the 55% and the hour and a half as descriptions of what a phone does to a person at bedtime misrepresents a good study.
Filtering the light does not reliably help
The commercial response to that study was blue-light filtering eyewear, and it has now been reviewed systematically.
Blue-light filtering spectacle lenses probably make no difference to eye strain, effects on sleep-related outcomes are unclear, and the authors state the findings do not support prescribing them to the general population.
Direction: No detectable effect. Strength of evidence: Strong.
Caveat Trials were small and short — under one day to five weeks — so long-term effects remain untested, and the review covers spectacle lenses rather than device-level night modes.
An earlier meta-analysis of blue-blocking glasses shows why the marketing claims survived: what people report and what instruments record part company.
Objective effects were small to medium and imprecise — sleep efficiency g = 0.31 (CI −0.05 to 0.66) and total sleep time g = 0.32 (CI 0.01 to 0.63) — while self-reported sleep quality showed a much larger effect at g = −1.25.
Direction: Cuts both ways. Strength of evidence: Mixed.
Caveat A large subjective effect alongside a marginal objective one is the classic signature of unblinded expectancy, and benefits concentrated in clinical groups such as insomnia, bipolar disorder and delayed sleep phase rather than healthy sleepers.
A lab test of ordinary evening phone use found little
When the exposure is set at something closer to real behaviour — an hour and a half of reading on a phone — the effect on falling asleep disappears.
Ninety minutes of evening smartphone reading did not significantly affect sleep onset latency in any condition — unfiltered screen, blue-light-filtered screen or printed book — nor sleep-dependent memory consolidation, and adolescents' melatonin recovered within about 50 minutes of exposure ending.
Direction: No detectable effect. Strength of evidence: Strong.
Caveat Male participants only, single-night lab conditions, and it did detect reduced deep (N3) sleep in the first two hours for adults using unfiltered screens, so it is a partial rather than a total null.
The children's evidence is nearly all cross-sectional
The most-cited source on devices and children's sleep is a meta-analysis of surveys, and it is rated low certainty by its own authors.
Bedtime device use was associated with inadequate sleep quantity (OR 2.17), poor sleep quality (OR 1.46) and excessive daytime sleepiness (OR 2.72), while mere access to a device without reported use carried almost as much risk (OR 1.79, 1.53 and 2.27).
Direction: Increase. Strength of evidence: Mixed.
Caveat Every included study was cross-sectional with self-reported exposure and outcome, GRADE certainty was rated low, heterogeneity was substantial and causality cannot be inferred.
The detail that should slow anyone down is the second half: simply owning a device, with no reported use, predicted nearly as much risk as using one. That pattern is far more consistent with confounding — household income, bedroom arrangements, parental routines — than with light or content. Anyone setting rules at home should read it alongside what the evidence says about children, parents and devices.
Objective measurement moves the blame from evening to bed
One study replaced questionnaires with wearable and stationary cameras plus wrist accelerometry, and it changes the picture.
Screen use in the two hours before bed had no association with most sleep outcomes, while every 10 minutes of interactive screen use in bed cost about 9 minutes of sleep, gaming about 17 minutes and multitasking nights about 35 minutes.
Direction: Decrease. Strength of evidence: Mixed.
Caveat Only 79 youths over four nights each, so the estimates are imprecise; the authors explicitly conclude that recommendations to restrict all screen time before bed seem neither achievable nor appropriate.
A large student survey points the same way on location while undercutting the assumption that social media is the culprit.
Each additional hour of in-bed screen use was associated with 59% higher odds of insomnia symptoms and about 24 fewer minutes of sleep, but the association did not differ between social media and other activities, and exclusive social-media users had the lowest insomnia rates at 27.6% against 37.1% for non-social-media users.
Direction: Increase. Strength of evidence: Mixed.
Caveat Cross-sectional with a 35.1% response rate and entirely self-reported sleep and screen measures, so reverse causation — poor sleepers reaching for phones — is fully compatible with the data.
What the intervention evidence supports
Very little has been trialled. A randomised pilot in 38 college students found that avoiding the phone for 30 minutes before bed over four weeks shortened sleep latency by about 12 minutes and increased sleep duration by about 18 (He et al., PLOS ONE, 2020), but it was unblinded, used self-reported sleep diaries rather than actigraphy and pre-selected people who already wanted to cut down. Sleep is also the one outcome that showed any persistence in the abstinence trials described in what a digital detox actually does. Taken together, the defensible position is that keeping interactive use out of the bed has better support than any rule about light, filters or a fixed number of screen-free hours.
The short version
- Four hours of evening e-reader use at maximum brightness suppressed melatonin by 55.1% and delayed circadian phase by more than 1.5 hours in 12 inpatients (Chang et al., PNAS, 2015), a protocol far more extreme than normal phone use.
- A Cochrane review of 17 trials concluded that blue-light filtering lenses probably make no difference and should not be prescribed to the general population.
- Ninety minutes of evening smartphone reading did not significantly change sleep onset latency in a polysomnography study of 68 young men, filtered or unfiltered (Höhn et al., Brain Communications, 2024).
- Objectively measured, every 10 minutes of interactive screen use in bed cost about 9 minutes of sleep, while pre-bed use showed no association with most sleep outcomes (Brosnan et al., JAMA Pediatrics, 2024).
- The children's evidence cannot establish cause: it is cross-sectional and low certainty, and device ownership without reported use predicted almost as much risk as use itself.