Blue light is the least of it

A considered reading of the research on phones and sleep — why the study that launched the blue-light story explicitly disclaims it, why the effects shrink to almost nothing whenever sleep is measured rather than remembered, and what is left standing once the colour and the content have both been set aside.

---

It is on the nightstand, or under the pillow, or face down on the mattress within arm's reach. The phone is the last thing most people look at before sleep and the first thing they reach for on waking, and no other object in the house holds that position. A lamp does not have a feed.

The explanation for what this does to sleep has been settled in public for about a decade, and it is a story about colour. Screens emit light rich in the short wavelengths near the blue end of the spectrum; that light suppresses melatonin; melatonin accompanies sleep; therefore the screen keeps you awake by shining the wrong colour into your eyes. The story arrived with a remedy attached almost immediately. Phones acquired a warm-tinted evening mode. An industry grew around amber lenses. The advice hardened into a rule repeated in nearly every article on the subject: no screens for an hour before bed, because of the blue light.

The photobiology underneath that story is real, elegant, and settled. Almost everything built on top of it is not. The single study that put the claim into public circulation says, in its own discussion section, that it cannot support the claim. The commercial remedies derived from it have been tested under blinding and have not been shown to work. And in 2024 the body that issues sleep guidance to the American public convened sixteen experts to vote on whether light from evening screens harms sleep, and they could not reach consensus — for children, for adolescents, or for adults.

What follows separates the pathways by which a phone can cost you sleep, and grades each by how much weight the evidence behind it can bear. There are three worth taking seriously. Two of them are far weaker than their reputations. The third is the dullest thing in this essay and the best supported.

Three pathways, not one

A phone at bedtime can plausibly damage sleep in three separable ways.

The first is photic: light from the screen reaches the eye, signals daytime to the circadian system, suppresses melatonin, and shifts the body clock later.

The second is arousal: the content on the screen — the argument in the group chat, the work email, the video that resolves into another video — engages the mind in a way incompatible with falling asleep, independently of any light.

The third is displacement: the phone consumes the hours in which sleep would otherwise have happened. Nothing physiological is required. The time is spent, and it is spent at the end of the day, the end that has no fixed floor.

These are usually collapsed into a single claim about screens. They should not be, because they imply completely different remedies. If the problem is photic, a filter helps. If the problem is arousal, the filter is irrelevant and what matters is what you were doing. If the problem is displacement, both are irrelevant and what matters is the clock.

One thing to hold on to first

Before any of the three, a pattern that runs underneath the whole literature and decides how much of it to believe. Studies that ask people about their sleep tend to find effects. Studies that measure sleep mostly do not. It is worth installing this now, because almost every finding that follows sorts along that line.

The clearest single demonstration comes from a daily-diary study of young adults with poor sleep, wearing a Fitbit. Bedtime device use on a given night predicted about twenty minutes less self-reported sleep that night — and showed no association with any variable measured by the device on the same wrist, on the same night (Chkhaidze, Millar, Revenson, & Mindlis, 2025). Same people, same nights, two instruments, two different answers.

The imbalance is structural. When the National Sleep Foundation assembled the evidence in 2024 it screened 2,209 articles and found that only 35 experimental or intervention studies existed in the entire literature since 2007. Of those 35, thirty — 86 per cent — used self-reported sleep; thirteen used polysomnography, the full electrode-based laboratory standard, and four used actigraphy, a wrist-worn motion sensor that infers sleep from stillness (Hartstein et al., 2024). Those categories overlap, since a study can carry both.

Two things follow, and the second matters more. The field that produced a decade of confident public advice rests on about three dozen experiments. And roughly half of them did measure sleep properly — which is precisely why the pattern is worth noticing rather than explaining away. This is not a literature that lacked good instruments. It is a literature where the studies using the good instruments kept coming back with less than the studies that asked.

The objective instruments have their own difficulty, and it cuts in an awkward direction. When five consumer and research sleep trackers were validated against polysomnography — the full electrode-based laboratory standard — in participants deliberately using phones around the sleep period, all of them misclassified motionless wakefulness as sleep to some degree (Willoughby et al., 2024). A person lying still in the dark, scrolling, can be scored as asleep. That undercuts the null results and the positive ones alike, and it means the objective studies in this field are less objective than they look.

None of which decides anything on its own. It simply sets the exchange rate: a self-reported effect is worth less than a measured one, and a measured null is worth less than it appears.

What light actually does

Start with the mechanism, because it is the part that survives.

The eye does not use light only to see. Alongside the rods and cones that produce vision, the retina contains a small population of intrinsically photosensitive retinal ganglion cells, which carry a pigment called melanopsin and which report ambient light levels to the brain's master clock rather than contributing to the image. Two independent groups mapped their sensitivity in 2001 and converged on the same answer: the action spectrum for melatonin suppression peaks in the short-wavelength region around 446 to 477 nanometres (Brainard et al., 2001; Thapan, Arendt, & Skene, 2001). That signal reaches the suprachiasmatic nucleus of the hypothalamus, which governs circadian timing, and one downstream effect is suppression of melatonin, the hormone whose evening rise accompanies the body's preparation for sleep.

This is why the blue-light story has such purchase. There is a real photoreceptor, tuned to a real part of the spectrum, wired to the real clock. Nothing in what follows disputes any of it.

It is worth noting how those spectra were obtained, though, because it matters for what they can be asked to prove. Both studies used monochromatic light, delivered in the middle of the night, through pharmacologically dilated pupils. That is the configuration designed to maximise sensitivity and isolate the receptor. It is not an evening with a phone.

The study everyone cites

The claim reached the public through one paper. In 2015, Anne-Marie Chang and colleagues at Brigham and Women's Hospital published a controlled comparison in PNAS: participants spent evenings reading either from a light-emitting e-reader or from a printed book. Reading from the device suppressed melatonin, delayed circadian timing by about an hour and a half, lengthened the time taken to fall asleep, and reduced alertness the following morning (Chang, Aeschbach, Duffy, & Czeisler, 2015).

That is a real result from a serious laboratory, and the circadian finding in particular is large and important. But the study has been asked to carry a claim it does not make, and four details explain why.

The first is the sample: twelve people, eleven of them analysable for melatonin.

The second is the dose. Participants read for roughly four hours an evening, on an iPad at maximum brightness, for five consecutive evenings, inside a fourteen-day inpatient protocol. The circadian delay of an hour and a half is the product of that regimen. It is not what one evening with a phone produces.

The third is the comparison condition. The printed book was lit at 0.91 lux, in a room of about 3 lux. That is not a person reading in a normally lit bedroom; it is close to darkness. The study contrasted a screen against near-blackness, and the reported difference is the difference between those two states.

The fourth is the one that should have ended the popular story before it began. The authors write that the study did not include a light-emitting device of longer wavelength for comparison, and that their findings may therefore be due to the difference in irradiance level rather than spectral composition. In plain terms: they could not tell whether the effect came from the light being blue or simply from the light being brighter. The paper universally cited as proving that blue light disrupts sleep states in its own discussion that it cannot distinguish blue from bright.

There is a fifth detail, and fairness requires reporting it too: the device was held at a fixed 30 to 45 centimetres, further than many people would choose, which the authors note reduced retinal exposure in their own protocol — and they add that users of smaller devices may hold them closer still. That cuts against the study's dose in one direction and against the phone in the other.

A companion study made the picture more realistic and introduced a complication. When Evan Chinoy, Jeanne Duffy and Charles Czeisler let participants choose their own bedtimes rather than fixing them by protocol, tablet use again suppressed melatonin and delayed its onset — and participants also went to bed about half an hour later than on paper evenings, 22:03 against 21:32 (Chinoy, Duffy, & Czeisler, 2018). That study had nine participants and the same near-dark comparison, so the numbers should be held loosely. But notice what happened the moment the experimenters stopped controlling bedtime. A second pathway walked into a light experiment. The device did not only change physiology. It changed what time people stopped.

The dose problem, which runs the wrong way

The obvious sceptical move here is to say that laboratory doses are larger than real ones, and that a phone is too small a light source to matter. That move is wrong, and the study showing it is wrong is the most interesting work in this section.

In 2019, Andrew Phillips and colleagues at Monash University mapped individual dose–response curves for light-induced melatonin suppression. Fifty-five participants were exposed in the evening to a dim control and to levels between 10 and 2,000 lux. The dose producing half-maximal suppression across the group was about 25 lux (Phillips et al., 2019). Twenty-five lux is not a laboratory extreme. It is a dim room. Exposure at 10, 30 and 50 lux delayed apparent melatonin onset by 22, 77 and 109 minutes respectively. Those were five-hour exposures, beginning four hours before habitual bedtime — and the same scepticism applied to Chang's four hours applies here. This is not the dose from glancing at a phone. It is the dose from an evening spent in a lit room, which is the point.

So the circadian system is not insensitive to ordinary domestic light. It is acutely sensitive to it, at levels below those of a normally lit living room. An international expert consensus recommends keeping evening exposure below 10 melanopic lux in the three hours before bed, and the sleeping environment below 1 lux (Brown et al., 2022) — thresholds most homes exceed by a wide margin with lights nobody has thought to question.

A note on that unit, because it matters twice more before the end. Melanopic lux is not the same measure as the ordinary lux figures above. It weights light by how strongly it drives the melanopsin system rather than by how bright it looks, and the two can diverge sharply: a warm, red-shifted lamp can be visually bright and melanopically dim. That divergence is the entire premise of the filtering products. For a white-ish screen, though, the two run close together — Chang's iPad measured 31.7 photopic lux against 31.0 melanopic, near enough to identical.

This does not vindicate the blue-light advice. It undermines it. Chang's iPad delivered about 31 lux. If a dim room is already in the effective range, then the phone is one contributor to a total ambient dose that also includes the ceiling fixture, the lamp, the television across the room, and the bathroom light above the mirror at the moment of tooth-brushing. Removing the phone while leaving that environment intact removes a fraction of the signal. The remedy has been aimed at the object people feel guilty about rather than at the variable governing the response, which is total light entering the eye. (The comparison between screen and room lighting is an inference from the dose–response data rather than a measured finding; there is surprisingly little published measurement of real-world household evening illuminance, which is itself worth knowing.)

A second finding in the same study should be fatal to screen advice delivered as a universal rule. Individual sensitivity varied enormously: the half-maximal dose ranged from 6 lux in the most sensitive participant to about 350 lux in the least, a spread the authors describe as more than fiftyfold among healthy young adults (Phillips et al., 2019). The same evening is registered by two people's circadian systems as entirely different stimuli. Advice calibrated to the average describes almost nobody. The fiftyfold figure is itself a ratio between two imprecise extremes — thirteen of the fifty-five individual curve fits did not meet the study's own precision threshold — but the direction is not in doubt.

How large is the light effect, really?

Melatonin suppression is a biomarker. A bad night is an outcome. The two are routinely treated as the same thing, and they are not.

The most useful compilation is a 2024 theoretical review in Sleep Medicine Reviews by Serena Bauducco and colleagues, which collected the independent tests of the bright-screen hypothesis and reported what each did to sleep onset latency. It is a theoretical review rather than a systematic one, so the set is not guaranteed exhaustive. Across eleven studies the differences run: +3.3, −1.9, +9.9, −2.0, +7.5, +0.4, +5.7, −3.9, +4.9, +5.0 and +0.5 minutes. Three of the eleven are negative — the bright screen was associated with falling asleep faster. The largest positive value in the whole set, 9.9 minutes, is Chang. The authors' summary is that screen light and arousal exert a small-to-negligible influence, with a delay of sleep onset of, at best, around ten minutes (Bauducco et al., 2024). The ten-minute figure is about how long people took to fall asleep; the review offers no equivalent number for how long they then slept.

The same review makes a pointed observation about how the public message was built: a single study — the one with the largest effect in the literature — was foregrounded in popular sleep writing and broadcasting until it became the general claim. That is not a failure of the science. It is a failure of the transmission.

A recent study shows what a realistic dose does. Christopher Höhn and colleagues gave male adolescents and young men ninety minutes of evening smartphone reading across three conditions — no filter, blue-light filter, and a printed book — with sleep measured in the laboratory by polysomnography, the full electrode-based standard, plus salivary melatonin. (Male-only samples are common in this literature and are a real limit on how far its findings travel.) Both smartphone conditions attenuated melatonin relative to the book, and the blue-light filter did not prevent it. Adolescents had fully recovered by bedtime fifty minutes later; adults had not. Sleep architecture showed no significant overall light effect, and overnight memory consolidation was untouched (Höhn et al., 2024). One detail cuts the other way and belongs on the record: in the first quarter of the night, adults reading without the filter lost some deep sleep, and those reading with it did not. That is the clearest filter-specific benefit anywhere in this essay's sources, and it is a single secondary outcome in a male-only sample.

That is the shape of the finding. A measurable dip in a hormone, mostly recovered before sleep, with no detectable consequence for the sleep that followed. Chang's own study found no difference in total sleep time or sleep efficiency either.

Two objections deserve answering before this is taken too far, because both are good.

The first is that these are different endpoints, and that this essay has swapped one for the other without saying so. Phillips found that 30 lux delayed melatonin onset by more than an hour. That is not a trivial biomarker wobble; a body clock pushed later while the alarm stays where it is is lost sleep, and a study measuring how many minutes someone took to fall asleep on a single night is structurally incapable of detecting it. The objection is correct, and it does not rescue the phone. It relocates the problem. The phase-delay evidence is about the total evening light dose reaching the eye over hours, which is exactly the quantity a screen contributes a minority share of in a lit room. Read properly, the strongest light finding in this literature is an argument about the lamp, the ceiling fixture and the bathroom mirror — not about the object in your hand.

The second is chronicity. Every controlled null described here is acute: a single night, a week, at most a fortnight. Nobody has run a randomised trial of ten years of nightly phone use, and nobody will. Ten minutes a night is a rounding error once; it is not obviously a rounding error across a decade, and an effect too small to surface in a two-week trial can still be real. That is a genuine limit on how far any of the null results can be pushed, and it applies to this essay's conclusions as much as to the claims it is correcting.

Does the fix work?

The blue-light account produced a specific, testable prediction: filter the short wavelengths and the problem should go away. It has been tested repeatedly, with unusually good designs, and it has not gone well.

Start with the phone setting. When researchers measured melatonin suppression across iPad Night Shift configurations, suppression did not significantly differ between the warm and cool settings; their conclusion was that changing a display's spectral composition without changing its brightness may be insufficient (Nagare, Plitnick, & Figueiro, 2019). A subsequent randomised trial put 167 emerging adults through seven nights of phone use in the hour before bed under three conditions — Night Shift on, Night Shift off, and no phone at all — with sleep measured by actigraphy — a wrist-worn motion sensor that infers sleep from stillness, and the standard objective instrument in this field outside a laboratory. There were no significant differences across the three groups on any sleep outcome. Not between the Night Shift settings, and not between using a phone and using nothing (Duraccio, Zaugg, Blackburn, & Jensen, 2021).

The most carefully controlled test of screen filtering is cleverer still, though it is single-blind and small — 55 participants. Every participant installed f.lux; only the experimental group had the blue-reduction feature switched on, and participants did not know which group they were in. Sleep was measured by actigraphy for a week. The experimental group showed no greater improvement in objective sleep, self-reported sleep, or mental health — and rated the software as more distracting, disabling it more often (Smidt, Blake, Latham, & Allen, 2022).

For the glasses, there are now two syntheses. A Cochrane review of seventeen randomised trials concluded that it is not known whether blue-light filtering spectacle lenses improve sleep quality: six trials comprising 148 participants split three to three between improvement and no difference, and the certainty of evidence was graded very low (Singh et al., 2023). A 2025 meta-analysis restricted to randomised crossover trials with actigraphy-derived outcomes found a pooled effect on sleep onset latency of −4.86 minutes (95% CI −20.23 to 10.52) and on total sleep time of +8.75 minutes (95% CI −35.31 to 52.82), with zero heterogeneity — and, importantly, only three double-blind trials totalling 49 participants (Luna-Rangel et al., 2025). The point estimates are small and the intervals are wide enough to be uninformative. This is not a demonstration that filters do nothing. It is an absence of good evidence that they do something.

One study rescues the mechanism while condemning the product. Using displays engineered to vary the melanopic component by a factor of three or four while holding brightness and apparent colour constant — the separation Chang could not achieve — researchers found that low-melanopic light did shorten sleep latency, attenuate melatonin suppression, and advance melatonin onset, in proportion to the dose (Schöllhorn et al., 2023) — in 72 men, which is the usual limit on this literature. So spectrum genuinely matters, if you change enough of it. A consumer warm-tint setting does not change enough of it. These are not contradictory results. They are a mechanism study and a product evaluation, and they disagree about the product.

The second pathway: the one that feels true

Take blue light away from someone and they reach for the next explanation without pausing, because it is obvious. It is not the light. It is that you are lying there at midnight reading something infuriating.

The observational literature agrees. In the National Sleep Foundation's 2011 poll of 1,508 Americans, interactive devices — computers, phones, consoles — used in the hour before bed were associated with difficulty falling asleep and unrefreshing sleep, while passive ones — television, music players — were not (Gradisar et al., 2013). Both emit light; only one tracked with trouble. Path analysis on 101 adolescents later found two routes running in parallel from fear of missing out to shorter sleep: a behavioural one that pushed bedtimes later, and a cognitive one that raised pre-sleep arousal and lengthened sleep onset (Scott & Woods, 2018).

Coherent, mechanistic, satisfying — and almost entirely self-reported questionnaire data, gathered at one moment, with no way to establish what came first.

The controlled tests do not support it.

The decisive experiment is small and well built. Selina Ladina Combertaldi and colleagues brought 32 volunteers into a sleep laboratory and compared thirty minutes of pre-bed social media against thirty minutes of progressive muscle relaxation and against a neutral night, with the effects of blue light excluded by design. Social media did not significantly raise arousal and disturbed neither objective nor subjective sleep. The relaxation condition, on the same nights in the same paradigm, did exactly what theory predicts — lower heart rate, better sleep efficiency, shorter sleep onset (Combertaldi et al., 2021).

That second half is what makes the first half informative. A study that finds nothing is usually a study that could not detect anything. This one demonstrably could. It looked at social media and found nothing there.

The pattern repeats. Twenty adolescents given three counterbalanced conditions — social media, a book on a smartphone, a physical book — showed no differences in sleep quality, pre-sleep arousal or memory consolidation, and rated the social media condition as less stimulating than the reading (Sennock et al., 2024). Fifty participants shown a suspenseful television series had their stress, heart rate and cortisol successfully raised, most of all after a cliffhanger — and then fell asleep faster than after a neutral series (Baselgia et al., 2023). Induced arousal did not lengthen sleep onset. It shortened it. Something did survive into the night — after cliffhangers specifically, one index of sleep quality was lower — so the finding is that arousal fails to do the thing it is always said to do, not that it does nothing at all.

The largest observational tests agree. Among 45,202 Norwegian university students, an extra hour of screen use after getting into bed was associated with substantially higher odds of insomnia and about twenty-four minutes less sleep — but the associations did not differ between social media and other activities (Hjetland et al., 2025). And a 2026 meta-analysis in JAMA Pediatrics pooling within-person associations across 25 studies and 4,562 participants found screen time associated with later sleep onset and with nothing else — no within-person association with total sleep time, sleep latency, efficiency, or subjective quality. Timing did moderate one thing — screen use after getting into bed was associated with worse subjective sleep quality where daily or evening use was not — while screen type moderated nothing at all (Bourke, Maddren, Sippel, & Thomas, 2026).

None of this makes pre-sleep arousal imaginary. When 22 healthy sleepers were simply instructed to sleep badly, their objectively measured sleep onset lengthened and they woke more often (Combertaldi & Rasch, 2020). A psychological state can wreck a night. The finding is narrower: phone content does not reliably produce enough of that state to show up. The feed is absorbing. It is not, on the evidence, more activating than a novel.

The third pathway: the hours themselves

Which leaves the explanation nobody finds interesting.

Liese Exelmans and Jan Van den Bulck named the arithmetic. They distinguished bedtime — getting into bed — from shuteye time, actually attempting sleep. Across 338 young adults the average gap was 39 minutes, and the longer the gap the worse the reported sleep: those with a shuteye latency over an hour were more than nine times as likely to be rated poor sleepers (Exelmans & Van den Bulck, 2017). In a representative sample of 584 adults aged 18 to 96, only about one in five tried to sleep as soon as they got into bed (Exelmans, Gradisar, & Van den Bulck, 2018). Among 4,010 Norwegian sixteen- and seventeen-year-olds, mean shuteye latency on school nights was 43 minutes and mean sleep was six hours forty-three; the authors call the long shuteye latency a main driver of short school-day sleep (Saxvig et al., 2021).

One detail resists the obvious reading. In the 2017 study, shuteye latency was related to media use before getting into bed, not to media use in bed. In the later representative sample, people with a shuteye latency over thirty minutes used both television and smartphones more than those under thirty minutes, with no equivalent difference for non-screen activities. The pre-bed hour appears to do more work than the in-bed minutes, and the delay is media-shaped rather than a general tendency to lie awake.

The behaviour has a literature. Bedtime procrastination — going to bed later than intended with nothing preventing sleep — was named by Floor Kroese and colleagues, who found in 2,431 Dutch adults that poorer self-regulation predicted insufficient sleep, mediated by bedtime procrastination (Kroese, Evers, Adriaanse, & de Ridder, 2016). The meta-analysis of that literature — 43 studies — found the expected correlations but graded the evidence base low under GRADE, and found that evening chronotype — being a night owl — correlated at least as strongly as self-control did, with the two intervals heavily overlapping (Hill et al., 2022). Whatever this is, it is not purely a character failing. Meanwhile a daily diary study of 108 employees found that on evenings when people had less self-regulatory resource, they procrastinated bedtime less — a result its authors describe as contradicting the idea that bedtime procrastination reflects depleted willpower (Kühnel, Syrek, & Dreher, 2018).

The size of this pathway is what distinguishes it. The gap between getting into bed and attempting sleep runs to tens of minutes — 39 in one adult sample, 43 on school nights across four thousand Norwegian adolescents — against under ten minutes for screen light and under ten for arousing content. Displacement is not a competing explanation of the same size. It is several times larger than the two mechanisms that absorb all the attention, and it is the only one of the three a person can verify without instruments, by looking at the clock.

One intervention study points the same way, and it should be read with care. When adolescents were asked to stop phone use an hour before bed for a school week, they stopped 80 minutes earlier, turned the lights out 17 minutes earlier, and slept 21 minutes longer (Bartel, Scheeren, & Gradisar, 2019). Note the exchange rate: eighty minutes of surrendered phone time bought twenty-one minutes of sleep. Note also the design — 63 adolescents, no control group, no randomisation, sleep by self-reported diary, and only about a quarter of those approached willing to take part. It is consistent with displacement being the operative mechanism. It is not evidence that a curfew works, and the trials that tested that question properly reached a different answer, described further down.

The phone that is switched off

If light were the whole account, one prediction follows: a phone you never look at should be harmless. A dark device on a bedside table is, photically, a paperweight.

That prediction fails. Ben Carter and colleagues' meta-analysis of twenty studies and 125,198 school-age children found that children using a portable screen device at bedtime were about twice as likely to get inadequate sleep as those who did not. That is the expected result. The unexpected one is what happened to children who merely had a device available in the room and did not report using it: they were about three-quarters more likely to get inadequate sleep, and on daytime sleepiness the two groups were close to indistinguishable (Carter, Rees, Hale, Bhattacharjee, & Paradkar, 2016).

In the review's own numbers, the odds ratios for use were 2.17 for inadequate sleep quantity, 1.46 for poor sleep quality and 2.72 for daytime sleepiness; for access without use, 1.79, 1.53 and 2.27. The second row is not much smaller than the first, and on two of the three outcomes it is barely smaller at all.

Every one of those twenty studies was cross-sectional, heterogeneity was severe, exposure was self- or parent-reported, and households that permit bedroom phones differ from those that do not in ways that affect sleep independently. It is an association in children, and it should not be extended to adults without argument. But whatever explains a dark phone costing sleep, it is not melatonin suppression by a screen nobody switched on, and it is not arousing content nobody read. In one survey of 454 adolescents, 62.9 per cent took their phone to bed, 56.8 per cent left it on overnight, and 7.9 per cent were woken by a message at least twice a week (Adachi-Mejia et al., 2014) — a single-clinic pilot, not a national estimate, but a reminder that a phone in a bedroom is a device that can interrupt.

What happens when people actually change it

The intervention evidence is thinner, and weaker, than the confidence of the advice built on it.

The best-measured randomised trial is the Danish SCREENS study: 89 families, household recreational screen media capped at three hours a week for a fortnight, with sleep measured by EEG and 97 per cent compliance among children. Children's physical activity rose by about 46 minutes a day; the adults' did not move. Sleep did not move either — no significant between-group differences on any EEG-based sleep outcome, in children or adults (Pedersen et al., 2022).

It is not alone. A four-week randomised trial imposing a 10pm curfew on high-school athletes found no improvement in sleep, performance or mood (Harris et al., 2015). The 167-person accelerometry trial described earlier found nothing in its primary analysis, even in the complete-abstinence arm (Duraccio et al., 2021). And in a randomised pilot of an online programme to cut evening screen use, every group improved by roughly thirteen minutes — including the active control (Hill et al., 2025).

The positive trials share a feature. In a randomised pilot, 38 students whose phones were locked by screen-time controls for thirty minutes before bed reported shorter latency, longer sleep, better mood and lower arousal (He, Tu, Xiao, Su, & Tang, 2020). Every sleep outcome was self-reported; the control group received no intervention whatsoever, while the treatment group got a nightly reminder text and verification calls from the researchers. That contrast is not restriction against no restriction. It is restriction plus sustained attention against nothing.

For scale: cognitive behavioural therapy for insomnia — the best-evidenced treatment there is, delivered over weeks by a clinician and tested across twenty randomised trials — reduces sleep onset latency by about 19 minutes, with a total-sleep-time effect statistically indistinguishable from zero (Trauer, Qian, Doyle, Rajaratnam, & Cunnington, 2015). That is the ceiling for this entire category of problem. Any phone intervention claiming to beat it by a wide margin, measured by questionnaire, in an unblinded design, is describing expectation.

Two details from that consensus panel belong here. The first cuts against this essay, so it should not be buried: the panel did reach consensus — one of only five statements out of ten to do so — that behavioural strategies can reduce the negative effects of screen use on sleep, particularly reductions in evening use (Hartstein et al., 2024). Sixteen experts could not agree that the light matters, and could agree that changing the behaviour helps. That is a coherent position, and it is roughly where this essay arrives from the other direction.

The second is scope. The panel records that it did not address questions of duration or timing of screen use. The rule about stopping an hour or two before bed — the most repeated piece of sleep advice of the past decade — was never adjudicated by the body whose consensus statement is cited for it. It remains a reasonable inference from mechanism and association. It has never been a finding.

Which way does the arrow run?

The longitudinal studies keep returning an uncomfortable answer.

Follow 942 university students annually for three years and sleep problems predict later media use, while media use does not predict later sleep problems; the authors read this as students reaching for media to cope with sleep they were already not getting (Tavernier & Willoughby, 2014).

Three larger studies then find the relationship running in both directions at once — 1,620 Swedish adolescents over a year (Mazzer, Bauducco, Linton, & Boersma, 2018), 4,641 American children in the ABCD cohort, analysed in separate discovery and replication samples (Zhao et al., 2024), and 2,500 Czech adolescents across three waves (Tkaczyk, Ksinan, & Smahel, 2026). In all three, the path from poor or late sleep back to screen use was the equal or stronger leg. Each set of authors draws the same practical moral, and it is not the familiar one: the Swedish group warns that young people turn to devices because they are struggling to sleep, and the Czech group recommends that interventions prioritise a consistent sleep schedule over reducing screen use. The ABCD analysis adds the complication that ought to temper everyone — externalising symptoms predicted both later sleep problems and later screen use, which is the third-variable structure any causal story here most needs to exclude and none has.

Then there is the experiment that settles the direction at least once. A randomised crossover trial put 105 children through a week of sleep extension and a week of sleep restriction, measuring screen use with wearable cameras. Randomising children to less sleep produced more screen time — around an hour more after school in the camera data, and about seventeen minutes more by parent report (Jackson et al., 2025). The camera result depends on an imputation rule for blocked images and only 49 children had usable footage, so it should be held with care. But the direction of that manipulation is not in question, because the experimenters controlled it.

Short sleep, in that trial, produced screen use. Whatever else is true, causation in this area runs in both directions — and the decade of advice was written as though it ran in one.

What the evidence supports

Well supported. The photobiology: a melanopsin-based receptor system peaking around 446–477 nm governs melatonin suppression (Brainard et al., 2001; Thapan et al., 2001). Evening light affects circadian timing, and the system is sensitive at ordinary indoor levels with more than fiftyfold variation between individuals (Phillips et al., 2019). Bedtime device use is consistently associated with later bedtimes and shorter sleep, and within individuals the surviving association is with when the screen is used, not what is on it (Bourke et al., 2026). Sleep is displaced by a large, measurable amount of time spent awake in and around bed (Exelmans & Van den Bulck, 2017; Saxvig et al., 2021).

Reasonably supported. The relationship is bidirectional, with the sleep-to-phone path at least as strong as the reverse (Tavernier & Willoughby, 2014; Zhao et al., 2024; Tkaczyk et al., 2026), and experimentally induced short sleep increases screen use (Jackson et al., 2025). Device presence tracks with worse sleep even without reported use (Carter et al., 2016).

Weakly supported, or contradicted. That screen light meaningfully harms sleep: effects under ten minutes across eleven tests with three pointing the wrong way (Bauducco et al., 2024), no sleep-architecture effect at a realistic dose (Höhn et al., 2024). A 2024 expert panel could not reach consensus that it does at any age (Hartstein et al., 2024) — which is an absence of agreement rather than evidence of absence, and is listed here for that reason and no stronger one. That phone content is unusually arousing: contradicted by every controlled experiment holding light constant (Combertaldi et al., 2021; Sennock et al., 2024; Baselgia et al., 2023) and by the largest observational tests (Hjetland et al., 2025; Bourke et al., 2026). That blue-light filtering helps: very low certainty (Singh et al., 2023), a pooled actigraphic estimate whose confidence interval spans zero (Luna-Rangel et al., 2025), and null under blinding (Smidt et al., 2022). That a screen curfew produces more sleep: null in the best-measured randomised trials (Pedersen et al., 2022; Harris et al., 2015; Duraccio et al., 2021).

Bounded. A preregistered analysis of 50,212 children puts total screen time at under two per cent of the variance in sleep outcomes (Przybylski, 2019). Whatever is happening is real and modest, and it is not the dominant determinant of anyone's sleep.

What to do, given the evidence

In descending order of how much the evidence behind them can bear:

1. Fix the schedule, not the screen. The one thing that reliably moves is when sleep starts (Bourke et al., 2026), and the loop between late bedtimes and phone use runs in both directions (Tkaczyk et al., 2026). A consistent bedtime and wake time is the larger lever, and it is what the field's own longitudinal researchers now recommend prioritising.

2. Count the hour, not the app. What predicts trouble is being awake at midnight, not what you were awake looking at. Swapping the feed for a novel on the same device changes very little (Sennock et al., 2024). If you set a rule, make it about the time.

3. If you manage light, manage the room. The circadian system responds to total light reaching the eye, and is half-maximally suppressed at levels well below a normally lit living room (Phillips et al., 2019). Dimming the overhead light in the last hour plausibly acts on more of that dose than any single screen setting — plausibly rather than demonstrably, since the screen-versus-room comparison is an inference from the dose–response curve and not something anyone has measured in real homes. Warm screen modes are not harmful; they are a small intervention presented as a decisive one.

4. Stop the phone waking you. The presence finding is cross-sectional and confounded, but a device that can interrupt a sleeping person is a device worth silencing or removing (Carter et al., 2016; Adachi-Mejia et al., 2014). This is the one phone intervention whose mechanism is not in dispute, because it does not require any theory at all.

5. Do not buy the glasses on the strength of the sleep claim. Very low certainty in Cochrane, a pooled actigraphic effect indistinguishable from zero, and null results under proper blinding (Singh et al., 2023; Luna-Rangel et al., 2025; Smidt et al., 2022).

6. If sleep is genuinely bad, treat it as a sleep problem. Persistent insomnia has a well-evidenced treatment, and it is not a screen rule. The most common failure mode here is spending a year adjusting device settings for a condition that needed a clinician.

A note on certainty

The blue-light story is not wrong because light is harmless. It is wrong in a more specific and more instructive way. It took a real mechanism, established under monochromatic light through dilated pupils at three in the morning, and carried it into a claim about a phone in a lit bedroom. It took a twelve-person inpatient study that compared an iPad at full brightness against a book in near-darkness, and whose authors wrote plainly that they could not separate blue from bright, and made it the settled public explanation — with a product attached at every step.

Three things survive the reading. The mechanism is real, and the circadian system is more light-sensitive than most people assume, which argues for dimming the room rather than tinting the screen. The effects on sleep itself, wherever sleep is measured rather than remembered, are small enough to disappear into the noise of a well-controlled trial. And the causation runs in both directions, with one randomised experiment showing that making children sleep less made them use screens more.

None of that makes the phone innocent. It makes it smaller and more specific than the alarm, and largely immune to the remedy that was sold for it. What is left, once the colour and the content have both been set aside, is a clock problem wearing the costume of a physiological one — which is why a decade of advice aimed at the retina has produced so little.

The dullest instruction in this essay is the best supported, and it is this. Whatever time you had intended to turn the light off, turn it off then.

---

References

• Adachi-Mejia, A. M., Edwards, P. M., Gilbert-Diamond, D., Greenough, G. P., & Olson, A. L. (2014). TXT me I'm only sleeping: Adolescents with mobile phones in their bedroom. Family & Community Health, 37(4), 252–257.
• Bartel, K., Scheeren, R., & Gradisar, M. (2019). Altering adolescents' pre-bedtime phone use to achieve better sleep health. Health Communication, 34(4), 456–462.
• Baselgia, S., Combertaldi, S. L., Fahr, A., Wirz, D. S., Ort, A., & Rasch, B. (2023). Pre-sleep arousal induced by suspenseful series and cliffhangers have only minor effects on sleep: A sleep laboratory study. Sleep Medicine, 102, 186–198.
• Bauducco, S., Pillion, M., Bartel, K., Reynolds, C., Kahn, M., & Gradisar, M. (2024). A bidirectional model of sleep and technology use: A theoretical review of how much, for whom, and which mechanisms. Sleep Medicine Reviews, 76, 101933.
• Bourke, M., Maddren, C. I., Sippel, F., & Thomas, G. (2026). Within-person association between daily screen use and sleep in youth: A systematic review and meta-analysis. JAMA Pediatrics, 180(5), 500–509.
• Brainard, G. C., Hanifin, J. P., Greeson, J. M., Byrne, B., Glickman, G., Gerner, E., & Rollag, M. D. (2001). Action spectrum for melatonin regulation in humans: Evidence for a novel circadian photoreceptor. The Journal of Neuroscience, 21(16), 6405–6412.
• Brown, T. M., Brainard, G. C., Cajochen, C., Czeisler, C. A., Hanifin, J. P., Lockley, S. W., Lucas, R. J., Münch, M., O'Hagan, J. B., Peirson, S. N., Price, L. L. A., Roenneberg, T., Schlangen, L. J. M., Skene, D. J., Spitschan, M., Vetter, C., Zee, P. C., & Wright, K. P. (2022). Recommendations for daytime, evening, and nighttime indoor light exposure to best support physiology, sleep, and wakefulness in healthy adults. PLOS Biology, 20(3), e3001571.
• Carter, B., Rees, P., Hale, L., Bhattacharjee, D., & Paradkar, M. S. (2016). Association between portable screen-based media device access or use and sleep outcomes: A systematic review and meta-analysis. JAMA Pediatrics, 170(12), 1202–1208.
• Chang, A.-M., Aeschbach, D., Duffy, J. F., & Czeisler, C. A. (2015). Evening use of light-emitting eReaders negatively affects sleep, circadian timing, and next-morning alertness. Proceedings of the National Academy of Sciences, 112(4), 1232–1237.
• Chinoy, E. D., Duffy, J. F., & Czeisler, C. A. (2018). Unrestricted evening use of light-emitting tablet computers delays self-selected bedtime and disrupts circadian timing and alertness. Physiological Reports, 6(10), e13692.
• Chkhaidze, A., Millar, B. M., Revenson, T. A., & Mindlis, I. (2025). Scrolling your sleep away: The effects of bedtime device use on sleep among young adults with poor sleep. International Journal of Behavioral Medicine, 32(4), 634–639.
• Combertaldi, S. L., Ort, A., Cordi, M., Fahr, A., & Rasch, B. (2021). Pre-sleep social media use does not strongly disturb sleep: A sleep laboratory study in healthy young participants. Sleep Medicine, 87, 191–202.
• Combertaldi, S. L., & Rasch, B. (2020). Healthy sleepers can worsen their sleep by wanting to do so: The effects of intention on objective and subjective sleep parameters. Nature and Science of Sleep, 12, 981–997.
• Duraccio, K. M., Zaugg, K. K., Blackburn, R. C., & Jensen, C. D. (2021). Does iPhone night shift mitigate negative effects of smartphone use on sleep outcomes in emerging adults? Sleep Health, 7(4), 478–484.
• Exelmans, L., Gradisar, M., & Van den Bulck, J. (2018). Sleep latency versus shuteye latency: Prevalence, predictors and relation to insomnia symptoms in a representative sample of adults. Journal of Sleep Research, 27(6), e12737.
• Exelmans, L., & Van den Bulck, J. (2017). Bedtime, shuteye time and electronic media: Sleep displacement is a two-step process. Journal of Sleep Research, 26(3), 364–370.
• Gradisar, M., Wolfson, A. R., Harvey, A. G., Hale, L., Rosenberg, R., & Czeisler, C. A. (2013). The sleep and technology use of Americans: Findings from the National Sleep Foundation's 2011 Sleep in America poll. Journal of Clinical Sleep Medicine, 9(12), 1291–1299.
• Harris, A., Gundersen, H., Mørk-Andreassen, P., Thun, E., Bjorvatn, B., & Pallesen, S. (2015). Restricted use of electronic media, sleep, performance, and mood in high school athletes — A randomized trial. Sleep Health, 1(4), 314–321.
• Hartstein, L. E., Mathew, G. M., Reichenberger, D. A., Rodriguez, I., Allen, N., Chang, A.-M., Chaput, J.-P., Christakis, D. A., Garrison, M., Gooley, J. J., Koos, J. A., Van den Bulck, J., Woods, H., Zeitzer, J. M., Dzierzewski, J. M., & Hale, L. (2024). The impact of screen use on sleep health across the lifespan: A National Sleep Foundation consensus statement. Sleep Health, 10(4), 373–384.
• He, J.-W., Tu, Z.-H., Xiao, L., Su, T., & Tang, Y.-X. (2020). Effect of restricting bedtime mobile phone use on sleep, arousal, mood, and working memory: A randomized pilot trial. PLoS ONE, 15(2), e0228756.
• Hill, V. M., Ferguson, S. A., Rebar, A. L., Meaklim, H., & Vincent, G. E. (2025). A randomised pilot trial for bedtime procrastination: Examining the efficacy and feasibility of the Reducing Evening Screen Time online intervention (REST-O). Sleep Medicine, 129, 306–315.
• Hill, V. M., Rebar, A. L., Ferguson, S. A., Shriane, A. E., & Vincent, G. E. (2022). Go to bed! A systematic review and meta-analysis of bedtime procrastination correlates and sleep outcomes. Sleep Medicine Reviews, 66, 101697.
• Hjetland, G. J., Skogen, J. C., Hysing, M., Gradisar, M., & Sivertsen, B. (2025). How and when screens are used: Comparing different screen activities and sleep in Norwegian university students. Frontiers in Psychiatry, 16, 1548273.
• Höhn, C., Hahn, M. A., Gruber, G., Pletzer, B., Cajochen, C., & Hoedlmoser, K. (2024). Effects of evening smartphone use on sleep and declarative memory consolidation in male adolescents and young adults. Brain Communications, 6(3), fcae173.
• Jackson, R. F., Meredith-Jones, K. A., Haszard, J. J., Galland, B. C., Morrison, S., Jaques, M., & Taylor, R. W. (2025). The impact of sleep loss on screen time in children: Secondary analyses of a randomised crossover trial using objective measures of screen time. Pediatric Obesity, 20(12), e70050.
• Kroese, F. M., Evers, C., Adriaanse, M. A., & de Ridder, D. T. D. (2016). Bedtime procrastination: A self-regulation perspective on sleep insufficiency in the general population. Journal of Health Psychology, 21(5), 853–862.
• Kühnel, J., Syrek, C. J., & Dreher, A. (2018). Why don't you go to bed on time? A daily diary study on the relationships between chronotype, self-control resources and the phenomenon of bedtime procrastination. Frontiers in Psychology, 9, 77.
• Luna-Rangel, F. A., Gonzalez-Bedolla, B., Salazar-Ortega, M. J., Torres-Mancilla, X. M., & Martinez-Cadena, S. (2025). Efficacy of blue-light blocking glasses on actigraphic sleep outcomes: A systematic review and meta-analysis of randomized controlled crossover trials. Frontiers in Neurology, 16, 1699303.
• Mazzer, K., Bauducco, S., Linton, S. J., & Boersma, K. (2018). Longitudinal associations between time spent using technology and sleep duration among adolescents. Journal of Adolescence, 66, 112–119.
• Nagare, R., Plitnick, B., & Figueiro, M. G. (2019). Does the iPad Night Shift mode reduce melatonin suppression? Lighting Research & Technology, 51(3), 373–383.
• Pedersen, J., Rasmussen, M. G. B., Sørensen, S. O., Mortensen, S. R., Olesen, L. G., Brønd, J. C., Brage, S., Kristensen, P. L., & Grøntved, A. (2022). Effects of limiting recreational screen media use on physical activity and sleep in families with children: A cluster randomized clinical trial. JAMA Pediatrics, 176(8), 741–749.
• Perrault, A. A., Bayer, L., Peuvrier, M., Afyouni, A., Ghisletta, P., Brockmann, C., Spiridon, M., Hulo Vesely, S., Haller, D. M., Pichon, S., Perrig, S., Schwartz, S., & Sterpenich, V. (2019). Reducing the use of screen electronic devices in the evening is associated with improved sleep and daytime vigilance in adolescents. Sleep, 42(9), zsz125.
• Phillips, A. J. K., Vidafar, P., Burns, A. C., McGlashan, E. M., Anderson, C., Rajaratnam, S. M. W., Lockley, S. W., & Cain, S. W. (2019). High sensitivity and interindividual variability in the response of the human circadian system to evening light. Proceedings of the National Academy of Sciences, 116(24), 12019–12024.
• Przybylski, A. K. (2019). Digital screen time and pediatric sleep: Evidence from a preregistered cohort study. The Journal of Pediatrics, 205, 218–223.
• Saxvig, I. W., Bjorvatn, B., Hysing, M., Sivertsen, B., Gradisar, M., & Pallesen, S. (2021). Sleep in older adolescents. Results from a large cross-sectional, population-based study. Journal of Sleep Research, 30(4), e13263.
• Schöllhorn, I., Stefani, O., Lucas, R. J., Spitschan, M., Slawik, H. C., & Cajochen, C. (2023). Melanopic irradiance defines the impact of evening display light on sleep latency, melatonin and alertness. Communications Biology, 6(1), 228.
• Scott, H., & Woods, H. C. (2018). Fear of missing out and sleep: Cognitive behavioural factors in adolescents' nighttime social media use. Journal of Adolescence, 68, 61–65.
• Sennock, S., von Lieres und Wilkau, K., Günther, A., Brandhorst, I., Zinke, K., Conzelmann, A., Renner, T. J., & Kurz, E.-M. (2024). Investigation of the influence of 45-minute pre-sleep social media use on sleep quality and memory consolidation in adolescents. Sleep Medicine, 124, 299–307.
• Singh, S., Keller, P. R., Busija, L., McMillan, P., Makrai, E., Lawrenson, J. G., Hull, C. C., & Downie, L. E. (2023). Blue-light filtering spectacle lenses for visual performance, sleep, and macular health in adults. Cochrane Database of Systematic Reviews, 8(8), CD013244.
• Smidt, A. M., Blake, M. J., Latham, M. D., & Allen, N. B. (2022). Effects of automated diurnal variation in electronic screen temperature on sleep quality in young adults: A randomized controlled trial. Behavioral Sleep Medicine, 20(5), 513–529.
• Tavernier, R., & Willoughby, T. (2014). Sleep problems: Predictor or outcome of media use among emerging adults at university? Journal of Sleep Research, 23(4), 389–396.
• Thapan, K., Arendt, J., & Skene, D. J. (2001). An action spectrum for melatonin suppression: Evidence for a novel non-rod, non-cone photoreceptor system in humans. The Journal of Physiology, 535(Pt 1), 261–267.
• Tkaczyk, M., Ksinan, A. J., & Smahel, D. (2026). Longitudinal between- and within-person associations among screen time, bedtime, and daytime sleepiness among adolescents: Three-wave prospective panel study. Journal of Medical Internet Research, 28, e78972.
• Trauer, J. M., Qian, M. Y., Doyle, J. S., Rajaratnam, S. M. W., & Cunnington, D. (2015). Cognitive behavioral therapy for chronic insomnia: A systematic review and meta-analysis. Annals of Internal Medicine, 163(3), 191–204.
• Willoughby, A. R., Golkashani, H. A., Ghorbani, S., Wong, K. F., Chee, N. I. Y. N., Ong, J. L., & Chee, M. W. L. (2024). Performance of wearable sleep trackers during nocturnal sleep and periods of simulated real-world smartphone use. Sleep Health, 10(3), 356–368.
• Zhao, Y., Paulus, M. P., Tapert, S. F., Bagot, K. S., Constable, R. T., Yaggi, H. K., Redeker, N. S., & Potenza, M. N. (2024). Screen time, sleep, brain structural neurobiology, and sequential associations with child and adolescent psychopathology: Insights from the ABCD study. Journal of Behavioral Addictions, 13(2), 542–553.