Blue Light Before Bed: The Light, or the Scrolling?

There is a small ritual a great many of us perform around ten at night. Brightness down. Amber filter on. Perhaps a pair of yellow-tinted glasses, if you have gone in properly. Then two more hours of the phone.

The ritual feels like maintenance, and it is worth asking whether the colour of the glow was ever the part doing the damage.

The belief, said kindly

Blue light from screens wrecks your sleep. The wavelength coming off the phone is the thing hurting your night, so filter it out and you have handled it.

This is a reasonable thing to believe and it did not come from nowhere. It grew out of some genuinely careful laboratory work on how the human body reads light. Then it travelled a very long way from what that work actually showed, picked up a product category on the way, and arrived in your evening as a rule about colour.

The part that is true

In 2000, a team exposed 23 healthy volunteers to a single six-and-a-half-hour bout of light in the early biological night and measured what it did to melatonin and to the timing of the body clock. Both responses followed a smooth dose-response curve, and the curve sat far lower than anyone had expected. Half of the maximum melatonin suppression arrived somewhere in the 50 to 130 lux band. Suppression was close to saturated by 200 lux. Ordinary dim room light of around 100 lux, the level you would describe as a bit gloomy, produced more than half the phase delay of a 9,000-lux blast.¹

That is the strongest evidence for the kernel, and it complicates the folk story rather than confirming it. The finding is about illuminance: how much light, for how long. It implicates the lamp on your side table as squarely as it implicates the phone. And it used six and a half hours of steady exposure, which is nobody’s idea of a scroll.

The study everyone quotes

Almost every version of the blue-light warning traces back to a single experiment. In 2015, twelve people were put in an inpatient sleep unit and asked to read for about four hours each evening for five evenings running, either on a light-emitting eReader at maximum brightness or on paper under dim light. The screen evenings suppressed melatonin by around 55 per cent, pushed the evening melatonin rise more than an hour and a half later, and lengthened the time it took to fall asleep from 15.8 minutes to 25.7.²

That last number rarely travels with the rest: about ten extra minutes.

And there is a line in the methods that changes how you should read the hormonal result. The device delivered only 31.7 photopic lux at the eye, below even the bottom of the 50-to-130-lux band where suppression turns half-maximal. The large effect came from four hours of it, five nights in a row. It was a dose experiment wearing the costume of a screen experiment. Twelve people, in a laboratory, doing something almost nobody does at home.

A more ordinary evening

A Swedish group ran something much closer to a normal night. Fourteen people read a novel from nine until eleven, on a self-luminous tablet or on paper, after a day that had included six and a half hours of bright light. Saliva melatonin, subjective sleepiness, time to fall asleep and sleep composition were all unaffected, and the full paper reports no change in EEG power during early slow-wave sleep either.³

Do not over-read that either. Fourteen participants, one night per condition, and a null result can never establish that nothing happened. The honest phrasing is that nothing was detectable at that dose. The daytime light is a real boundary condition too, and the authors were explicit about it: these people had had a properly bright day beforehand. Set alongside the four-hour version, though, it gives you the shape of the thing. Two hours of tablet at the end of a well-lit day did not register.

The fix does not do the job it is sold for

If the wavelength were the villain, changing the wavelength ought to help. That is the entire logic of night mode, and of the amber lenses. It is also where the evidence goes quiet.

A 2019 study tested Night Shift directly, comparing two Night Shift settings, one warm and one warmer, against a dim control and a deliberately high-stimulus condition. Melatonin suppression did not differ significantly between the two Night Shift settings, and the authors concluded that altering the spectrum while leaving brightness alone is insufficient.⁴ It is a small laboratory sample and one manufacturer’s implementation, so read it as a finding about that particular lever rather than about spectrum in principle.

The lenses have had a much longer run at it. A Cochrane review in 2023 gathered 17 randomised trials of blue-light-filtering spectacle lenses. Six assessed subjective sleep quality: three reported an improvement, three found none. Certainty of the evidence was graded very low, and the reviewers said plainly that they do not know whether these lenses are equivalent or superior to ordinary ones for sleep quality.⁵

The honest reading of that is “no reliable evidence of benefit”, which is a weaker and more careful statement than “useless”. The trials are small, short and mixed in design. What you can say with confidence is that the shelf of products ran a long way ahead of the evidence underneath it.

What survives when you look at everybody

Laboratories give you clean doses. To find out what actually predicts a bad night, you have to go and look at people’s lives. A 2025 Norwegian survey of 45,202 students aged 18 to 28 asked what they did on screens after getting into bed, and for how long. Each additional hour was associated with 59 per cent higher odds of insomnia symptoms and with 24 minutes less sleep. The striking part is what made no difference: the association did not vary significantly by what people were doing. Students who used only social media reported the lowest rates of insomnia symptoms and the longest sleep of anyone.⁶

The authors read this as displacement. The hour is the mechanism they land on. Time spent awake in bed with a device is time not spent asleep, and they note in passing that light exposure and arousal have shown minimal effects in experimental work, which is roughly where we have arrived here as well.

The usual caution applies with some force. This is cross-sectional and self-reported, so it cannot tell you which way the causal arrow runs, and people who sleep badly have more waking hours in bed to fill with something. Take it as the pattern at scale, not the proof of a cause.

The replacement villain is shaky too

It is tempting to swap one culprit for another and decide it was never the light, it was the scrolling: the arousal, the content, the small hit of alertness. That story deserves the same scrutiny, and it does not come through especially well. A sleep-laboratory study gave 32 young adults half an hour of social media before bed, with blue-light effects controlled, against progressive muscle relaxation and a neutral condition. Social media did not significantly raise arousal and disturbed neither measured nor reported sleep. The only change was slightly less stage N2, with slow-wave and spindle activity untouched. Relaxation, for its part, improved sleep efficiency and shortened the time taken to fall asleep.⁷

Thirty minutes, healthy young people, neutral content. That says nothing about an hour of a bad-tempered comment thread at midnight. What it does say is that arousal is a plausible secondary character rather than the established replacement for the villain we have just retired. The hour appears to be doing most of the work, and the rest is honestly still open.

The wonder

There is a fifth photoreceptor in the human eye, and it has nothing whatever to do with seeing. Alongside the rods and the three kinds of cone sits a small population of retinal ganglion cells containing a pigment called melanopsin. Their job is not to assemble an image. They are a light meter, wired more or less straight to the body clock, and they report a single number: how bright it is out there.

They are also extraordinarily sensitive. Around 100 lux, a level you would call a bit dim, a room you would happily read in, is already enough to move the clock more than half as far as a 9,000-lux flood.¹ Your eyes have been quietly filing brightness reports for as long as there have been evenings, long before there was anything on them worth scrolling.

Your body clock reads brightness and the hour far more than it reads colour. On the evidence so far, the phone looks likelier to cost you sleep by keeping you awake than by glowing blue.

What this might mean for your evening

Nothing above says the screen is harmless, and nothing above makes the amber filter a scandal. It says the lever most of us have been pulling is a small one, with two larger ones sitting beside it.

The first is brightness, across the whole room. The circadian evidence is about how much light reaches your eyes, so the ceiling light and the kitchen spots count every bit as much as the thing in your hand. Turning the room down is a bigger move than turning the screen orange.

The second is the hour. The clearest signal in the population data is simply time spent awake in bed with a device. If one thing is going to change, it is more likely to be when you put it down than what colour it was while you held it.

The day matters as well, which is the least intuitive part. The study that found nothing at all had given its participants a properly bright morning first. A body that has seen real daylight seems to read the evening differently, and that is a boundary condition worth respecting rather than a throwaway detail.

Some of this is why the wind-down work we make at Mythrae is audio, meant for a dark room with nothing to look at, a decision we wrote about in Why We Left the Screen Out. That is a design preference and a reading of the evidence above, and it is not a claim about what it will do for you. If you would like to hear where it goes, there is a free taster on the sleep page.

← Back to The Lab