Skip to content
terminally offline

research

Friction works until you turn it off

31 min read

Friction works until you turn it off

A considered reading of the research on making phones harder to use — the greyscale screen that shortens the session but not the reach, the app that works by asking a single question, and the trial in which only a quarter of the people who agreed to go without mobile internet for a fortnight kept the block in place.


The September essay here ended on an instruction it did not have the evidence to finish. Nearly nine in ten phone pick-ups are self-initiated (Heitmayer & Lahlou, 2021), it said, so the lever is the pick-up rather than the alert: rearrange the environment rather than resolving to try harder. That is the premise this site runs on, and it is a claim about mechanism. It deserves to be tested like one.

There is no shortage of things to test. The phone can be set to greyscale, which drains the colour from every icon and feed. Apps can be moved off the home screen into a folder, or deleted, or put behind a timer. Face ID can be switched off so that every unlock needs a passcode. An app can be installed that makes you wait and breathe before Instagram opens. The phone can be replaced, in effect or in fact, with one that only calls and texts. The common word for all of this is friction — making the unwanted thing slightly harder to do — and the case for it is borrowed from behavioural science, which has argued for the better part of two decades that changing the environment beats exhorting the person.

Here is the first surprise, and it is a pleasant one. The advice to switch off notifications went through a preregistered trial with objective logging and came out flat (Dekker, Baumgartner, Sumter, & Ohme, 2025). Friction has mostly been tested the same way, by measuring what people actually did with their phones, and what people did changed. Greyscale reduced measured screen time, though mostly in studies that compared people with their own baseline. Time limits reduced it in a large randomised trial. Blocking mobile internet roughly halved it.

The second surprise is less pleasant. The changes are mostly in how long people use the phone rather than how often they reach for it; the strongest friction proved the hardest to keep switched on; and when it comes off, most of the behaviour comes back. What follows is an attempt to say how much of the advice survives that.

Why friction should work

The argument begins with what a habit is. Wendy Wood and colleagues, who have spent more than two decades on the question, describe habits as responses that form when people repeat the same action in the same context, until the context alone is enough to set it off. Their review describes habits as the efficient, default mode of response — what happens when nothing intervenes (Wood & Rünger, 2016). The practical consequence is that a habit is vulnerable to anything that disrupts its cues.

The clearest demonstration is still one of the oldest. Wood, Tam and Witt followed students as they transferred to a new university and tracked three ordinary habits: exercising, reading the newspaper and watching television. The habits survived the move only when the context they depended on came with them — reading the paper with other people, for example. When the context changed, the habit was disrupted, and in some cases behaviour fell under the control of what people actually intended (Wood, Tam, & Witt, 2005). Disrupt the cue and you get a moment of choice back.

Social media offers its own example. Anderson and Wood reanalysed posting data from around a 2008 change to Facebook's design and found that it slowed posting only among the most habitual users — the people whose behaviour depended most on the old cues — even though the change increased posting overall (Anderson & Wood, 2021). The same paper names the industry's interest directly. Friction, in their definition, is the delays and hurdles people experience when accessing a platform, and platforms work to remove it, automating log-ins and speeding up load times. Lukoff and colleagues surveyed 120 YouTube users and found that the features most associated with a lost sense of agency were autoplay and recommendations — the two that make the choice of what to watch next on the user's behalf — while search and playlists, which leave that choice with the user, supported it (Lukoff et al., 2021).

So the theory is coherent, and the commercial incentive points the opposite way, which is a reasonable sign that the theory has teeth. If engagement is built by removing small costs, adding small costs back should reduce it.

The evidence friction borrows

Before the phone studies, a word on the general literature, because it is where most of the confidence in friction comes from and it is weaker than its reputation.

A large meta-analysis of nudges — choice-architecture interventions, of which friction is one kind — pooled more than 450 effect sizes from over 200 studies and two million participants, and reported an average effect of d = 0.45, small to medium. Interventions that changed the structure of a decision outperformed those that supplied information or helped people follow through on their intentions with reminders and commitment devices. Interventions that changed the effort involved — in that literature, usually by making the desirable option easier rather than the undesirable one harder — came in at d = 0.43 across twenty-three effects (S. Mertens, Herberz, Hahnel, & Brosch, 2022).

The same paper reported evidence of publication bias. Maier and colleagues reanalysed the data with a method designed to correct for it, and the average effect fell to 0.04, with evidence against an average effect for interventions that inform, remind or help people commit. Structural interventions, the family that greyscale and hidden apps belong to, were the one category where the evidence came out undecided rather than null (Maier et al., 2022). Commitment devices — the category a self-set time limit falls into — sit in one of the groups where the corrected evidence ran against an average effect, which makes the well-powered limit trial described below more notable, not less. An independent check points the same way. DellaVigna and Linos compared nudge trials published in academic journals with every trial run by two large government nudge units — 126 trials covering twenty-three million people. The published trials averaged an effect of 8.7 percentage points. The nudge-unit trials, whose complete records left no room for selective reporting, averaged 1.4. Selective publication, made worse by low statistical power, accounted for about seventy per cent of the gap (DellaVigna & Linos, 2022).

None of that is about phones. But it sets a prior. When a behavioural intervention is reported to work, the published estimate is probably an overestimate, and the best of the family — changing the structure of the choice — is better described as promising than as proven.

Greyscale: a shorter session, the same reach

Greyscale is the cleanest test of friction available, because it removes nothing. Every app still works. It simply makes the screen less rewarding to look at.

It reduces screen time, although most of the evidence compares people with themselves. Holte and Ferraro switched seventy-three undergraduates to greyscale for eight to ten days and found that their total screen time fell by an average of thirty-eight minutes a day compared with their own starting point, with social media and browsing down and video, oddly, unaffected (Holte & Ferraro, 2023). A follow-up with 133 students found reductions in screen time, problematic use and anxiety, though the authors noted that the change in anxiety may not be clinically meaningful, and depression did not change (Holte, Giesen, & Ferraro, 2023). Zimmermann and Sobolev ran a preregistered field experiment with 112 participants comparing greyscale against self-set time limits and a monitoring-only control. Greyscale produced an immediate, significant reduction in objectively measured screen time compared with the control; the time limits produced a smaller and more gradual one (Zimmermann & Sobolev, 2023). A later review of the literature puts that trial's reduction at about fifty minutes a day. It is the cleanest randomised evidence in the set. The behavioural result recurs on measured data, but only that trial shows it against a randomised control, and in the other controlled study, described below, the control group cut down too.

What it does not do is the more interesting part.

Dekker and Baumgartner tracked eighty-four people through a normal week and then a week in greyscale, logging both screen time and the number of times each phone was unlocked. Screen time fell by about twenty minutes a day. The number of unlocks did not change at all (d = 0.04). People picked their phones up as often as before and put them down sooner (Dekker & Baumgartner, 2024). The authors read this as evidence of deep-rooted checking habits, and after the September essay it is hard to read it any other way. The two earlier studies they review that counted unlocks had disagreed, one finding a decrease and one finding none, so this is not settled. But on the most careful measurement so far, greyscale acts on what the pick-up delivers and does little or nothing to the pick-up itself.

The study has other details worth knowing. There was no separate control group: the comparison is each person's ordinary week against their greyscale week, which leaves room for the novelty of being in a study. Participants reported more control over their phone use, less perceived overuse and less stress, but no change in productivity or sleep. They rated greyscale as more annoying than helpful. And when asked at the end whether it had reduced their screen time, they were on average unsure — when it had. The authors conclude that greyscale is an effective strategy, and add a caveat of their own: a twenty-minute reduction may not be substantial when daily use is still above four hours.

The well-being results across the set are thinner than the behavioural ones. In Holte and Ferraro's study, anxiety and depression improved in the control group too, so greyscale could not take the credit. Zimmermann and Sobolev found no immediate effect of the reduction on well-being or academic performance. Ochs and Sauer assigned ninety-seven students to greyscale, to moving their problem apps off the home screen, or to neither, for a week — and found that objective use fell in all three groups, the control included, while self-reported problematic use went up (Ochs & Sauer, 2022). The authors suggest that heightened awareness may explain both: being in the study may have been enough to change the behaviour, and attending to it may have made people rate it as more of a problem.

Whether people keep greyscale on varies a good deal. Olson and colleagues offered participants a menu of ten phone-reduction strategies; greyscale had some of the lowest compliance in their first study, kept by thirty-eight per cent, although around seven in ten of a second sample were still following a bundle of display changes including greyscale after two weeks (Olson, Sandra, Chmoulevitch, Raz, & Veissière, 2023). Dekker and Baumgartner, reviewing the earlier trials, note that Zimmermann and Sobolev's greyscale group reported keeping the setting on for only around half of the intervention period — while their own participants, most of whom wanted to cut down, reported keeping it on about ninety per cent of the time.

The summary is narrow but solid. Greyscale makes sessions shorter, by roughly twenty to fifty minutes a day in the studies that report a figure, most of them before-and-after comparisons. In the one study that counted unlocks carefully, it did not change how often people reached for the phone. It is a mild irritant, which is how it works, and how long people tolerate it depends on who they are.

Friction at the moment of reaching

If greyscale works on the session, the obvious next question is whether anything works on the reach itself.

The best evidence that something can comes from one sec, an app that intercepts a chosen app as it opens, puts up a screen asking the user to breathe in and out for a few seconds, and then offers a choice: continue, or don't. Grüning, Riedel and Lorenz-Spreen logged six weeks of use from 280 people. On average, thirty-six per cent of attempts to open a target app ended with the user backing out at the prompt. Over the six weeks, the attempts themselves fell, from 166 per user in the first week to 105 in the sixth, a drop of thirty-seven per cent. Combined, actual openings of the target apps fell by fifty-seven per cent (Grüning, Riedel, & Lorenz-Spreen, 2023).

That is a large effect, and unlike greyscale it lands on the reach: people tried to open the chosen apps less often as the weeks went on. It comes with caveats, most of them stated by the authors. The participants had downloaded one sec of their own accord and were recruited through the app itself. There was no control group and no measurement from before the app was installed, so the declines are measured from a first week in which one sec was already running; the authors identify the resulting self-selection as a confound and call for a randomised trial. One of the authors, Riedel, created the app, which the paper discloses. And the 280 were those who stayed in the study for all six weeks. Of 719 new users recruited, everyone who deleted the app or left the study early was excluded from the main analysis — the authors report that including them gave similar results — so fewer than two in five of the people who started are in the headline figure.

The more useful finding may be the experiment the authors ran alongside the field data. They separated the app's three ingredients — a message prompting reflection, a short enforced delay, and an explicit option to back out — and tested them in a preregistered online experiment in which 500 participants watched real viral video clips. The option to back out had the strongest effect. The delay helped on its own but added nothing once the option to back out was present. The reflective message did nothing. The setting was artificial — skipping videos in an online task, not opening apps on a phone — but the implication is suggestive: the active ingredient in one sec may not be friction in the sense of effort at all. It may be a question, put at the moment of reaching, with a single tap to say no.

The same pattern appears elsewhere. Lyngs and colleagues gave fifty-eight students a browser extension that either removed the Facebook news feed, or asked them to type why they were opening Facebook and then reminded them of the answer, or — as a control — turned the background white. The goal reminder cut median daily time on Facebook from about twenty-seven minutes to fifteen, and visits from twenty-nine a day to eleven. Removing the feed was associated with shorter visits. Both came with complaints: the reminders were annoying, and the missing feed left some people worried about what they were missing (Lyngs et al., 2020). Hiniker and colleagues' MyTime, in a small two-week deployment with twenty-three people, reduced time in the apps participants themselves considered a poor use of time by twenty-one per cent while leaving the apps they valued untouched (Hiniker, Hong, Kohno, & Kientz, 2016). In a randomised trial of seventy people, half of whom used it for three weeks, an app called Wellspent — which put up a full-screen choice to quit or continue whenever a session ran past a self-set limit — cut reported daily time on each user's most problematic app by around twenty-nine minutes relative to the control group, without a significant change on a standard measure of problematic social media use (L. Mertens et al., 2026).

Lyngs's study also caught something the others could not. The extensions only ran on laptops, and almost every participant also used Facebook on a phone. Eighty-six per cent of the goal-reminder group said the intervention had changed how they split their Facebook use between the two devices, and some described moving the activities the extension interrupted onto their phones, more deliberately. The logs could not see it, so this rests on what participants said. The September essay closed with a study in which moving the phone away sent people to their computers (Heitmayer, 2025). This is the same effect running in the other direction.

Limits set in advance

The strongest version of the self-binding idea is the time limit set ahead of the moment, by a person who is not, at that point, the one being tempted.

The most rigorous test is Allcott, Gentzkow and Song's experiment with around two thousand American adults, which built its own app so that limits could not simply be waved away. The limits built into most phones are flexible; in this study, participants set daily limits for individual apps, changes took effect the next day, and in most cases the limits could not be overridden on the spot. When a limit ran out, participants could ask for more time, but most of them only after an enforced delay of two, five or twenty minutes, assigned at random; a fifth could not extend it at all, and a fifth could extend it straight away. Over twelve weeks, they reduced use of six of the most tempting apps — Facebook, Instagram, Twitter, Snapchat, web browsers and YouTube — by twenty-two minutes a day, or sixteen per cent. The effect declined only slightly over time. Seventy-eight per cent of participants set binding limits and were still using them at the end, without any incentive to do so, and people were on average willing to give up $4.20 for three weeks of access to the limit function (Allcott, Gentzkow, & Song, 2022).

The authors read this as clear evidence of self-control problems that people partly recognise, and their model attributes about thirty-one per cent of social media use to them. That is the case for friction made by economists, and it is a strong one. Several details belong alongside it. The limits barely moved subjective well-being: an effect of 0.04 standard deviations, not statistically significant, although participants did report less of the use that felt compulsive — using the phone longer than intended, or mindlessly. The study's surveys walked participants through setting limits that matched their own stated ideal, and the authors expect that simply offering the feature would have had smaller effects. The sample used Android phones, and the trial ran early in the pandemic. And of the people who signed up, only five per cent were already using an app to limit their phone (they were excluded from the experiment), which the authors flag as a puzzle: people value commitment, and almost none of them were using it. One of the explanations they offer is that their self-selected sample overstates demand for it.

Smaller studies point the same way with less force. Okeke and colleagues made the phone vibrate every five seconds once a daily Facebook limit had passed; use fell by more than twenty per cent over three weeks, and the reduction did not persist once the vibration was removed (Okeke, Sobolev, Dell, & Estrin, 2018).

And then there is the study that asks where the time went. Collis and Eggers randomised students to keep Facebook, Instagram and Snapchat to ten minutes a day across all their devices for two and a half months — an instructed limit rather than a hard block — while software tracked their activity on computers and Android phones, with iPhone users checked by screenshot. The students moved the time to instant messaging, spent no less time on their devices overall, and showed no significant change in well-being or grades (Collis & Eggers, 2022). A limit aimed at a single app is a limit on a single route.

The dumbphone, simulated

If moderate friction moves behaviour modestly, the obvious experiment is maximal friction: a phone that cannot do the things that make it compelling.

The strongest evidence here is recent. Castelo and colleagues recruited 467 iPhone users in the United States and Canada for a preregistered randomised trial in which an app blocked all mobile internet for two weeks. Calls and texts still worked. Computers still had the internet. Half the participants blocked first and the other half two weeks later, which gave the study a built-in replication (Castelo, Kushlev, Ward, Esterman, & Reiner, 2025).

The effects are large for this literature. Average daily screen time, among those who supplied screenshots of it, fell from 314 minutes to 161. Counting every participant who completed the surveys, whether or not they kept the block on, the intervention improved subjective well-being (d = 0.45), a composite of mental-health symptoms (d = 0.56) and an objective measure of sustained attention (d = 0.23). These are within-person effect sizes pooling both groups' before-and-after changes, which are not directly comparable with between-group effects; the delayed group on its own showed effects of 0.36, 0.52 and 0.28 when its turn came. The authors offer two comparisons that make for striking headlines — the attention improvement is about the size of ten years of age-related decline, and the effect on depression symptoms is larger than the meta-analytic effect of antidepressants — and, in the same paragraph, a caveat: their sample and intervention are quite different from those studied in clinical settings. Mediation analyses suggested that much of the well-being effect ran through what people did with the time — more socialising in person, more exercise and time outdoors, a greater sense of self-control, slightly more sleep. None of those explained the improvement in attention.

That is the strongest single argument for friction published to date. It is also the strongest argument about its limits, because of what happened to compliance.

Of the 467 people who had agreed to block their mobile internet for two weeks, 272 redeemed the code for the blocking app, 266 set the block up, and 119 — twenty-five and a half per cent of those who signed up — met the study's definition of compliance, which allowed the block to be switched off no more than four times. Put another way, about four in ten never installed the blocker at all, and of those who did and finished the study, a little under half kept it on. The authors call compliance "evidently difficult". The benefits held up when the non-compliers were counted, which is a point in the study's favour: even a partly kept block helped. But the measured benefit belongs to an intervention that most of the people who volunteered for it did not sustain for a fortnight.

The other caveats are ones the authors list themselves. The sample was motivated: eighty-three per cent rated their desire to cut down at five or more out of seven. Participants knew they were in a study about phones and well-being, and the authors acknowledge that expectancy and demand effects could have contributed to the self-reported outcomes — which is why the objective attention measure, the smallest of the three effects, is the hardest to explain away. And the people who stayed in the study to the end had better mental health and attention at baseline than those who left it.

Pieh and colleagues found a similar shape with a cruder instrument. They randomised 111 students and asked half of them to keep smartphone screen time at or below two hours a day for three weeks, down from an average of about four and a half, and found small-to-medium improvements in depressive symptoms, stress, sleep quality and well-being relative to the control group. The trial was not blinded. And screen time rose quickly once the three weeks were over; by the follow-up six weeks later it was approaching where it started (Pieh et al., 2025).

What about the real thing — people who have actually swapped their smartphones for basic phones? The research is almost entirely qualitative. Rothschild and Lindqvist interviewed fifteen people: eight long-term users of traditional dumbphones, five users of newer "designer" dumbphones and two of their developers. Their reasons included mental and physical health, religious devotion and political disaffiliation; the costs included the privilege required to opt out of a world that assumes a smartphone, and the interpersonal complications that follow (Rothschild & Lindqvist, 2024). It is an interesting study of a small, self-selected group. It is not evidence that dumbphones work, and this publication could find no controlled trial of an actual switch to a basic phone.

The compliance problem

Lay the studies side by side and adherence turns out to be a recurring weak point, though not a uniform one.

Some frictions were kept well. Seventy-eight per cent of Allcott's participants kept their limits to the end. Olson's second sample was still following most of the ten strategies six weeks on, which the authors put partly down to several of them being changes to phone settings that only need making once; in the first study, seventy-nine per cent were still using a passcode instead of fingerprint unlocking after two weeks, a friction met at every single unlock. Dekker and Baumgartner's participants reported keeping greyscale on about ninety per cent of the time. Others were not kept. Greyscale was kept by thirty-eight per cent of participants in Olson's first study and for about half the time in Zimmermann and Sobolev's. And the most restrictive intervention in the set, the mobile-internet block, was kept in place by a quarter of the people who signed up for it.

The studies measure adherence in different ways — self-report, logged settings, retention in the study — so the comparison is rough, and the pattern is not a neat ranking. But the strongest friction was the hardest to keep, and that should not be surprising. Friction works by making something you habitually want slightly worse to get, which means every moment it is working is a moment some part of you would prefer it were off. The more a setting restricts, the more often it is in the way, and the more attractive the switch that removes it. A two-week trial can measure how well a friction works without having to find out how long anyone would put up with it.

It is also worth being clear about who the trials are run on. One sec's participants had downloaded one sec. Castelo's were motivated enough to sign up for two weeks without mobile internet, and the ones who stayed were among the better-functioning at baseline. Eighty-five per cent of Dekker and Baumgartner's participants already intended to reduce their phone use. Allcott's sample selected itself in through social media advertisements — although there, reweighting the sample to look more like the national population made the estimated self-control problem larger, not smaller. Volunteers who want to cut down are the people most likely to keep a friction switched on. It would be unwise to assume that everyone else would keep it as well.

The people being studied seem to have found their own version of a friction worth keeping. Tran and colleagues interviewed thirty-nine people about compulsive phone checking and found that they periodically reflected on it and deleted the apps that drove the checking without giving back enough to justify it. The authors' recommendation to designers was to build for what users find meaningful rather than to build lock-out mechanisms (Tran, Yang, Davis, & Hiniker, 2019). That is a smaller and more durable kind of friction: not a hurdle in front of the app but the absence of the app.

What happens when it comes off

Assume the friction is kept for as long as the trial asks. What happens after?

Mostly, the behaviour comes back. Castelo's participants went from 161 minutes a day to 265 within two weeks of the block ending — still below the 314 they started at, but most of the way back. Pieh's participants were approaching their starting screen time six weeks after the restriction ended. Okeke's vibration effect did not outlast the vibration.

Mostly is not entirely, and the exceptions are informative. Castelo's gains in well-being and mental health were still partly present two weeks after the block ended, which the authors suggest may be because screen time stayed below baseline. Lyngs's goal-reminder group made fewer Facebook visits after the reminders were removed. And Allcott's study contains the cleanest evidence in this literature that phone use is habit-forming in the technical sense. When participants were paid to cut their use for three weeks, use fell by fifty-six minutes a day, and the reduction partly persisted after the payments stopped — by nineteen minutes a day over the following three weeks and twelve over the three after that (Allcott, Gentzkow, & Song, 2022). Less use today makes for less use tomorrow, with decay.

That is the real basis for optimism here, and also the reason that short trials cannot settle the question. Lally and colleagues followed ninety-six volunteers forming a new daily habit and found that the time it took to reach ninety-five per cent of their eventual automaticity ranged from eighteen days to two hundred and fifty-four (Lally, van Jaarsveld, Potts, & Wardle, 2010). Breaking an established habit is harder to measure, and there is no particular reason to expect it to be quicker. Most of the friction trials reviewed here ran for between one and six weeks.

What the evidence supports

Well supported. Friction reduces measured phone use while it is in place. A randomised trial found that greyscale immediately reduced logged screen time against a control (Zimmermann & Sobolev, 2023), and before-and-after studies put the reduction at roughly twenty to fifty minutes a day (Holte & Ferraro, 2023; Dekker & Baumgartner, 2024). Self-set limits that are hard to override on the spot reduce use of tempting apps by around sixteen per cent over twelve weeks, and most people who set them keep them (Allcott, Gentzkow, & Song, 2022). Blocking mobile internet roughly halves screen time while the block is in place (Castelo et al., 2025). Habits depend on their cues, and disrupting the cue can return behaviour to intention (Wood, Tam, & Witt, 2005; see also Anderson & Wood, 2021).

Reasonably supported. A pause at the moment of opening an app, with an explicit option to back out, is associated with fewer attempts to open it over six weeks, not just shorter sessions — though only in self-selected users with no control group — and in a separate experiment the option to back out did more of the work than the delay (Grüning, Riedel, & Lorenz-Spreen, 2023). A limit or friction applied to one app or one device can displace use to another (Collis & Eggers, 2022; Lyngs et al., 2020), although blocking mobile internet reduced media use across devices (Castelo et al., 2025). People value commitment devices and partly know they need them (Allcott, Gentzkow, & Song, 2022). Most of the reduction disappears within weeks of the friction being removed (Castelo et al., 2025; Pieh et al., 2025; Okeke et al., 2018).

Weakly supported, or mixed. That friction improves well-being. A large preregistered trial of blocking mobile internet found improvements in well-being, mental health and objectively measured attention (Castelo et al., 2025), as did a non-blinded trial of reduced screen time (Pieh et al., 2025). Greyscale, app-specific caps and time limits mostly did not move well-being (Zimmermann & Sobolev, 2023; Collis & Eggers, 2022; Allcott, Gentzkow, & Song, 2022), and the broader detox literature is inconsistent (Radtke, Apel, Schenkel, Keller, & von Lindern, 2022), with methodological weaknesses that may partly explain why (Calvert et al., 2026). The general nudge literature that lends friction its reputation shrinks sharply once publication bias is accounted for (Maier et al., 2022; DellaVigna & Linos, 2022).

Not supported by current evidence. That greyscale reduces how often you check your phone: the one-week study that logged unlocks found no change, and the earlier studies it reviews disagreed with each other (Dekker & Baumgartner, 2024). That switching to a dumbphone improves well-being or attention: this publication could find no controlled trial of an actual device switch, and the published research on dumbphone users consists of interviews with a small, self-selected group (Rothschild & Lindqvist, 2024).

What to do, given the evidence

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

  1. Set limits in advance, and make them hard to dismiss. The best-designed trial in this area found a sixteen per cent reduction in use of the most tempting apps, sustained over twelve weeks, from self-set limits that most participants could only extend after an enforced delay (Allcott, Gentzkow, & Song, 2022). The limits built into phones are easy to wave past by default; adding a passcode or a waiting period brings them closer to what was tested.

  2. Put the friction at the reach, and make it a question. A pause with a clear option to back out was associated with fewer attempts to open target apps over six weeks, in a self-selected sample, and in a separate experiment the option to back out did more work than the delay (Grüning, Riedel, & Lorenz-Spreen, 2023). Switching off Face ID so that each unlock needs a passcode is a cruder version of the same idea. Most people in one study kept it (Olson et al., 2023), though its effect has not been tested on its own.

  3. Use greyscale knowing what it does. It shortens sessions — by twenty to fifty minutes a day in the studies that report a figure — and in the one study that logged unlocks, it left the number of pick-ups untouched (Holte & Ferraro, 2023; Dekker & Baumgartner, 2024). That is a real and modest effect. It is also irritating, and adherence varies widely.

  4. Expect the time to move, and decide where. A limit on three social apps sent students to their messaging apps (Collis & Eggers, 2022); a laptop-only intervention, by participants' own account, moved some use onto their phones (Lyngs et al., 2020). Much of the benefit of blocking mobile internet ran through what people did instead — seeing others, exercising, going outside (Castelo et al., 2025). Friction removes an option. It does not supply one.

  5. Delete what you would not miss. The strategy people describe using on their own is periodic reflection followed by removing the apps that drive checking without giving much back (Tran et al., 2019). An app that is not installed cannot be reached for, and there is no setting to switch back on.

  6. Treat a total block as an experiment, not a lifestyle. Two weeks without mobile internet produced the largest benefits in this literature, and only a quarter of the people who signed up kept the block in place (Castelo et al., 2025). Doing it once and paying attention to what changes is a reasonable use of that evidence. Expecting to sustain it is not supported by any of it.

  7. Plan for the friction to come off. Most of the effect leaves with it. What remains is whatever habit changed while it was on: in the one study built to measure that, people paid to cut down kept about a third of the reduction three weeks after the payments stopped, and about a fifth three weeks after that (Allcott, Gentzkow, & Song, 2022).

A note on certainty

This is the most encouraging of the recent essays here, and it should be read with that in mind. Friction is the rare piece of popular attention advice that reliably changes measured behaviour. Greyscale works on screen time. Limits work. A blocked mobile internet works, in the narrow and important sense that people use their phones much less while it is in place, and in the one large trial that looked, they felt and concentrated better too.

What it does not do is change the person. The checking habit that the September essay identified as the main event was untouched by greyscale in the one study that counted it, and the one tool associated with fewer attempts to open apps did so by putting a question in their path. Friction works on the session and, at best, on the moment of reaching; it does not appear to reach whatever produces the reach. And its strongest versions are fragile in a predictable way: the more a friction restricts, the more often it is in the way, and the more tempting it is to switch it off.

The research is also young, short and mostly run on the motivated. Many of the trials here lasted a week or two, most samples were students or volunteers who wanted to cut down, and the well-being outcomes are largely self-reported by people who knew what the study was about. A systematic review of digital self-control tools called for longer studies and standardised measures (Monge Roffarello & De Russis, 2023), and a recent scoping review of forty-five social media reduction studies found methodological limitations pervasive enough that mixed results may partly reflect weak designs rather than real differences in effect (Calvert et al., 2026). The strongest single result in this essay has one trial behind it.

What survives is useful, and smaller than the advice. Making the phone a little worse to use does make people use it a little less, for as long as it stays worse. The practical skill is choosing a friction mild enough to keep, aimed at the moment of reaching rather than the length of the session — and deciding in advance what the time is for. Friction works. Then you turn it off.


References

  • Allcott, H., Gentzkow, M., & Song, L. (2022). Digital addiction. American Economic Review, 112(7), 2424–2463.
  • Anderson, I. A., & Wood, W. (2021). Habits and the electronic herd: The psychology behind social media's successes and failures. Consumer Psychology Review, 4(1), 83–99.
  • Calvert, E., Vaidya, S., Ilagan, G., Swords, C. M., Goldberg, S. B., Ferguson, C., Linardon, J., & Torous, J. (2026). Mapping the methodological landscape of social media detox intervention through a scoping review of evidence relevant to social media restriction policies. Communications Psychology. Advance online publication.
  • Castelo, N., Kushlev, K., Ward, A. F., Esterman, M., & Reiner, P. B. (2025). Blocking mobile internet on smartphones improves sustained attention, mental health, and subjective well-being. PNAS Nexus, 4(2), pgaf017.
  • Collis, A., & Eggers, F. (2022). Effects of restricting social media usage on wellbeing and performance: A randomized control trial among students. PLOS ONE, 17(8), e0272416.
  • Dekker, C. A., & Baumgartner, S. E. (2024). Is life brighter when your phone is not? The efficacy of a grayscale smartphone intervention addressing digital well-being. Mobile Media & Communication, 12(3), 688–708.
  • Dekker, C. A., Baumgartner, S. E., Sumter, S. R., & Ohme, J. (2025). Beyond the buzz: Investigating the effects of a notification-disabling intervention on smartphone behavior and digital well-being. Media Psychology, 28(1), 162–188.
  • DellaVigna, S., & Linos, E. (2022). RCTs to scale: Comprehensive evidence from two nudge units. Econometrica, 90(1), 81–116.
  • Grüning, D. J., Riedel, F., & Lorenz-Spreen, P. (2023). Directing smartphone use through the self-nudge app one sec. Proceedings of the National Academy of Sciences, 120(8), e2213114120.
  • Heitmayer, M. (2025). When the phone's away, people use their computer to play: Distance to the smartphone reduces device usage but not overall distraction and task fragmentation during work. Frontiers in Computer Science, 7, 1422244.
  • Heitmayer, M., & Lahlou, S. (2021). Why are smartphones disruptive? An empirical study of smartphone use in real-life contexts. Computers in Human Behavior, 116, 106637.
  • Hiniker, A., Hong, S. R., Kohno, T., & Kientz, J. A. (2016). MyTime: Designing and evaluating an intervention for smartphone non-use. Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems, 4746–4757.
  • Holte, A. J., & Ferraro, F. R. (2023). True colors: Grayscale setting reduces screen time in college students. The Social Science Journal, 60(2), 274–290.
  • Holte, A. J., Giesen, D. T., & Ferraro, F. R. (2023). Color me calm: Grayscale phone setting reduces anxiety and problematic smartphone use. Current Psychology, 42(8), 6778–6790.
  • Lally, P., van Jaarsveld, C. H. M., Potts, H. W. W., & Wardle, J. (2010). How are habits formed: Modelling habit formation in the real world. European Journal of Social Psychology, 40(6), 998–1009.
  • Lukoff, K., Lyngs, U., Zade, H., Liao, J. V., Choi, J., Fan, K., Munson, S. A., & Hiniker, A. (2021). How the design of YouTube influences user sense of agency. Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems, 1–17.
  • Lyngs, U., Lukoff, K., Slovak, P., Seymour, W., Webb, H., Jirotka, M., Zhao, J., Van Kleek, M., & Shadbolt, N. (2020). 'I just want to hack myself to not get distracted': Evaluating design interventions for self-control on Facebook. Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems, 1–15.
  • Maier, M., Bartoš, F., Stanley, T. D., Shanks, D. R., Harris, A. J. L., & Wagenmakers, E.-J. (2022). No evidence for nudging after adjusting for publication bias. Proceedings of the National Academy of Sciences, 119(31), e2200300119.
  • Mertens, L., Brockmeier, L. C., Roitzheim, C., Radtke, T., Dingler, T., & Keller, J. (2026). Promoting self-regulated social media use on smartphones with a mobile intervention app (Wellspent): Randomized controlled trial. JMIR mHealth and uHealth, 14, e56824.
  • Mertens, S., Herberz, M., Hahnel, U. J. J., & Brosch, T. (2022). The effectiveness of nudging: A meta-analysis of choice architecture interventions across behavioral domains. Proceedings of the National Academy of Sciences, 119(1), e2107346118.
  • Monge Roffarello, A., & De Russis, L. (2023). Achieving digital wellbeing through digital self-control tools: A systematic review and meta-analysis. ACM Transactions on Computer-Human Interaction, 30(4), 1–66.
  • Ochs, C., & Sauer, J. (2022). Curtailing smartphone use: A field experiment evaluating two interventions. Behaviour & Information Technology, 41(16), 3598–3616.
  • Okeke, F., Sobolev, M., Dell, N., & Estrin, D. (2018). Good vibrations: Can a digital nudge reduce digital overload? Proceedings of the 20th International Conference on Human-Computer Interaction with Mobile Devices and Services, 1–12.
  • Olson, J. A., Sandra, D. A., Chmoulevitch, D., Raz, A., & Veissière, S. P. L. (2023). A nudge-based intervention to reduce problematic smartphone use: Randomised controlled trial. International Journal of Mental Health and Addiction, 21(6), 3842–3864.
  • Pieh, C., Humer, E., Hoenigl, A., Schwab, J., Mayerhofer, D., Dale, R., & Haider, K. (2025). Smartphone screen time reduction improves mental health: A randomized controlled trial. BMC Medicine, 23, 107.
  • Radtke, T., Apel, T., Schenkel, K., Keller, J., & von Lindern, E. (2022). Digital detox: An effective solution in the smartphone era? A systematic literature review. Mobile Media & Communication, 10(2), 190–215.
  • Rothschild, A., & Lindqvist, J. (2024). Swapping 5G for 3G: Motivations, experiences, and implications of contemporary dumbphone adoption. Proceedings of the ACM on Human-Computer Interaction, 8(CSCW1), 1–40.
  • Tran, J. A., Yang, K. S., Davis, K., & Hiniker, A. (2019). Modeling the engagement-disengagement cycle of compulsive phone use. Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems, 1–14.
  • Wood, W., & Rünger, D. (2016). Psychology of habit. Annual Review of Psychology, 67, 289–314.
  • Wood, W., Tam, L., & Witt, M. G. (2005). Changing circumstances, disrupting habits. Journal of Personality and Social Psychology, 88(6), 918–933.
  • Zimmermann, L., & Sobolev, M. (2023). Digital strategies for screen time reduction: A randomized field experiment. Cyberpsychology, Behavior, and Social Networking, 26(1), 42–49.

Pass it on

The quickest way to tell people to put their phones down is, regrettably, on their phones.

So that is where the reminders go. Follow on Instagram, send one to someone who needs it, then put yours down.

Follow @terminallyofflineclub