To build habits, the technique with the strongest evidence is the if-then plan (an "implementation intention"), with a medium-to-large effect of d = 0.65 across 94 tests (Gollwitzer and Sheeran, 2006). Changing your environment and tracking progress come next (d = 0.55 and 0.40 in large meta-analyses), followed by small rewards for consistency, temptation bundling, commitment devices and habit stacking, which have field-experiment or single-study support.
- If-then plans rank first: d = 0.65 across 94 tests.
- Environment design (d = 0.55) and progress tracking (d = 0.40) follow, both from large meta-analyses.
- Reward the comeback: a bonus for returning after a missed workout was the best of 54 gym programmes (+27%).
- 9.1x: how over-optimistic expert forecasts of those programmes were.
- 66 days, not 21, is the median time for a daily habit to plateau.
The ranking criterion is the quality and size of the evidence. Techniques with a meta-analysis are ranked by its pooled standardised effect size (Cohen's d or Hedges' g, where 0.2 is small, 0.5 medium and 0.8 large). Techniques without one come after, ordered by the size of the best randomised field experiment.
One number to hold in mind before you start: in Phillippa Lally's 2010 study at University College London, people forming a new daily habit took a median of 66 days to reach a plateau of automaticity, with modelled estimates from 18 to 254 days.
A 2024 systematic review of 20 studies by Singh and colleagues put typical medians at 59 to 66 days. Twenty-one days is a myth.
Which habit techniques have the strongest evidence?
If-then plans, environment design and progress monitoring are the three techniques backed by meta-analyses of dozens to hundreds of randomised studies. The others rest on field experiments, and one of those started as a researcher's private trick.
Katherine Milkman, a behavioural scientist at Wharton, had trouble getting herself to the gym. She also loved escapist audiobooks, the kind of page-turner she would never count as work. So she made a rule: she could listen to them, The Hunger Games among them, only while exercising. By the account Knowledge at Wharton published in 2013, she started going five days a week.
Years later she gave the trick a name, temptation bundling, and tested it properly. Over nine weeks, 226 students and staff at a university gym who wanted to exercise more were split into groups; one got iPods loaded with tempting novels that they could use only at the gym. That group went far more often at first, the effect faded after the Thanksgiving break, and afterwards 61% of participants opted to pay for gym-only access to iPods holding tempting audiobooks. A private habit became rank 5 on this list.
Knowledge at Wharton: How researchers used The Hunger Games to encourage healthier choices (Nov 2013)The chart compares the pooled effects of the three meta-analysed techniques.
See the data as a table
| Pooled effect size (d) | |
|---|---|
| If-then plans (implementation intentions) | 0.65 |
| Defaults (choice architecture) | 0.62 |
| Changing decision structure (environment design) | 0.55 |
| Choice architecture, all types | 0.45 |
| Monitoring goal progress | 0.4 |
Source: Gollwitzer and Sheeran 2006; Mertens et al., PNAS 2022; Harkin et al., Psychological Bulletin 2016 · Choice-architecture estimates shrink after publication-bias correction (to about d = 0.31 under a moderate correction). Effects come from different meta-analyses and outcome types.
| Rank | Technique | Best evidence | Effect | Evidence type |
|---|---|---|---|---|
| 1 | If-then plans | Gollwitzer and Sheeran 2006 | d = 0.65 (94 tests) | Meta-analysis |
| 2 | Environment design | Mertens et al. 2022 | d = 0.55 for structure changes (200+ studies) | Meta-analysis |
| 3 | Progress tracking and accountability | Harkin et al. 2016 | d = 0.40 (138 studies, 19,951 people) | Meta-analysis |
| 4 | Rewards for consistency | Milkman et al. 2021 | +27% gym visits, top arm | Megastudy, 61,293 people |
| 5 | Temptation bundling | Milkman, Minson and Volpp 2014 | +51% gym visits at first, then decayed | Field experiment |
| 6 | Commitment devices | Giné, Karlan and Zinman 2010 | +3 percentage points quitting smoking | Field experiment, Philippines |
| 7 | Habit stacking | Judah, Gardner and Aunger 2013 | Stronger habit when flossing followed brushing | Exploratory study, 50 people |
1. If-then plans (implementation intentions)
An if-then plan links a specific situation to a specific action, in the form "If it is 6:30 am and I have finished my tea, then I will put on my running shoes." Peter Gollwitzer and Paschal Sheeran's 2006 meta-analysis of 94 independent tests found a medium-to-large effect on goal attainment, d = 0.65.
The plans helped at four points:
- getting started;
- staying on track when distracted;
- letting go of a failing approach;
- saving self-control for later.
The mechanism is that the "if" part makes the cue stand out, and the "then" part makes the response close to automatic when the cue appears. That is the same cue-response link that a habit is, built deliberately instead of by slow repetition.
A goal intention ("I want to exercise more") leaves the when and where open; an if-then plan closes them.
- Effect: d = 0.65 across 94 tests
- Cost: two minutes of writing
- Write it down: "If [time/place/after X], then I will [first action]"
2. Environment design
Environment design means changing what is easy, visible or default around you, rather than relying on willpower. The 2022 PNAS meta-analysis by Stephanie Mertens and colleagues, covering more than 200 studies, 450 effect sizes and about 2.15 million people, found that changing the structure of a choice (defaults, effort, ordering, physical layout) worked better (d = 0.55) than giving information (d = 0.38) or reminders and commitments (d = 0.31). Defaults alone came out at d = 0.62.
The caution matters: the authors found publication bias, and a moderate correction cut their overall estimate from 0.42 to 0.31; a severe correction took it near zero. Food choices responded most.
In practice:
- keep the phone charging in another room during study hours;
- put the guitar on a stand instead of in its case;
- make the unwanted habit one or two steps harder.
These are the same levers I use in the deep work focus routine.
- Effect: d = 0.55 for structural changes (0.31 after moderate bias correction, all types)
- Scale: 200+ studies, about 2.15 million participants
- Strongest domain: food choices

3. Progress tracking and accountability
Tracking progress toward a goal is an effective behaviour-change technique in its own right: Benjamin Harkin and colleagues' 2016 meta-analysis of 138 randomised studies with 19,951 participants found that interventions which increased monitoring improved goal attainment by d = 0.40. The effect was larger when progress was reported to others or made public, and when it was physically recorded.
That gives accountability and the humble habit tracker the same evidence base. A paper calendar with crosses, a shared spreadsheet with a friend, or a weekly message to a study partner all qualify. What matters is that you record the behaviour, not just think about it, and that someone else can see it.
- Effect: d = 0.40 (138 studies, 19,951 people)
- Bigger when: recorded physically, reported publicly
- Simplest version: one tick per day on a wall calendar
4. Rewards for consistency, including for coming back
Small, immediate rewards can raise a behaviour while they last, and the best-tested example is Katherine Milkman and Angela Duckworth's 2021 megastudy in Nature. Thirty scientists designed 54 four-week programmes for 61,293 members of an American gym chain, all tested on the same outcome: weekly visits.
A baseline of planning prompts, reminders and small point rewards raised visits by 9% over a placebo group that averaged 1.48 visits a week. The top arm added a bonus of 125 points (about US$0.09) for returning after a missed workout and raised visits by 27%. That detail matters for anyone obsessed with streaks: rewarding the comeback beat rewarding the unbroken chain, and Lally's data also showed a single missed day did not derail habit formation.
See the data as a table
| Increase in visits (%) | |
|---|---|
| Bonus for returning after a missed workout (125 pts) | 27 |
| Larger reward per visit (490 pts) | 25 |
| Told most Americans exercise, share rising | 24 |
| Bonus for returning after a missed workout (225 pts) | 23 |
| Choice of gain or loss framing | 19 |
| Planning, reminders and micro-rewards (baseline) | 9 |
Source: Milkman et al., Megastudies improve the impact of applied behavioural science, Nature 2021 · Placebo group averaged 1.48 visits a week. The authors warn that top-performer estimates are probably inflated by selection.
Two sobering results from the same study:
- only 8% of the 54 programmes still had a significant effect in the four weeks after they ended;
- when 301 lay people, 156 public-health professors and 90 behavioural-science practitioners were asked to predict the results, their forecasts were uncorrelated with reality and on average 9.1 times too optimistic (Milkman et al., Nature 2021).
Your favourite productivity influencer is not better at this than those professors.On the 9.1x forecasting error in Milkman et al. 2021
- Effect: +9% to +27% visits during the programme
- After rewards stop: only 8% of programmes had lasting effects
- Best design: reward the return after a lapse
5. Temptation bundling
Temptation bundling pairs something you want with something you should do, and allows the treat only during the chore. In Milkman, Julia Minson and Kevin Volpp's 2014 field experiment, participants who could listen to gripping audiobooks only at the gym visited 51% more often than controls at first; those merely encouraged to restrict their listening visited 29% more than controls.
The effect faded over the weeks, especially after the Thanksgiving holiday break. Still, 61% of participants opted to pay for gym-only access to iPods holding tempting audiobooks, which suggests people value the device. A local version: allow yourself the next episode of a series only on the exercise cycle, or the good filter coffee only at the study desk.
- Effect: +51% gym visits at first, fading over time
- Demand: 61% paid to keep the bundle afterwards
- Weakness: novelty wears off
6. Commitment devices
A commitment device puts something at stake that you lose if you fail. The best-known test is CARES, a savings account for smokers in the Philippines run by Xavier Giné, Dean Karlan and Jonathan Zinman: people deposited money for six months, then took a urine test, and forfeited the balance to charity if they failed.
Only 11% of smokers offered the account took it, and being offered it raised the chance of passing the six-month test by 3 percentage points over controls, an effect that held in surprise tests at 12 months. That is modest at the population level but large per user, because so few signed up. Commitment devices work for the people who want them and need a reason to stick.
- Take-up: 11% of those offered
- Effect: +3 percentage points quitting at six and twelve months
- Best for: one behaviour you have repeatedly failed to change
7. Habit stacking
Habit stacking means attaching a new behaviour to the end of an existing routine, and it has the thinnest direct evidence on this list. The relevant study is Gaby Judah, Benjamin Gardner and Robert Aunger's 2013 exploratory study of 50 adults learning to floss: half were told to floss before brushing, half after.
Those who flossed after brushing tended to form stronger habits and, at eight months, flossed more often. The authors called for more rigorous designs. Habit stacking is really a special case of the if-then plan at rank 1, with an existing routine as the "if", which is why it is worth using even though its own evidence is small.
- Evidence: one exploratory study, 50 people
- Finding: placing the new behaviour after an existing one built stronger habits
- Best use: as the cue in an if-then plan
Fifty-four four-week digital programmes designed by 30 scientists from 15 US universities were tested on 61,293 members of a US gym chain, with weekly gym visits as the single outcome. 45% of programmes raised visits by 9% to 27% against placebo; the best paid a small bonus for returning after a missed workout. Only 8% of programmes had significant effects in the four weeks after ending, and expert forecasts were uncorrelated with results. Limit: one behaviour, one country, paid gym members who had already chosen to join a programme.
Read the paper (free)A meta-analysis of 94 independent tests of if-then plans across health, academic, environmental and interpersonal goals. The pooled effect on goal attainment was d = 0.65. Plans helped with starting, staying on course, disengaging from failing means and conserving self-control, and worked by making the cue more accessible and the response more automatic. Limit: many included studies were short laboratory or self-report studies; long-term maintenance was less tested.
Read the paper (free)138 randomised studies with 19,951 participants compared interventions that prompted progress monitoring with controls. Interventions raised monitoring frequency (d = 1.98) and goal attainment (d = 0.40), and the rise in monitoring explained the gain. Effects were larger when outcomes were reported to others or made public and when information was physically recorded. Limit: wide range of goals and populations, so the pooled number hides large variation.
Read the paperMore than 200 studies, 450 effect sizes and about 2.15 million participants. Changing decision structure (d = 0.55, defaults 0.62) beat decision information (0.38) and decision assistance such as reminders (0.31). Food choices showed the largest effects. Limit: the authors detected publication bias; a moderate correction lowered the overall estimate from 0.42 to 0.31, and a severe one to 0.03.
Read the paper (free)How long does it take to build a habit?
About two months on average, with very wide variation. Lally's study followed 96 volunteers for 84 days as they repeated a chosen daily behaviour; among those whose data fitted the model well, the median time to reach 95% of their automaticity plateau was 66 days, and model estimates ranged from 18 to 254 days.
Singh and colleagues' 2024 review of 20 studies with 2,601 participants reported median times of 59 to 66 days, with individual times ranging from 4 to 335 days, and simple behaviours forming faster than complex ones. Plan for a two-to-three-month runway and judge your habit at day 60, not day 21.
What should you combine, and in what order?
Stack the evidence: an if-then plan for the cue, an environment that makes the action easy, a visible tracker, and a small reward for returning after a miss.
- Write one if-then plan. Use an existing routine as the cue. Keep the first action under one minute.
- Remove one obstacle and add one. Make the new habit one step easier and the competing one a step harder.
- Track it where you can see it. One mark per day, shared with one person.
- Reward the comeback. After a missed day, give yourself something small for restarting, not for the streak.
- Review at day 60. If it still feels effortful, simplify the action rather than adding willpower.
What's new
- Review of 20 studies finds healthy habits take about two months to form, not 21 days News-Medical, Jan 2025
- How to make a habit stick: means of 106 days for a morning stretch and 154 for an evening one in one study BPS Research Digest, Mar 2025
- Think it takes 21 days to form a habit? A University of South Australia review says about two months Gizmodo, Jan 2025
Questions people ask
Does it really take 21 days to form a habit?
No. No habit study supports 21 days. Lally's 2010 study found a median of 66 days, and a 2024 review of 20 studies found medians of 59 to 66 days.
Does missing one day break a habit?
In Lally's data, missing a single opportunity did not significantly affect habit formation; being highly inconsistent did. Milkman's megastudy found that rewarding a return after a missed workout was the most effective programme of 54.
Is a dopamine detox a good way to reset habits?
None of the meta-analyses here tested a "dopamine detox". The techniques with evidence change cues, environments and rewards in a targeted way. A weekend of abstinence may reduce exposure to cues briefly, but without an if-then plan and a changed environment the old cues return on Monday.
Why do I keep failing even with an app streak?
Streaks reward continuity, so a single break can feel like total failure and lead people to quit. Track progress, but reward restarting. Harkin's meta-analysis found tracking works best when it is recorded and shared, which a private app streak often is not.
Sources
- Milkman K.L. et al.: Megastudies improve the impact of applied behavioural science, Nature 600:478-483 (2021)
- Gollwitzer P.M. and Sheeran P.: Implementation intentions and goal achievement: A meta-analysis of effects and processes, Advances in Experimental Social Psychology 38 (2006)
- Harkin B. et al.: Does monitoring goal progress promote goal attainment? Psychological Bulletin 142(2):198-229 (2016)
- Mertens S. et al.: The effectiveness of nudging: A meta-analysis of choice architecture interventions, PNAS 119(1) (2022)
- Milkman K.L., Minson J.A. and Volpp K.G.M.: Holding the Hunger Games Hostage at the Gym: An Evaluation of Temptation Bundling, Management Science 60(2):283-299 (2014)
- Giné X., Karlan D. and Zinman J.: Put Your Money Where Your Butt Is: A Commitment Contract for Smoking Cessation, AEJ: Applied Economics 2(4):213-235 (2010)
- Judah G., Gardner B. and Aunger R.: Forming a flossing habit, British Journal of Health Psychology 18(2):338-353 (2013)
- Lally P. et al.: How are habits formed: Modelling habit formation in the real world, European Journal of Social Psychology 40(6) (2010)
- Knowledge at Wharton: How researchers used The Hunger Games to encourage healthier choices (2013)
- News-Medical: Healthy habits take longer to form than expected (on Singh et al., Healthcare, 2024) (2025)
- The Behavioral Scientist: What Lally et al. (2010) actually found
