Where 66 days came from
The number traces back to a study by Phillippa Lally and colleagues, published in 2010.
Ninety-six volunteers chose one eating, drinking or activity behaviour they wanted to repeat daily in response to the same kind of cue—for example, eating fruit with lunch or running before dinner. They reported the behaviour and completed self-report automaticity items during a 12-week study.
The researchers were not looking for a day when everybody suddenly became habitual. They modelled each person’s automaticity as an asymptotic curve: growth rises more quickly at first, then increases by smaller amounts as it approaches a personal plateau. They defined the modelled formation time as the point at which the curve reached 95% of that plateau.
The median modelled time among the 39 participants whose curves met the researchers’ “good fit” criteria was 66 days. Individual estimates in that group ranged from 18 to 254 days.
Those details change the meaning of the headline. Sixty-six was a summary of varied individual curves under particular study conditions. It was not an assigned challenge length, a minimum dose, or a universal biological timetable.
What the study did find
Three findings are more useful than the slogan.
Automaticity tended to grow gradually
For many participants, the curved model fit better than a straight line. Early repetitions tended to contribute larger gains, while later gains became smaller. Habit formation looked more like approaching a plateau than filling 66 identical boxes.
People and behaviours varied substantially
The 18-to-254-day range was not a minor footnote. It was a central result. Participants also reached different maximum levels of automaticity. The researchers observed that the exercise group took longer to approach its plateau than the eating and drinking groups, but the study was not powered to make firm subgroup conclusions. Lally et al. (2010).
One omission was not catastrophic
In the opportunities the researchers could analyse, missing a single planned occasion did not materially alter the longer-term formation process. That does not show that repeated non-performance has no effect; repetition was essential to the model. It does challenge the idea that a single empty square sends a person back to day one.
What it did not find
The paper did not show that:
- a behaviour automatically becomes a habit on day 66;
- 66 consecutive completions are required;
- every person can form every habit with the same schedule;
- reaching 95% of a modelled plateau is a universal definition of success;
- the behaviour will transfer intact to a new context;
- complex routines follow the same curve as a simple cue-linked action;
- automaticity guarantees that the action always occurs.
There were important limits. Although 96 people enrolled, 82 supplied sufficient data for analysis. The asymptotic model fit 62 participants and fit well for 39. The study relied on self-reported automaticity and performance, ran for 12 weeks, and focused on once-daily behaviours cued by a daily event. The authors themselves called for more work on complexity, consistency and objective measures. Lally et al. (2010).
This does not make the research useless. It makes it research: informative within a stated design, not a law detached from it.
What the 2024 review adds
A 2024 systematic review and meta-analysis examined 20 habit-formation intervention studies involving 2,601 adults and several health-related behaviours. Across the four studies that directly reported time to formation, reported medians were 59 to 66 days, reported means were 106 to 154 days, and individual estimates ranged from 4 to 335 days. Singh et al. (2024).
That wider range reinforces the main lesson: there is no single reliable deadline.
It also needs its own caveats. Only four of the 20 studies directly reported formation time. The studies differed in design, behaviour, intervention, measurement and follow-up. Eleven were rated at high risk of bias, and many had small samples. The review could not meta-analyse time-to-automaticity because the duration evidence was too limited and heterogeneous. Singh et al. (2024).
So “the newer science says 59 days” would repeat the same mistake with a different number. The review offers a better estimate of uncertainty, not a replacement countdown.
A better use for day 66
If a calendar marker helps, use day 66 as a review—not a verdict.
Ask five questions:
- What is the actual cue? Name the event or setting that precedes the action. “Evening” is broad; “after I plug in my phone” is observable.
- Does the cue bring the action to mind? Notice whether you remember before an alert, list or streak prompts you.
- Is beginning becoming more familiar? Look for less decision-making, not zero effort.
- Where does the pattern fail? If misses cluster on travel days or late meetings, the context—not your entire character—may be the useful unit of analysis.
- Is the action still worth supporting? Automaticity is not a reason to keep an irrelevant behaviour.
Then choose one next move:
- Continue if repetition is useful and automaticity is still growing.
- Adjust if the cue appears but the version is too large.
- Add an off-routine plan if the usual setting is doing most of the work.
- Taper one support if the cue is reliably prompting the start.
- Pause or end if the behaviour no longer fits the reason you chose it.
No option requires resetting a counter to manufacture a cleaner story.
What the calendar can—and cannot—tell you
The calendar can show exposure to opportunities. It can reveal that Tuesdays repeatedly fail, that the cue rarely occurs, or that a smaller version makes busy days more workable. It can help you choose when to review support.
It cannot directly observe the mental association between cue and action. It cannot tell you whether a completion was effortless, reluctantly forced, or performed only to protect a streak. And it cannot decide whether the habit still serves you.
Keep the number in its proper role: evidence about time can set patient expectations, but your next decision still comes from the pattern in front of you.
Evidence notes and sources
- 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, 998–1009. The often-quoted 66 days was a median modelled time to 95% of a personal automaticity plateau among the 39 “good fit” curves; individual estimates and curve fit varied widely.
- Singh, B., Murphy, A., Maher, C., & Smith, A. E. (2024). Time to Form a Habit: A Systematic Review and Meta-Analysis of Health Behaviour Habit Formation and Its Determinants. Healthcare, 12, 2488. Twenty studies were included, but only four directly reported formation time; study heterogeneity and risk of bias limit any universal duration claim.
- Keller, J. et al. (2021). Habit formation following routine-based versus time-based cue planning: A randomized controlled trial. British Journal of Health Psychology, 26, 807–824. Among participants meeting the study’s modelled success criterion, the median estimate was 59 days and the range was 4–335; only 23% of the analysed sample met that criterion.