About 66 days, if you want one number. But 66 is a median from a single study of 39 people, the same study found anywhere from 18 to 254 days, and it never claimed a day you cross and are done.
In 2009 a team at University College London asked 96 volunteers to pick one simple daily thing — a glass of water with lunch, a piece of fruit, a short run — do it in the same situation each day, and report every evening how automatic it felt. The study ran 84 days.
| 96 | volunteers, mostly postgraduate students, average age 27 |
| 84 days | of daily reporting — the whole study |
| 82 | gave enough data to analyse |
| 39 | fit the curve well enough to estimate a plateau. 66 is the median of these 39. |
| 18–254 | days across them — and these are model estimates, not observed finish dates |
That last line is the one everybody drops. The study stopped at day 84, so every estimate past it — including the 254 — is the curve extended beyond what anyone actually watched. And the people whose data did not fit the model are not in the 66 at all.
The researchers were careful about what they measured, too: self-reported automaticity, how little the thing took deciding — not “habit” as a finished state.
Because of a 1960 book by a plastic surgeon, Maxwell Maltz, who noticed his patients took about three weeks to stop being startled by their own new face. It was an observation about adjusting to a change, in a book about self-image. No one ran a study. It became a rule anyway, and it has outlived every attempt to correct it.
In the data, missing a single day did not measurably set people back. What did matter was being erratic: people who performed the behaviour inconsistently, week after week, did not get to automatic at all.
So the honest version is not “never break the chain”. It is: one missed day is noise, and a month of missed days is the signal. What a missed day actually costs.
A streak counter does the opposite of what the research suggests. It prices one missed day at everything you have — back to zero — and prices the twentieth missed day at nothing, because there is nothing left to lose. The day the streak breaks is the day people delete the app.
It also puts the number in the wrong place. 66 is roughly when something stops taking a decision, for about half the people it was measured on. It is not a finish line, and nothing bad happens if you take 120.
RIPEN is a habit tracker that draws your habits as one tree, and it is built on 66 — but on leaves, not on a streak.
That is the research turned into a shape: one miss is cheap, a long drift is expensive, and 66 is a thing that appears on the tree rather than a counter resetting to zero. You can be at 64 leaves, miss a Tuesday, and still be at 63 on Wednesday.
There is no account and no analytics, and Pro is bought once rather than every month. More about how the tree works, or get the app.
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 — and UCL’s write-up of it. On how the 66 gets overstated, see this breakdown of what the study actually found. The 21 days trace to Maxwell Maltz, Psycho-Cybernetics (1960).