A New Habit Takes About Two Months. Some Take Longer.
By Martin Bergmann | AI Project Lab | November 2026
It is week three. The habit you started has stopped feeling new and has not yet become automatic. Do you keep going?
The last two posts each ended with a habit to try: make the AI quiz you, and write down what you decide and what you hand over before you open it. If you tried one the week I posted it, you are somewhere around week two or three now.
What a study of 96 people found.
In a study published in 2010, Phillippa Lally and colleagues at University College London asked 96 volunteers to pick one new healthy behavior, such as eating fruit with lunch, and to do it every day after a daily cue. For 12 weeks, the volunteers logged how automatic it felt.
Among the volunteers who built a habit, the typical time was about two months. Some got there in under three weeks. For others, the estimate was more than eight months, a projection past the end of the study. About half of the volunteers did not do the behavior consistently enough to build a habit at all. And one missed day barely mattered.
A 2024 review of health-habit studies reached a similar conclusion: habits can start forming within about two months, but the timing varies widely between people. Only a few of those studies measured the time directly, so the evidence is still thin.
Progress was fastest at the start and slowed after that. My reading, not the study's: that is why week three can feel like nothing is happening.
What this means for an AI habit.
The study covered eating, drinking, and exercise. It says nothing about work habits or AI. My inference is that a new AI habit needs a long runway too, so the useful question at week three is not "is it working?" but "have I done it enough days?"
That changes some of my own advice. My AI Adoption Playbook says to commit to a few actions for 30 days, then retake the Radar. After this study, I would call 30 days the first check-in, not the finish line.
How to form one.
This is my suggestion, not the study's, and it is untested.
Pick one habit from the last two posts or anything you want to develop
Give it a cue. Use something you already do, like "after I open my laptop, I write two lines about the day's biggest task: what I decide and what I hand over." Or put a task on your calendar. Rocío Fernández writes that routines are held together by context, not willpower. Everyone in the study tied their behavior to a daily cue, though the study did not test whether the cue helped. The "after X, I do Y" form is the if-then plan from my post on closing loops.
Keep it small. Jeremy Utley tells people learning AI to pick one low-stakes task, do it every day even for 10 minutes, and count reps, not results.
Set a review date about three months out, not at week three.
Miss a day? Do it the next day. Do not start over.
Week three is not the verdict. Keep going until your review date.
Martin Bergmann is the Director of an IT PMO and founder of the AI Project Lab. He writes about AI, project management, and the future of intelligent work.
Comments