The gap that is not closing
Three years of data from a US survey that keeps repeating:
| Daily | Several times a week | A few times a year or more often | |
|---|---|---|---|
| May 2023 | 4% | 11% | 21% |
| May 2025 | 8% | 19% | 40% |
| May 2026 | 15% | 30% | 52% |
The ratio between "I have tried it" and "I use it daily" was more than five to one in 2023 and three and a half to one in 2026. It is improving — but slowly. It is still true that barely one in three people who try it gets as far as daily use.
Another survey, of 5,119 people, gives higher numbers: 49% have used a chatbot, 24% daily — the two differ because the survey above asks employees about use at work, while this one asks adults about any use at all. And the spread across sectors is wide: daily use runs at 42% in tech, 27% in finance and 22% in professional services.
What we call "regular use"
This is the most sobering number in the article, and it is worth reading twice.
A study of nearly 6,000 senior executives found that more than two thirds of leaders use AI regularly — and that this use averages an hour and a half a week.
That is not a routine. That is an occasional look-in.
The same study adds that nine out of ten respondents perceived no effect whatsoever on employment or productivity over the past three years.
So the goal is not to "use it more", but to spread that hour and a half across the week so that it repeats — because what repeats can be measured, and what you measure tells you whether it is worth it.
The drop-off
The best data on this comes from a randomised corporate experiment: more than 6,000 employees, 56 companies, six months, with access handed out at random.
Roughly 40% of the people given access used it regularly over the six months.
Which means sixty per cent did not — even though they got the paid tool for free and knew the experiment was being watched.
(On long-term, individual-level drop-off — what happens after one month or three — I found no peer-reviewed research. This is the closest data there is.)
What separates a successful rollout from a failed one
A survey of 11,749 people across 14 markets gives the best answer here, and its most important finding is not about the tool.
Strategic clarity is worth more than tool access. The survey's own finding is that people with a clear strategy but a limited toolset do better than people with good tools and no direction — the table below only gives the two extreme cases for this, not the mixed ones.
| Reports a measurable effect | |
|---|---|
| Clear direction + good tool access | 83% |
| Weak direction + weak access | 55% |
Two more numbers from the same place, which together explain a lot:
- Only 36% feel they have had enough training — while 88% say they will need significant reskilling within five years.
- 66% get little or no guidance on what to do with the time they save, and more than half of them do not put it into more valuable work either.
That second one is why "how much do I save with it" is misleading on its own. If nobody says what goes into the freed-up time, the saving shows up nowhere.
Why it is not the feeling that decides
There is one measurement worth reading before any decision about an AI routine.
In a randomised experiment, experienced professionals working on their own familiar work took 19% longer when they were allowed to use AI. Beforehand they had expected a 24% speed-up. After finishing the task — after living through the slowdown — they still estimated a 20% speed-up.
Thirty-nine percentage points between the feeling and the stopwatch.
One thing follows from this for the weekly routine: it needs measurement in it. Nothing complicated — it is enough to look, once a week, at how much time went into something and what came out of it. But you cannot build on the feeling, because the direction of the error is predictable: it errs upward. The real time savings post breaks the same question down by task type.
How long until it becomes a routine?
Here I am using an analogy, and it is worth saying so out loud: I found no research on AI habit formation. What exists is health behaviour research — but the mechanism transfers, and its numbers are sobering.
The best meta-analysis works from 20 studies and 2,601 participants:
- the median time for a habit to form is 59–66 days;
- the mean is higher than that, 106–154 days;
- individual variation runs between 4 and 335 days.
So "it takes 21 days to form a habit" is not true. If you measure your AI routine in weeks, you give up early.
Three transferable pieces of advice from the same place:
- do it in the morning, not in the evening — morning practice measurably beds in faster;
- you pick what you practise — a habit you chose yourself is stronger than an assigned one;
- in the same context — same time, same place, tied to the same trigger.
The weekly schedule
This part is the concrete one. The goal is not to use it a lot — it is to use it predictably.
MONDAY MORNING - 20 minutes
Writing out the week. What are the three things
that have to get done, and what is the first step
for each. The machine splits it up, you order it.
TUESDAY-THURSDAY - 10 minutes a day, same minute
One single recurring task. Do not keep switching.
Draft replies, tidying notes or a summary -
pick ONE for six weeks.
WEDNESDAY - 15 minutes
Whatever came up twice this week gets a template.
This is the one step that compounds.
FRIDAY AFTERNOON - 15 minutes
Measuring. What went faster, what went slower,
what had to be rewritten. Three lines, same file.
FRIDAY, LAST 5 MINUTES
One thing you will NOT do with a machine next
week, because it did not work out.
That adds up to roughly an hour and a half a week — exactly what executives already spend today. The difference is not the amount, it is that the same thing repeats, and that on Friday you look at it.
Three rules that make the schedule above work:
1. Do not switch tasks for six weeks. The median for habit formation is two months. Anyone who tries something new every week starts from day one every week.
2. The Friday measurement is three lines. What I did, how long it took, what had to be fixed. Not a project — three lines, into the same file. After six weeks it turns into data.
3. What does not work, stop doing out loud. A routine holds you if dropping something from it does not feel like a failure.
Shadow AI: what the routine is really about, only nobody says it
There is one phenomenon that overrides all routine planning if you do not face up to it.
The numbers vary widely, and the variation is itself instructive:
| What was asked | Result |
|---|---|
| "Have you used it knowing you were not allowed to?" (1,250 people) | two thirds |
| "Do you use an unapproved tool?" (1,000 people) | 81% |
| Telemetry-based aggregate | eight in ten employees |
The hardest and least disputable figure, though, comes from that same survey of 1,250 people in the first row of the table: 86% of respondents work at an organisation that in their view HAS an AI policy — and two thirds of them break it anyway.
It is not the policy that is missing. It is the usable alternative.
Two further numbers that back this up: most corporate AI use happens on personal rather than company accounts, and unapproved tools stay unnoticed for more than a year on median.
One thing follows from this for your own routine: if there is an approved tool at your workplace, use that one — not out of rule-following, but because the data handling is sorted there. If there is not, then leave out of your routine anything whose input is company data, until there is. Brainstorming, drafting and preparation still work regardless.
Common misconceptions
- "I'm already a regular user." Maybe — but by the measurement that means an hour and a half a week on average. Worth checking what it is for you.
- "If the tool is good, it builds itself in." In a randomised experiment 60% of participants did not use the free paid tool regularly over six months.
- "It takes 21 days to form a habit." The best meta-analysis has a median of 59–66 days, with individual variation from 4 to 335 days.
- "I can feel that I'm faster." In the most precise measurement the feeling was 39 percentage points out — in the wrong direction.
- "The tool is what matters." By the largest survey's own finding clear direction is worth more: more people reach a measurable effect with weak tools and a clear direction than the other way round — that is the report's own wording, the table above only shows the two extreme cases.
- "We have a policy, so we're fine." 86% of respondents have a policy, and two thirds of them break it.
What to take away
A routine does not become good by being large. It becomes good by being the same, at the same time, and by being looked at on Friday.
Three things come out of the measurements unambiguously: a habit takes two months, not three weeks; direction is worth more than the tool; and your own feeling on this topic is a demonstrably bad instrument.
That is why those fifteen minutes sit at the end of the weekly schedule. Not admin — it is the single point at which you find out whether your hour and a half really brings anything back.
Where exactly in your office day you can win time is mapped out in the seven task types post, with the studies; and for the first concrete piece of the weekly routine, the inbox triage post hands you a ready recipe.