Deciding is the hard part.
Decision fatigue is not why.

The short answer

You are probably not short of motivation, and your willpower did not run out. Between wanting to train and training there is a stack of unresolved questions, and every one of them is a place to stop. What reliably helps is making the next step cheap and already specified. What does not reliably help is being told to want it more.

Start with the number that matters

Sit with that for a second, because it reframes everything. The people failing to train are, overwhelmingly, people who already wanted to. They are not unconvinced. The failure happens after the intention exists.

Which is why motivational content mostly does not work on them. Reviews of the intention-behaviour relationship find that experimentally producing medium-to-large increases in intention yields only small changes in behaviour. You cannot fix a translation problem by adding more of the thing that was never missing.

Now the part we got wrong

The natural next move is to reach for decision fatigue: the idea that choosing depletes a finite mental resource, so by evening you have none left. It is a satisfying story. It is also, as far as anyone can currently show, not true.

The famous study is the Israeli parole board data from 2011, where favourable rulings appeared to collapse across a session and reset after a break. Two problems surfaced quickly. Case ordering was not random, with unrepresented prisoners tending to be heard last. And favourable rulings take longer to deliver, so a judge approaching a break preferentially slots in the quick cases, which are the refusals. The pattern can be reproduced without anyone being tired.

The field evidence has been mixed. Studies of clinicians do find decision quality drifting across a session. But the best-designed test to date, a registered report analysing 231,076 real medical judgments by 174 triage nurses, found strong evidence for no decision fatigue on every pre-specified test.

Its authors made a point worth repeating, because it is the trap this whole area falls into. They observed that urgency ratings rose later in the day, and noted it would have been easy to present that in good faith as evidence of fatigue, when it is more likely confounded by other things that vary with time of day.

You should also drop the glucose version entirely. The idea that a sugary drink restores self-control has not held up.

What survived, and it is better

Something real sits underneath the folk theory. Mental effort is genuinely unpleasant, and people avoid it: given two versions of a task for the same reward, they reliably pick the easier one, and they require compensation to do harder cognitive work.

But the modern accounts describe a price, not a fuel gauge. Exerting control shifts your motivation and your attention rather than draining a tank, and the feeling of mental fatigue functions as a signal that what you are doing has become expensive relative to the alternatives.

In plain terms: you do not run out of willpower. You keep a running tally of what things cost, and an unresolved decision is expensive. When the price of starting gets high enough, you find something cheaper to do. That is arithmetic, not weakness, and it responds to changing the price rather than to being lectured.

Too much choice: real, but only sometimes

The jam study is the other pillar of this argument in popular writing: a tasting booth with 24 varieties drew more visitors but sold to about a tenth as many of them as a booth with 6.

It did not generalise. Across 63 experimental conditions and 5,036 participants, the average effect of more options is virtually zero. A direct field replication in a supermarket found no negative effect at all.

The larger analysis, 99 observations across 7,202 people, explains when it does appear. Choice overload shows up reliably when four things are true at once:

• The options are hard to compare, with no obvious structure or standout
• The task itself is difficult, with time pressure and no help
• You do not already know what you want, and have no expertise here
• Your goal is to be finished rather than to be right

Read that list again as a description of a person opening a training app for the first time, or deciding dinner at seven in the evening. Two hundred exercises with no basis for ranking them. No way to know whether today should include Romanian deadlifts. Tired, and wanting the question to go away.

So the honest claim is not "too much choice is bad." It is that this specific corner of the problem is exactly where choice overload replicates.

The lever that actually moves

That ratio is the most useful thing in this entire article, and we are quoting the ratio rather than the absolute numbers deliberately. The nudge literature as a whole has a publication bias problem serious enough that a bias-corrected reanalysis of its main meta-analysis concluded the evidence for an average effect largely disappears. Comparisons within one analysis survive that critique far better than headline effect sizes do, because both arms are biased the same way.

The real-world demonstrations are hard to argue with. Switching a retirement plan from opt-in to opt-out transformed participation without changing a single person's preferences. Redesigning a court summons so the date and consequences appeared at the top cut failures to appear by 13%, and adding text reminders cut them by 21%, producing roughly 30,000 fewer arrest warrants over three years. Nobody in that group was unmotivated to avoid arrest. The information was in the wrong place.

Where this gets complicated, and we are not going to hide it

The obvious conclusion from all of the above is: remove the decision, fix the schedule, done. Two good field experiments say otherwise.

One paid gym members for visits only inside a planned, fixed daily two-hour window, against a group paid for a visit at any time that day. The fixed-window group made fewer visits, during the study and after the payments stopped. Another randomised 877 gym members to pre-select the days and times they intended to attend, and found a tightly estimated null.

The plausible reading, and we present it as a reading rather than a finding, is that a rigid schedule creates a binary failure condition. Miss the window and the day is a write-off. A missed day on its own does not appear to damage habit formation, but a rule converts a missed day into a violation, and violations are what people quit over.

The distinction we draw from this, which the evidence supports in direction but has not tested head-on: pre-decide the content, keep the timing flexible. Knowing what today's session is removes preference uncertainty and comparison difficulty, which are the conditions that actually produce choice overload. Committing to a particular hour adds a way to fail without removing any of them.

Two numbers to stop repeating

"You make over 200 food decisions a day"

We used this ourselves before we checked it, and we have since taken it out of an earlier piece. Nothing was ever counted. Participants estimated their daily food decisions, averaging 14, then estimated components for a typical meal, and the components were multiplied out and summed to reach 227. The difference was labelled "mindless."

A 2025 analysis identified it as a textbook subadditivity effect: ask people about anything in pieces and the pieces total more than the whole. Nothing about food is required to produce it. The paper's lead author also resigned from his university following a research misconduct finding, with 18 papers retracted.

The point survives without the number. Eating recurs several times a day, in changing contexts, usually under time pressure and usually without a plan. That is a description of the conditions where choice overload is real, and it does not need a fake statistic attached.

"It takes 21 days to build a habit"

No empirical basis. It traces to a 1960 self-help book by a plastic surgeon writing about how long patients took to adjust to their new appearance.

The research-backed figure people reach for instead, 66 days, is a median from a single study of 96 people, and quoting it as a rule is nearly as bad. The actual finding is the spread: individuals ranged from 18 to 254 days, and a 2024 meta-analysis put the range at 4 to 335. Repetition in a consistent context matters more than elapsed time, and exercise forms habits more slowly than eating or drinking does.

One last piece of honesty

The largest experiment ever run on this question put 30 scientists and 53 different interventions in front of 61,293 gym members. Just under half the interventions worked, raising weekly visits by 9 to 27%.

And the effects largely faded once the interventions stopped.

Anybody, including us, who tells you that one design change produces lasting behaviour change is overclaiming. Short-run behaviour is cheap to move. Durable behaviour is not. The most that software can honestly promise is to stop being one of the reasons you did not start.

← Back to the library