Adherence 01: The first month

Short answer. How often people trained in their first 28 days was the strongest predictor of whether they were still training six months later. The data are observational, so early consistency may be a marker and not a cause. Our position is that the first month should be built for completion over stimulus: start below what you could do, and keep sessions short enough to survive a bad day.

---

What it is

Whether you keep training. Not how you train, not what you take, not when you rest. Whether you are still doing it in six months.

Every other entry in Field Notes examines a variable that changes outcomes by a few percent among people who continued. This one examines whether they continued, which decides whether any of those few percent arrive.

The proposed mechanism

Habit formation. Repeating a behavior at regular intervals strengthens automaticity and reduces the amount of conscious motivation required to perform it. On this account the early period matters disproportionately because it is when the behavior is still entirely motivation-dependent.

Self-efficacy through mastery. Completing progressively challenging sessions provides evidence of your own capability, which is proposed to be the mechanism by which early success predicts later persistence.

Basic psychological needs. Self-determination theory holds that satisfying needs for autonomy, competence and relatedness produces more self-determined motivation, which sustains behavior better than external pressure.

The three are compatible and predict the same thing: the beginning is where the outcome is decided.

What has actually been tested

The largest dataset we have found on this question. An observational cohort followed 522,994 adult users of a digital fitness app for six months from their first recorded workout, defining adherence as at least one workout per week with up to three missed weeks allowed (Conti et al., 2026).

Early consistency was the strongest predictor across every analysis performed. Training frequency during the first 28 days predicted sustained adherence more strongly than any other variable examined, and the association held after modeling time-varying effects. The protective effect was strongest early in follow-up and attenuated gradually.

Adherence declined steadily throughout, and the absolute numbers are worth printing. At six months, 18.3% of intermediate users and 25.8% of advanced users were still adherent by the study's definition, with beginners lower still. Three in four of the most experienced people in the dataset had stopped inside half a year. Demographic and experience characteristics shaped the trajectories, but with relatively modest differences by age and sex compared with the behavioral predictors.

Three other behaviors were associated with lower dropout. Greater diversity in equipment used, a higher emphasis on resistance exercise specifically, and longer workout duration among users who trained more frequently. Users training in commercial gyms showed higher adherence probability across the full follow-up.

Past behavior predicts future behavior, strongly. A separate analysis used structural equation modeling, which tests a chain of statistical relationships at once. Adding a consistent exercise routine over the preceding six months raised the share of variation in adherence the model explained from 9.2% to 64%, with satisfaction of basic psychological needs and autonomous motivation mediating the relationship (Cid et al., 2025). That 64% is explained variance, a measure of how well the model fits, and not a 64% higher likelihood of continuing, although the paper's own conclusion words it that way.

Our reading

This finding is uncomfortable for us and we are going to say so plainly.

The other entries in this publication concern variables producing standardized effect sizes (differences expressed in standard deviations) in the range of 0.1 to 0.3. Rest interval duration. Proximity to failure. Load selection. Frequency. We have spent considerable effort establishing where those effects are real and where they are overstated, and that work stands.

None of it competes with showing up. A person who trains consistently for two years with an indifferent program will be further ahead than a person with a well-designed one who stops at week seven. No study has run that comparison, and the cohort above did not measure progress at all. What it does show is how many people stop, and how early the stopping can be predicted.

The practical implication points the opposite way from most program design. If early consistency is the strongest predictor, then the first month should be built for completion rather than for stimulus. A program that is slightly worse on paper and gets done beats a better one that gets abandoned, and the early weeks are when abandonment happens.

That means starting below what you could do, keeping sessions short enough that they survive a bad day, and treating a completed easy session as a success, not a wasted one. The early weeks are not when you are building muscle. They are when you are building the thing that will build muscle.

And the variables in the rest of this publication become relevant later, not sooner. Fine-tuning rest intervals in month one is solving a problem you do not have yet. (→ Loading fundamentals sub-hub)

OMNID's beginner program is built this way: its first week carries a little over half the weekly sets it reaches by its fourth, and it builds from there. The dose-response curve describes people who kept training. (→ Loading fundamentals 07: Weekly volume)

Where the goals diverge

Starting out. The finding applies most directly here.

Returning after a layoff. Past behavior predicts in both directions: you have prior evidence of your own capability and prior evidence of stopping. What a layoff costs is covered separately, and it is less than most people assume. (→ Loading fundamentals 10: Time off and coming back)

Already consistent for years. The adherence question is largely settled for you, and the marginal variables the rest of this publication covers are where your remaining gains are. (→ Loading fundamentals 02: Reps in reserve)

Training around unstable circumstances. The finding suggests protecting frequency over session quality when something has to give, because frequency is what the habit is built from.

The case against

This is observational data and the causal direction is not established. People who were going to adhere may simply have trained more in their first 28 days because they were already the adhering type. Early consistency may be a marker rather than a cause, and no randomized trial has manipulated early frequency to test it.

The data come from a commercial fitness application. That is the only way to obtain a cohort of this size. It also means the population is people who downloaded a fitness app, the outcome is app-recorded workouts rather than training, and the organization holding the data has a commercial interest in engagement findings.

App-recorded adherence is not training. A logged session and a productive session are different things, and this dataset cannot distinguish them.

The commercial gym finding is almost certainly confounded by income, prior commitment and the sunk cost of a membership.

The author network runs through this entry too. The cohort analysis shares an author with most of the training-variable research cited elsewhere in this publication, which is the same concentration we flag in that branch rather than an independent check on it. (→ Loading fundamentals sub-hub)

And our own framing overstates the contrast. We have set adherence against programming as though they were competitors. In practice a program that produces visible progress supports adherence, so the variables the rest of this publication covers feed back into this one.

What is not known

Whether early consistency can be manipulated to produce later adherence, which is the question the data cannot answer and the only one that would justify a prescription.

What the mechanism is. Habit automaticity, self-efficacy and needs satisfaction are all proposed and none has been isolated in this context.

Whether the 28-day window is meaningful or an artifact of how the analysis was constructed.

Whether any of this transfers outside app users to people training without one.