The Best Time of Day to Exercise — and Why the Studies Disagree

In the space of six weeks this spring, three large studies announced the best time of day to exercise. They named three different times.

One said seven in the morning. One said the middle of the day. The third — the only one that randomised anything — declined to name an hour at all, and quietly suggested the question had been asked backwards.

That disagreement is not a failure of science. It is the finding. And once you can see why each study landed where it did, the practical answer gets simpler rather than more complicated, in much the same way the evidence on when you eat turned out to be about your own rhythm rather than the clock on the wall.

Three studies, three different best hours

In March, at the American College of Cardiology meeting, researchers presented an analysis of 14,489 adults in the All of Us programme who had worn a fitness tracker for roughly a year and agreed to have that data linked to their medical records. Morning exercisers came out ahead on nearly everything measured: 31% lower odds of coronary artery disease, 30% lower for diabetes, 35% lower for obesity, with the sharpest signal sitting in the seven-to-eight o'clock hour. The authors called the work hypothesis-generating. The internet called it settled.

In April, the European Journal of Preventive Cardiology published something larger and slower: 75,509 UK Biobank participants who had worn research-grade accelerometers, followed for a median of 8.7 years, with death as the outcome. Among people already meeting activity guidelines, the timing pattern with the lowest all-cause mortality was not morning. It was midday — a hazard ratio of 0.79 (95% CI 0.65–0.97) against a pattern of activity spread evenly through the day, and that was after adjusting for how much activity people did in total.

Also in April, Open Heart published the only randomised trial in the set. One hundred and fifty sedentary adults aged 40 to 60 in Lahore, each carrying at least one cardiovascular risk factor, were assigned to walk briskly for forty minutes, five times a week, for twelve weeks. Some trained between eight and eleven in the morning, some between six and nine in the evening. One hundred and thirty-four finished.

Three research groups. Three defensible methods. Three different hours. When that happens, the problem is usually not the data — it is the question.

Why the morning result is probably about morning people

Observational timing studies compare people, not protocols. The seven a.m. exerciser and the nine p.m. exerciser differ in a hundred ways that have nothing to do with hormone curves: shift patterns, childcare, commutes, stiff joints, sleep debt, income, and whether the body in question is willing to leave the bed at six.

Almost all of those differences sort in the same direction. People whose knees hurt at dawn, who work nights, who sleep badly, or who are already unwell get quietly filed into the later categories. Which means that "morning exercisers are healthier" and "healthier people can exercise in the morning" produce exactly the same table.

The UK Biobank result is the tell. Bigger cohort, harder outcome, much longer follow-up, adjusted for total volume — and it crowned a different hour. When two large observational datasets disagree about which hour wins, the honest reading is not that one of them is right. It is that the hour is a weak variable carrying a great deal of other people's lives on its back.

The only randomised answer points at your clock, not the clock

The Lahore trial did the one thing the cohorts could not: it assigned the time. Participants were first sorted into morning or evening types, then randomised to train either at their preferred end of the day or at the other one. Same exercise, same dose, same supervision, same twelve weeks. Only the alignment differed.

The aligned group moved further on nearly every measure. Systolic blood pressure fell by 10.8 mm Hg in the aligned group against 5.5 in the mismatched one (p=0.002), with diastolic following the same pattern. So did RMSSD — a beat-to-beat index of heart rate variability — along with peak VO₂, LDL cholesterol, fasting glucose and self-reported sleep quality.

Read carefully, that trial says nothing at all about mornings. It says that being made to train against your own rhythm cost these participants roughly half the benefit of identical training.

What chronotype is, and what the trial could not measure

Chronotype is simply where your internal day sits relative to the clock on the wall. Some people's circadian timing runs early, some late, and most of us sit somewhere in the unglamorous middle — a group this trial deliberately excluded.

Independent researchers reviewing the study made a sharper point. It measured preference, using a questionnaire alongside some temperature monitoring, rather than circadian phase itself. Preference and biology overlap, but they are not the same object: your preferred training time is also a product of your job, your family and your habits. The trial may have shown that training when you want to beats training when you don't. That is a real and useful finding. It is not quite the one in the headline.

It was also 150 people in a single city over twelve weeks, with sleep captured by questionnaire rather than measured. Take the direction seriously. Hold the decimals loosely.

The one timing rule that has a dose–response

There is a piece of exercise-timing advice that does survive scrutiny, and it has nothing to do with which camp you belong to. In 2025, Nature Communications published an analysis of 14,689 physically active people across 4,084,354 nights of wearable data, comparing each night against that person's own baseline rather than against other people.

The result is graded rather than binary. Sessions that ended four or more hours before sleep onset showed no association with sleep at all, at any intensity. Closer than that, and strain starts to matter: the hardest sessions finishing nearest to bedtime tracked with sleep onset up to 80 minutes later, sleep up to 13.9% shorter, sleep quality 5.6% lower, a higher overnight resting heart rate and lower overnight heart rate variability.

So the rule is not "don't train at night." An easy evening walk is close to free. A brutal interval session an hour before bed is not — and the cost turns up in two numbers you can already read the next morning.

The variable that dwarfs all of this

Put the effect sizes side by side and the hierarchy is faintly embarrassing. Going from sedentary to meeting activity guidelines is associated with mortality reductions in the tens of percent, in essentially every cohort ever assembled. Shifting sessions you already do from six in the evening to noon bought the UK Biobank's guideline-meeters a hazard ratio of 0.79. Real, probably. A rounding error next to the first step, certainly.

The optimisation instinct always prefers the fine adjustment to the large one, because the fine adjustment is more interesting to discuss and considerably easier to buy. It is more pleasant to debate the ideal training hour in a room with good lighting than to go outside and do the unremarkable aerobic work that most of the benefit actually comes from.

The honest ranking, then: train at all; train consistently for years; train at an hour you don't resent; keep the hardest sessions out of the last four hours before sleep; and only then, if you enjoy this sort of thing, start experimenting with your chronotype.

How to run the experiment on yourself

None of the studies above can tell you what your body does, because none of them contained you. What they can do is tell you where to look. Two weeks per window is usually enough to see a signal if one exists.

Hold everything else steady — same sessions, same intensity, same weekly dose — and move only the hour. Then watch three things you can measure at home without much ceremony: your overnight resting heart rate, your HRV trend, and how long you take to fall asleep. The Agen Band tracks all three overnight, and the same caveats apply here as everywhere else in what a wearable can and cannot measure: these are trends worth noticing, not measurements worth arguing about.

Read them like weather rather than verdicts. You are looking for a persistent shift across ten to fourteen nights, not one good Tuesday. A single bad night after an evening session tells you about that night; a fortnight of later sleep onset and a higher overnight pulse tells you about the schedule.

And keep the tiebreaker that isn't a number at all: which window did you actually turn up for in week six, when it was raining? Numbers are for correcting fantasy, not for replacing experience. If the morning slot buys you a slightly better overnight heart rate and you hate every minute of it, the evening slot is the better training, because it is the training that will still be happening next year. That is the same logic that runs through the rest of building a protocol you can keep.

One caution worth stating plainly: if you have a cardiovascular condition, or take medication that affects heart rate, blood pressure or sleep, talk to your doctor before making significant changes to how or when you train. Shift workers are a genuinely different case — none of this research addresses a body clock that is being asked to move every few days.

The bottom line

Three large 2026 datasets each crowned a different best hour, which is about the clearest available evidence that the hour is not the point. The only randomised trial in the set found that matching training to your own rhythm outperformed identical training at the wrong end of the day — with the honest caveat that what it measured was preference rather than biology.

Keep one rule and let the rest go: finish the hard sessions roughly four hours before you intend to sleep, and the evening is yours. Everything else on this page is a refinement of a decision you already made correctly, on the day you decided to train at all.