Most health writing, including an earlier version of mine, works by piling up studies until the conclusion feels supported. That is the wrong move, because studies are not interchangeable. A well-run randomised trial and a survey of self-reported habits are not two votes on the same question.
So this page does something different. For each question, it picks the single best available study, says why that one wins, and states plainly what it does not settle. Where the best evidence is disappointing, that is what gets reported.
(The personal half, my own training history and the audit that started this, is on Twenty-Eight Years of Training, Audited. It is a worked example, not evidence.)
Where to start, depending on who you are
| If you are | Go to |
|---|---|
| Just wanting the plan | The plan that falls out of this, at the end |
| Sceptical of health claims | How I picked, then judge the rest for yourself |
| Wondering if exercise really extends life | Question 1, and brace slightly |
| Over fifty, or caring for someone who is | Question 4, the strongest evidence on this page |
| Choosing between hard intervals and easy miles | Question 3 |
| Never trained and unsure where the bar is | The floor |
How I picked: the filters
You do not need a statistics degree to rank evidence. You need about six questions, and they knock out most of what gets quoted at you.
1. Is there a control group? Without one you cannot know what would have happened anyway. A study of people who exercise, with nobody to compare them to, tells you nothing about the exercise.
2. Were people randomly assigned? This is the big one. If people chose to exercise, then everything else about people-who-choose-to-exercise is baked into the result: income, sleep, diet, health at the start, whether they can afford a gym. Randomising breaks that link. Studies where nobody randomised anything are observational: useful for finding patterns, weak for proving cause.
3. What was actually measured? A hard endpoint (did they die, did they break a hip) beats a surrogate like a lab value that we believe connects to something that matters. Surrogates are cheaper and faster, and they are also how a lot of confident advice ends up wrong.
4. How many people, and for how long? Sample size gets the attention, but what actually matters is how many events happened. A five-year trial of healthy people in which almost nobody dies cannot answer a question about dying, no matter how many were enrolled. This will matter enormously in a moment.
5. Was it pooled properly? A systematic review gathers all the studies to a pre-declared rule, rather than the ones an author liked. Cochrane reviews are the strictest version, and they publish a certainty rating (high, moderate, low, very low) using a framework called GRADE. When a Cochrane review says high certainty, that is close to the best this field offers.
6. Who paid, and was the plan published first? Pre-registering a protocol before collecting data stops the result being chosen after the fact.
⚠ One structural limit worth knowing before any of this. You cannot blind an exercise trial. Everyone knows whether they are exercising, so a placebo is impossible, and expectation is baked into every result. And nobody can randomise a person to thirty years of anything. So the long-term questions will never have the kind of evidence a drug gets. That is not a reason to dismiss the field. It is a reason to be honest that its ceiling is lower.
Question 1: does exercise actually make you live longer?
The best study: the Generation 100 trial. BMJ, 2020. Researched · randomised controlled trial · n=1,567 · 5 years · mortality endpoint
Why it wins. It is the rarest thing in this field: a randomised controlled trial with death as the endpoint, run for five years. 1,567 adults aged 70 to 77 in Trondheim were randomly assigned to twice-weekly high-intensity intervals, twice-weekly moderate continuous training, or a control group following national activity guidelines. Randomised, controlled, long, hard endpoint, published in a major journal. Nothing else on this question comes close on design.
What it found, and this is the uncomfortable part. All-cause mortality did not differ between the control group and the exercise groups combined. Analysed separately, the high-intensity group showed an absolute risk reduction of about 1.7 percentage points against a control mortality of 4.7%. Suggestive, but this was not the headline comparison.
Why the null result is not the same as “exercise doesn’t work.” Three things:
- The control group exercised. They were told to follow national physical-activity guidelines, and many did. This was a trial of structured supervised training versus general advice, not exercise versus nothing.
- Almost nobody died. 95.4% of participants survived the five years. A trial needs events to detect a difference in events, and this one had very few. The sample was large, the number of deaths was not.
- Five years is short for an outcome that accumulates over decades.
What the huge numbers you have seen actually come from. The famous figures, such as an 80% difference in mortality risk between the least and most fit, come from observational work, most notably a study of 122,007 adults who did treadmill tests. Enormous, consistent, and it found no upper limit to the benefit. But nobody was randomised, so fitter people differ in many other ways, and some of that difference is doing the work. ⚠ Reverse causation is a real problem here too: being ill makes you unfit, not only the other way round.
The honest synthesis. Exercise almost certainly helps, and the observational signal is enormous and consistent. But the best randomised evidence is far more modest than the popular framing, and anyone quoting an 80% mortality reduction at you (including a previous version of my own health map) is quoting an association, not a demonstrated effect.
Question 2: how much, then?
The best source: the WHO 2020 guidelines on physical activity. Researched · guideline · 27-expert panel · GRADE process
Why it wins. This is not a study, and that is the point: it is a synthesis of the studies, built by a 27-person guideline development group drawn from epidemiology, physiology and behaviour, working from systematic reviews and grading the evidence with GRADE. When a single trial and a guideline disagree, the guideline usually has seen more evidence than the trial did.
What it says. For adults: 150–300 minutes of moderate activity a week, or 75–150 minutes of vigorous, or a mix, plus muscle-strengthening on at least two days a week. For older adults, add balance and strength work. And the messages underneath it: some activity is better than none, and more is generally better.
Why I would start here rather than with any single study. Every individual trial on this page has a flaw you can name. The guideline has already weighed those flaws across a much larger body of work than any one of them.
Question 3: hard intervals or easy volume?
The best evidence: meta-analyses of randomised trials comparing interval training with steady moderate training. Researched · pooled RCTs · ⚠ surrogate endpoint (VO2 max), not mortality
Why it wins. These pool randomised comparisons, so the two groups are alike at the start. ⚠ But note the endpoint is VO2 max, a surrogate. We have good reason to think fitness matters; this is not the same as measuring whether anyone lived longer.
What they found. Interval training produces larger fitness gains than moderate continuous training; in one pooled analysis roughly 4.9 versus 1.9 mL/kg/min. The benefit is not automatic, though: it showed up for longer intervals (≥2 minutes), enough total hard volume, and programmes run for at least four to twelve weeks. Short, token intervals did not deliver it.
What this does not say. It does not say intervals beat easy work for health, only for measured fitness. And Generation 100, the one trial that looked at death, found the high-intensity group’s advantage small and not the primary result.
On the 80/20 rule, corrected 31 August 2026. I previously wrote that this came only from observing elite athletes. That undersold it. Polarised training has support from systematic reviews and meta-analyses of randomised trials, with pooled results showing it producing the greatest improvements in VO2 peak, time to exhaustion and peak power compared with threshold-focused or single-intensity approaches.
⚠ The real limits are narrower than I made them. The superiority appears mainly in interventions under 12 weeks (past that, distributions converge), and it is studied in endurance athletes, so it is a well-supported default rather than a proven prescription for general health. The fuller argument, including the case for and against Zone 2’s mechanism, is on the Zone 2 page.
Question 4: the strongest evidence on this page
The best study: the Cochrane review of exercise for preventing falls. Researched · Cochrane systematic review of RCTs · GRADE high certainty
Why it wins, and it wins clearly. It is a systematic review of randomised trials, conducted to Cochrane’s standards, and it rates the overall effect of exercise on falls as high certainty under GRADE. Hard endpoint (falls actually happening), randomised, pooled across many trials, independently conducted. On the quality filters above, this scores better than anything else I found in the whole field.
What it found. Exercise prevents falls. Programmes built around balance and functional exercise are the ones that work. Tai Chi and the Otago home programme are the best-supported examples, with reductions in the range of roughly 20–58% depending on programme and population.
⚠ And the finding that corrected my own writing: strength training alone did not prevent falls. It worked when combined with balance training. If you are training for later life, or helping a parent, “lift weights” is not a falls strategy on its own.
Why this matters more than it sounds. Falls are the leading cause of injury and injury death in adults over 65 (CDC), around one in four fall each year, and the WHO ranks falls the second leading cause of unintentional injury death worldwide. This is the one place on this page where high-certainty evidence meets an outcome that plainly changes a life.
Question 5: how much strength work?
The best available: a meta-analysis of 16 prospective cohort studies, British Journal of Sports Medicine. Researched · ⚠ observational, self-reported activity; no RCT exists
Why it wins by default, which is a weaker claim. It pools many studies and finds a consistent signal: 30–60 minutes a week of muscle-strengthening activity associated with 10–20% lower all-cause mortality, on a J-shaped curve. The benefit plateaus, with no good evidence that more than about an hour a week buys additional longevity.
⚠ But every study in it is observational, and the activity was self-reported. People who lift weights differ from people who do not, in the same way fit people differ from unfit people. There is no randomised trial of strength training with a mortality endpoint, and given the design constraints there probably never will be. So: a useful number, held more loosely than the falls evidence.
The practical read stands anyway, because it points the same way as the guideline. WHO says at least two strengthening days a week; the pooled cohorts suggest half an hour to an hour is where most of the return sits. Two short sessions is not a beginner’s version of strength training. On this evidence it is the dose.
What actually changes with age
The four capacities matter at every age; their priority order moves, and the confidence behind each claim differs.
- Twenties and thirties. 📖 Build the aerobic base. This is the least evidence-backed of the age claims. It is a reasonable inference from how adaptation works, not a trial result.
- Forties. 📖 Strength stops being cosmetic. Also inference, supported by the guideline’s two-days-a-week floor.
- Fifties and beyond. 📊 Balance moves to the top. This is the best-evidenced statement on the page: high-certainty Cochrane evidence, hard endpoint, and a serious outcome.
- Caring for an older parent. 📊 Balance work specifically, not walking alone, and enough protein. The best-evidenced programmes are free and unglamorous.
Nobody is too old to start. Benefits appear in every age band studied, and Generation 100’s participants were in their seventies when they began.
The floor, if you have never trained
Ignore everything above and do this:
Walk most days, for a length of time you would not call ambitious. Then add two short sessions a week where you pick things up and put them down. That is the whole thing, and per the pooled strength evidence, those two sessions are not a stepping stone toward a real programme. They are the programme.
The most common failure here is not laziness. It is starting with a plan designed for someone with more time, failing in three weeks, and concluding you are the problem.
The plan that falls out of this
| Element | What to do | Confidence |
|---|---|---|
| Weekly aerobic | 150–300 min moderate, or 75–150 vigorous, or a mix | 📊 High. WHO guideline, GRADE process |
| Strength | 2 days a week, 30–60 min total | 📊 Moderate. Guideline is firm; the dose-response is observational |
| Balance | Explicitly, from your fifties, and sooner if you like | 📊 Highest on this page. Cochrane, high certainty |
| Intervals | Optional. Longer intervals (≥2 min), sustained over months, if you want VO2 max specifically | 📊 Moderate. Randomised, but a surrogate endpoint |
| Easy-to-hard split | Mostly easy, some hard | 📊 Moderate. Meta-analytic RCT support in athletes; ⚠ mostly in blocks under 12 weeks |
| Restarting | Roughly half your old load for two weeks | 🔧 My own rule. Load-spike research exists but the popular model is contested |
| Protein alongside | 1.4–2.0 g/kg/day; more, in larger doses, past ~60 | 📊 Moderate–high. ISSN position stand; anabolic-resistance literature |
📊 evidence · 📖 framework or inference · 🔧 my own rule
The three things this turns into, in detail
Each of these has its own page with the research, the protocol, and a week-by-week ramp for someone starting from nothing.
| Guide | What it covers |
|---|---|
| Zone 2 | The easy aerobic base: both sides of a contested mechanism, and an eight-week start from walking |
| VO2 max training | The Norwegian 4×4, the trial behind it, and a twelve-week ramp to earn the right to do it |
| Resistance training | The dose-response research, six movements, and twelve weeks starting at home with nothing |
If you do only two things: hit the WHO aerobic minimum, and do two strength sessions a week. If you are over fifty, make one of them include balance work. That is the single best-evidenced sentence on this page.
Message me
If you think I have picked the wrong study for any of these, or read one badly, I would genuinely like to hear it. That is a better use of this page than agreement. And if you want help judging a study somebody has quoted at you, the six filters near the top are most of what I use.
Email: abhishek@terrak.ai LinkedIn: Abhishek Chaudhry
What’s changed
- 31 August 2026. Corrected the 80/20 entry. I had written that polarised training was merely observed in elite athletes; it in fact has meta-analytic support from randomised trials, and I had undersold it while trying to be appropriately cautious.
- 31 August 2026. First published, split out from the personal training page so the evidence could be weighed on its own.