How we check a fitness claim
The same rules for every claim, written down so you can hold us to them.
What counts as evidence
Pooled evidence comes first. The longer the bar, the more weight a source type carries in a verdict.
WHO, the US Department of Health and Human Services, ACSM and ISSN. They weigh many studies and say plainly what they recommend.
Studies that pool many trials into one estimate with an interval. Our first choice for training and supplement questions.
For outcomes trials can’t test, such as how daily steps relate to the risk of death over years.
Only to explain or to show where evidence disagrees, never as the basis of a verdict on their own.
Within each level we prefer the newest high-quality review, and we say when newer evidence changed the picture.
How we decide a verdict
We judge the claim the way people usually say it, not the most charitable version. “1 gram per pound” is judged as a requirement, because that is how it is posted. When the direction is right but the number is wrong, the verdict is “Partly true”.
- Verdict: Holds up
Holds up
The best evidence supports the claim as people usually state it.
- Verdict: Partly true
Partly true
Part of it holds up, but the number, the scope or the certainty is off.
- Verdict: Doesn’t hold up
Doesn’t hold up
The best evidence contradicts the claim, or the effect is too small to matter.
- Verdict: Too early to say
Too early to say
There isn’t enough good evidence yet to call it either way.
How we show uncertainty
Every estimate comes with its interval when the source reports one. A 95% confidence interval is the range of values reasonably compatible with the data. When an interval crosses the no-difference line, we say the result is not statistically significant.
We label the unusual ones. One cold-plunge estimate comes from a Bayesian model, so its interval is a credible interval, not a confidence interval, and the chart note says so. When an interval wasn’t reported, the chart says “no interval” instead of inventing one.
We say when we flip a sign. To keep every bar in a chart pointing the same way, we occasionally show a published result with its sign reversed. Twice so far: fat burned during fasted exercise (reported as fed minus fasted) and one 2026 result on training to failure. The caption says so each time.
We round, and we say so. “About 2,500 steps” stands for a published 2,517. The exact figure is in the chart.
Ties to industry, said out loud
When the authors of a review we rely on report ties to a company that sells the thing being studied, we say so in the text. For example, several authors of the creatine reviews report ties to creatine makers, which the creatine check mentions alongside the results.
FitClaims itself has no sponsors, sells nothing and uses no affiliate links.
Lines we don’t cross
- No individual advice. Every page is general information; your clinician knows your health.
- No weight-loss targets, diets or calorie goals.
- No supplement doses beyond what an official position stand states, and we cite it.
- No quoting or naming the people who post a claim. We check ideas, not creators.
- No invented numbers. If a source doesn’t report something, neither do we.
Corrections and updates
Every check shows the month we last reviewed it. We revisit a claim when a better meta-analysis or a new guideline appears, and we change the verdict if the evidence changes.
Every page is researched, written and drawn by go4 with AI assistance (Claude, by Anthropic), then checked number by number against the linked sources.