How we review research
How Body CEO finds exercise studies, checks them, weighs them and turns them into advice. Every summary is checked against the study’s own abstract.
Where do the studies come from?
Every study comes from free, public research databases. We find studies on PubMed, through Europe PMC. Then we check each one with Crossref, OpenAlex and Unpaywall. These tell us if a paper was taken back, how often it is cited, and where you can read it. If we can’t link to a study, we don’t use it.
How do we pick studies?
A study counts only when the exercise is what it tests. AI reads each study’s title and summary to check it’s really about the exercise. If it only mentions the exercise, a person checks it before it can appear. Studies of children, patients or a single person can appear on their exercise’s page, but are never featured.
- The exercise is the subject. It’s in the title, or it’s what the study trains or compares.
- AI checks the fit. It reads each study’s title and summary, and says if the exercise is its subject, what it tests, or only mentioned.
- A study that only mentions it waits for a person. It never goes live by itself.
- Supplement trials are left out when the exercise is only the test, like caffeine before a bench press.
- Children’s fitness tests are left out, like push-up tests in school PE.
- Patient and rehab groups are left out, unless the exercise is the treatment being tested.
- Some studies are never featured. Single cases, and studies of children or patients, can appear on their exercise’s page. They never appear in Latest science, the app’s examples or workout tips.
How do we weigh studies?
Not all studies are equal, so each one gets an evidence score out of 100. The kind of study counts most. Then we look at how well it fits the advice, how well it was done, and how much other researchers use it, compared fairly. Retracted studies are never used.
The kind of study, strongest first
- 1A review that pools many studies (a meta-analysis or systematic review)
- 2A randomised trial (people put into groups by chance)
- 3Another controlled trial (groups compared, but not by chance)
- 4A study that follows people over time (a cohort study)
- 5A one-off lab test (like sensors that measure how hard a muscle works)
- 6A single case, or an expert’s opinion (including reviews that don’t pool studies)
What else counts
| Part | What we check |
|---|---|
| Fit | The exercise is the subject. And what the study measured matches the advice: muscle activity alone can’t back “builds muscle”. |
| Quality | How many people, and for how long. Who: adults who train rank above children, patients or elite athletes only. A plan registered before the study started. The authors’ own rating of their evidence. A peer-reviewed journal we can find in OpenAlex. |
| Fair citations | How often it’s cited compared with papers of the same age and field, never the raw count. An old paper has simply had more years to collect citations. Papers under two years old are too new to judge this way. |
| Newest reviews | When a newer review already includes an older trial, we cite the review instead. |
| Never retracted | A study that was taken back, withdrawn or flagged with a concern is never cited. |
| Who paid | Company funding is shown, so you can judge it. It doesn’t lower the score. |
We don’t yet check journals against a list of “predatory” publishers (journals that publish almost anything for a fee), because we haven’t found a list we can check reliably.
What the labels mean
Each study gets its own label, from its evidence score.
A high score: usually a well-run review or a large trial that fits the exercise.
A middle score: usually a good trial, or a smaller review.
A low score: usually a small or lab study. Shown only when there’s nothing stronger, and labelled Early.
Studies disagree shows when good studies point different ways. We tell you, instead of picking a side.
How does a study become advice?
AI writes a first summary in plain words, from the study’s abstract only. It appears only after automatic checks: every number from the abstract, no words like “proven”. A summary that fails twice stays off the page. Approved takeaways from Strong or Moderate studies guide your plan’s sets, reps and rest. Eliza, the assistant, reads them before answering training questions.
- 1 · FoundA weekly search on PubMed, with the filters above.
- 2 · ScoredThe evidence score and its reasons, in plain words.
- 3 · SummarisedAI drafts plain words from the abstract. Any number that isn’t in the abstract is flagged.
- 4 · CheckedEvery number must come from the abstract, with no promises. Only then does it show.
- 5 · Shapes your planTakeaways backed by Strong or Moderate studies guide your plan’s sets, reps and rest. In the app’s workouts, one shows under the set advice, linked to its study.
- 6 · Eliza cites itFor questions on technique, sets, reps or rest, the assistant reads our approved studies first and links the one it uses.
What do we never do?
We never give medical advice, copy a study’s own words, or make up a source. Every study links to a place where you can read it yourself. And we never say research “proves” something, because studies show what is likely, not what is certain for every person.
- No medical claims or diagnosis. This is general fitness information, not medical advice.
- No copied abstracts. We write our own summary and link to the study.
- No first pages without permission. We show a paper’s first page only when its licence allows it (CC BY or CC0).
- No made-up citations. Every study links to PubMed or its DOI.
- No “proven”. We say what a study found, not what it promises.
Who reviews the research?
Study summaries go live after automatic checks. Answers and common questions that pass a second AI’s check go live on their own. Takeaways, steps, cues and mistakes wait for Body CEO’s team. A page names a reviewer only when a confirmed reviewer approved it. Until then it says “Written with AI from published research, checked automatically”.
Written with AI from published research, checked automatically.
How do we keep it up to date?
We search for new studies on each exercise every week. At the same time, we check every study we use again for retractions, so a paper that was taken back comes off the page. The numbers below come straight from our database, so they are always current.
See every study we’ve looked at, including the ones we didn’t use, and why
Questions people ask
Is this medical advice?
No. It’s general fitness information based on research. If you have a health condition, an injury or pain, talk to a doctor or physio before you start.
Why do some exercises have no studies?
Some exercises haven’t been studied much on their own. We only show studies where the exercise is the subject. So a page shows no studies rather than weak or off-topic ones.
Why aren’t the most-cited studies always shown?
Older papers have had more years to collect citations. So we compare each paper with others of the same age and field. And a newer review that includes an older study usually tells you more.
Does AI write the advice?
AI writes the first draft from the study’s abstract. Study summaries show only after automatic checks: every number from the study, no words like “proven”, no retracted studies. An exercise page’s answers and common questions go live on their own once a second AI finds nothing wrong. The takeaways that guide your training are checked by a second AI. Then Body CEO’s team approves them: each flagged one is read, and the rest are spot-checked. The team can take anything down at once.
What does “Early” evidence mean?
The study’s evidence score is low: usually a small or lab study. It may still help, but bigger studies could change the picture. An exercise page shows an Early study only when it has fewer than three Strong or Moderate ones, and labels its card Early.
What happens if a study is retracted?
It comes off the page. Every week we check each study again for retractions, withdrawals and expressions of concern.
Found a mistake?
Please tell us. Email support@mindvalley.com with the page’s address and what looks wrong. We check it against the study itself. If we got it wrong, we fix the page and the advice that rests on it. We don’t have a form for this yet; one is coming.
316 studies approved so far. Body CEO gives general fitness information, not medical advice. Stop if anything hurts.