How We Read a Study

Last revised: 7 October 2026

BodyFat.Tips is built on a simple rule: claims come from sources, and we tell you what they are. This page explains where our information comes from, how a claim gets checked, what our ratings and numbers mean, and what we do when we get something wrong.

Where our information comes from

We start from the research itself, not from the news story or the press release about it. That research comes in two layers, and it helps to know which one you are looking at.

Primary research is the original studies, the ones that measured something themselves: trials, observational studies, lab and animal experiments. Reviews sit on top of them. A systematic review sets out to find every study on a question that meets its rules; a meta-analysis pools their results into one estimate. Either way, a review is a summary of other studies. That is how the US National Library of Medicine defines both a systematic review and a meta-analysis, and how the Cochrane Handbook describes them. So a review is secondary research, however good it is.

Most of our verdicts lean on reviews, because a well-run review sets out to gather every study that fits its question, not just the ones that suit an argument. We read them in full wherever we can get the full text, and we go to the original trials wherever a finding turns on one. When we could read only part of a paper, such as its abstract or its figures, the source list says so. We also read trial registries and regulators’ own documents. When we rely on reporting or a press release, we say so. And when a claim traces back to a single mouse study or a company’s own marketing, we tell you that too — because where a claim comes from is half the story.

The research itself is the floor, not the ceiling. We read widely to connect things: a new trial next to the older ones it revises, a supplement’s label dose next to the dose that was actually tested, a viral headline next to the study it claims to be based on. The goal is that every piece adds something that wasn’t there before — a comparison, a correction, a clearer frame.

How a claim gets checked

For each claim we look at what the evidence actually is, not just whether a study exists. That means reading for study design, how many people were involved, how long they were followed, who funded it, whether it has been replicated, and what the researchers themselves said the limits were. A single small trial and a large replicated meta-analysis are both “a study” — and they carry very different weight.

That reading produces two things you’ll see on every verdict: an evidence panel laying out what the piece rests on, and a confidence rating — Established, Supported, Preliminary, or Unsupported — that says how much weight the evidence can bear. The next section says what each one means.

What each rating means

Every verdict carries one of four ratings. This is what each one means:

• Established. Multiple good studies agree. Treat as settled.
• Supported. Evidence is consistent, with known limits.
• Preliminary. Early, small, conflicting, or animal-only. Not actionable alone.
• Unsupported. We looked, and the evidence does not back the claim: either it was tested and did not hold up, or nobody has tested it and nothing that was tested supports it. The verdict says which.

A rating is a judgement about the evidence, made by reading it for the things listed above. It is not a promise about what will happen to your body.

Each rating goes with one sentence, and it helps to know which. On an Established or Supported verdict, it goes with the finding the verdict states. On a Preliminary or Unsupported verdict, it goes with the claim as people repeat it: the one the verdict quotes under “How this claim travelled”, at the step headed “What it became”.

Ratings can move. When the evidence changes, or a correction changes what we can say about it, we rate the claim again, and a rating that moves is reported in the verdict’s correction note.

How we read the numbers

Three things come up on our verdicts again and again. Here is what each one means, in plain words.

The range around a result

A study’s result usually comes with a confidence interval: the range the true answer probably sits in. The result itself is the single best estimate; the range shows how far from it the real answer could plausibly be. A narrow range means the answer is pinned down. A wide one means there is more room for doubt. And when the range takes in the point that means no effect at all (zero for a difference between groups, 1 for a ratio), then “does nothing” is still on the table.

What the range is not: the range your own result will land in. It describes the average effect across the people studied. One person’s result can land well outside it, in either direction, so it should never be read as a forecast for you.

The p-value

A p-value answers a narrow question: if the treatment really did nothing, how often would chance alone produce a result at least this big? Researchers conventionally call anything under 0.05 “statistically significant”, meaning probably not chance, a line the Cochrane Handbook itself calls arbitrary.

It gets misread in two directions, and we try to avoid both. A large p-value does not show that the treatment does nothing; it shows the study could not tell its result apart from chance. A small one does not show that the effect is big enough to matter; for that you need the size of the effect and its range.

Trials and observational studies

In a randomised trial, chance decides who gets the treatment and who gets the comparison, so the two groups should differ only in the treatment and in luck. In an observational study, the researchers assign nothing: they compare people who already differ in what they eat, how much they move or what they take. That can show that two things go together. It is much weaker at showing that one causes the other, because whatever led people to their choice can also drive the result.

So when a verdict leans on observational evidence, read it as a link, not a promise that changing your own habit will change your result. The Study design row of every evidence panel names the kinds of study the verdict rests on, so you can see which you are dealing with.

These plain-word versions follow the Cochrane Handbook’s chapters on interpreting results and on non-randomised studies, and the US National Library of Medicine’s definitions of a randomised controlled trial and an observational study.

What we don’t do

We don’t accept payment to reach a conclusion, and no advertiser sees or shapes a verdict. We don’t present opinion as fact — when something is our judgment call, we say so. We don’t claim precision we don’t have: research has gaps, conflicts, and things that simply aren’t settled yet, and where that’s the case we tell you rather than pretend. And we don’t knowingly print a number we can’t trace. A figure about the research comes from a source in the page’s own source list, where you can check it. A figure for how often people search for something, or what advertisers pay for a click, comes from our own dated pulls of a paid keyword database, which you can’t open yourself; a page that quotes one should name that source and the date of the pull. The figure is Google’s estimate averaged over the latest twelve months the database holds, and rounded, so a month when the subject was in the news can lift it well above a typical month. When we last checked every page, on 9 October 2026, every such figure did, and each one matched a pull we had saved. When we find a figure we cannot trace, we take it out and say so on the page.

Updates and corrections

Research moves. New trials land, findings get overturned, and a verdict that was right last year can be wrong this year. We revisit our claims as the evidence changes, and we date significant revisions. When we get something wrong, we correct it on the page and leave the correction visible — it’s part of the record, not something to quietly erase. If you spot an error, tell us. Our full policies live on the Editorial Policy page.