Methods

How we check a claim

So far we have checked 89 claims and 5 full episodes from longevity podcasts against 910 studies. Code computed every badge from what those studies found. Nobody types in a badge or edits one.

Where the claims come from

Every claim starts as words someone said on a podcast or in a recorded interview on YouTube.

The quote is copied word for word from the transcript. A quote runs 20 seconds or less, and its link opens the video at that second.

Before testing, we read two minutes before the quote and three minutes after it. Each hedge stays, like "in my view" or "linked to." We test the claim as said, never stronger or weaker.

Each main point of a claim gets tested alone. A side detail we can't test is listed as not tested.

The claim goes to the person who made it: the guest in an interview, the host on a solo episode. We don't count a host's remark inside an interview.

We pick claims by rule, never by who said them. A claim check covers one claim, most often one that shows repeat across episodes. An episode check covers a whole episode, chapter by chapter, and shows the chapters we did not check in grey.

How we find the studies

We search PubMed, the index of medical research run by the US National Library of Medicine.

We start with the landmark trials and reviews a critic would name, then add every study the speaker cited.

Every study that tests the point gets ranked. First comes how closely it tests the point, then landmark status, then how often other studies cite it.

We read down that list until each point has ten studies that test it, or the list runs out. Each full text comes from legal sources only.

A study that tested another dose, another group of people or another outcome does not count against the claim. We mark it as not testing the claim.

Two readers mark each study

Two readers read each study, and each marks it as agrees, partly agrees or disagrees with the point. Both readers are AI models, and neither sees the other's answer.

When they split, a third reader decides without seeing either answer. Two of three wins.

Each mark rests on one sentence from the study. The page shows that sentence word for word, with a link to the study.

Which studies count

A study in this list gets a hollow dot on the page and does not count toward the badge.

  • Read from the abstract. It could change the badge, so it counts once read in full.
  • The readers did not agree on its stance.
  • An animal or lab study.
  • A review or commentary, not a study.
  • Retracted.

How code computes the badge

Each counted study is one source. On reports and answers, a researcher on tape is a source too. Each researcher counts once, by their newest view. On a claim check, the quote never counts as evidence for itself.

The rule runs these steps in order and stops at the first one that fits.

  1. Not enough evidence

    Fewer than 3 sources count, fewer than 2 different sources, or fewer than 3 take a side.

  2. Conflicting

    At least 25% of sources agree, and at least 25% disagree.

  3. Strong consensus for or against

    Agree minus disagree is at least 60% of all sources. At least 4 sources count, and 15% or fewer take the other side. At least 3 sources back the leading side. If only that last part fails, the badge is Leaning.

  4. Leaning for or against

    The net share is at least 25%. That share is agree minus disagree over all sources, with partly agree counted as half.

  5. Mixed

    Everything else.

What Precis is not

It is not medical advice. Talk to your doctor before you change anything.

It is not a systematic review. We check the studies that test one claim, not all research on a topic.

It is not a verdict on any person. The person named on a claim is where we heard it.

Who pays for this

No brand, sponsor or ad network pays us. How Precis stays independent.

Found an error?

Tell us, and we will check it. How to report an error.