How We Check Claims
Every claim in this series is a specific sentence a specific person said at a specific second, matched to a specific paper. This page explains how that is done, what the verdicts mean, where the method is weak, and how to tell us we got something wrong.
What gets checked
For each figure we select their most-watched longevity appearances — weighted by views, restricted to full-length material rather than clips — and work from the transcript. Every factual claim is pulled verbatim, with the timestamp at which it was said, so that each quote links back to the second it came from. You should never have to take our word for what someone said. Follow the link and hear it.
Opinions, predictions and jokes are not checked. A claim qualifies only if it asserts something about the world that could in principle be shown right or wrong: a study result, a number, a mechanism, a date.
The four verdicts
- Supported
- The claim matches what the cited evidence actually shows, at the strength stated.
- Overstated
- There is a real result underneath, described larger, more certain, or more general than the source supports. This is the most common verdict in the series, and it is a criticism of the sentence, not of the speaker's honesty.
- Contradicted
- The evidence points the other way, or the stated fact is checkably wrong — a date, a sample size, a journal.
- Unverifiable
- No checkable source was found. This does not mean the claim is false. Many personal claims — someone's own lab values, their own sleep scores — are unverifiable by construction, because no external dataset exists that could confirm or refute them. We say so rather than pretending to adjudicate.
How citations are verified
This is the part most claim-checking gets wrong, including our own first pass.
It is easy to remember a landmark study, describe it accurately, and attach the wrong identifier — a recent paper by the same author, or a different paper on the same topic. The description reads correctly and the citation is wrong. We found six such errors in our own draft material before publishing any of it.
So every citation now goes through a script that pulls the actual record from the National Library of Medicine and checks our written note against it: does the author we named appear on that paper, does the year match, is it the journal we said, and is it a primary research report rather than an erratum, a comment, or a retraction. Citations that fail are fixed or pulled before publication.
Automation catches provenance, not meaning. A citation can name the right authors, the right year and the right journal and still not support the sentence it is attached to — we found cases where the cited paper concluded the opposite of the claim it was filed under. Those only come out by reading the paper against the note, which is done by hand, prioritising the claims where the verdict carries the most weight.
Where this method is weak
Stating the limitations plainly is part of the method.
- Transcripts are machine-generated and contain recognition errors. "HRV" becomes "HIV", "NAD" becomes "Ned", "PFAS" becomes "PAS". Where a quote is garbled we flag it rather than quietly cleaning it up, because the cleaned version is no longer verbatim.
- Selection is by viewership, not by representativeness. These are the appearances most people actually watch, which is the right sample for "what is this person telling the public" — but it is not a random sample of everything they have said.
- Guests are not the host. A podcast episode contains claims by whoever is talking. Claims made by a guest are attributed to that guest and are not counted toward the host's record. Getting this wrong makes a host look far worse or better than the transcript supports.
- Absence of a source is not evidence of error. Our search failing and a claim being unsourceable are different things, and we distinguish them.
- We are reading abstracts and papers, not re-running analyses. Where a dispute is live in the literature, we say it is live rather than declaring a winner.
Conflicts of interest
Magellan Longevity earns affiliate commission on some products elsewhere on this site. That creates an obvious problem for a section that evaluates claims about supplements, so:
Pages in this series carry no affiliate links and no product placements. Not to the compound under discussion, not to anything else. If a claim check makes you want to buy or avoid something, that decision should not route through a page that gets paid either way.
Where a figure sells a product related to a claim we are checking, we say so. Where they have publicly reported a result that cuts against their own commercial interest, we say that too — it is relevant evidence about how they handle inconvenient data.
Corrections
We will get things wrong. When that happens the correction is made on the page, dated, and described — not silently edited. If you believe something here is inaccurate, including any of the figures whose claims we check, write to us and include the specific sentence and what the record actually says. Substantiated corrections are made regardless of who asks.
This series is evidence appraisal, not medical advice, and reading it does not create a physician–patient relationship. Nothing here is a judgment about any individual's character or intent — only about whether a particular sentence matches a particular body of evidence.