What we research from — and what we leave out
Where an episode’s research comes from, which sites it never draws on, and how a client’s own documents are used. The fact check compares the script with exactly these captured sources — nothing else.
- Research captures the full text of each source it keeps and stores it with the episode.
- Forums, social media, video, open publishing platforms and supplement retailers are never searched.
- A client can supply their own documents, alone or alongside web research, and they are labelled as theirs.
- The fact check reads these captured sources and nothing else — not a live web search, not a model’s memory.
Research is what the check reads. The check itself is on the fact-check page.
A fact check can only be as good as what it reads. Ours compares an episode’s script with the sources that episode was researched from, so the honest question is not only how the comparison works but what goes into it — and what is kept out.
Where the research comes from
Research starts from the brief the creator writes: a set of questions is planned from it and searched on the web. For every source kept, the full text of the page is captured and stored with the episode — whole pages rather than opening paragraphs, so that a paper’s results table is still there later and not only its abstract.
One of the searches looks only at journals, research archives and trial registries, and places in the episode’s sources are held for original research and official publications, so the record itself is not crowded out by faster-moving coverage of it. That is a preference in where research looks, not a rule about what the check accepts.
Each captured source is stored with a limit on its length, so the far end of an exceptionally long document may not be kept. The episode page lists every source the script drew on.
What research never draws on
Never searched
Forums and question-and-answer sites, social media, fan and how-to wikis, video platforms, open publishing platforms, and supplement and health retailers. A forum post is unattributed; a fan wiki can change after it is quoted; a video page captures as page furniture rather than what was said; a retailer sells the thing being discussed.
- reddit.com
- quora.com
- stackexchange.com
- stackoverflow.com
- facebook.com
- instagram.com
- x.com
- twitter.com
- pinterest.com
- fandom.com
- tvtropes.org
- wikihow.com
- youtube.com
- youtu.be
- tiktok.com
- vimeo.com
- medium.com
- substack.com
- amazon.com
- ebay.com
- walmart.com
- iherb.com
- gnc.com
- vitaminshoppe.com
- swansonvitamins.com
- bodybuilding.com
- puritan.com
- lifeextension.com
- nowfoods.com
- thorne.com
Short on purpose. A site not on this list is not vouched for — it is simply not excluded.
Dropped after search
Pages that can never be the evidence for a claim: shop storefronts and their previews, document dumps with no publisher, answer aggregators that summarise other sources, and sites that republish other people’s articles.
- shopifypreview.com
- myshopify.com
- scribd.com
- consensus.app
- custommapposter.com
Applied to web research for evidence episodes. A source a client supplies is never filtered out.
A client’s own documents
A creator can supply sources for an episode — links or pasted text. The episode can be researched from those alone, or from those together with web research. Either way they are captured and stored the same way, and shown on the episode page as provided sources, so a listener can tell the client’s material from what research found.
The fact check reads supplied material exactly as it reads anything else. That shows the script matches what the client supplied; it does not show that the client’s material is right.
What this does not tell you
- Being found by research is not an endorsement. A source is used because it bears on the brief. The check looks for sentences a source contradicts and risky claims no source supports; it does not judge whether a source is right.
- The exclusion lists are short. A site that is not on them is not excluded, and nothing here ranks the sites that remain.
- A page that refuses to be captured cannot be read. If the text of a source could not be stored, the check has nothing from it to compare against.
- The lists change as we find new kinds of page that cannot be evidence. An episode researched before a change was researched under the lists that existed then.
Where this fits
Research is the step before the script is written and before it is checked. By default, the finished script, its hook and its show notes are then compared with exactly the sources on this page by a separate AI model, and anything a source contradicts — or a risky claim no source supports — stops the episode until a person has dealt with it.