We index what experts say. We don't grade it.

Our posture: an index, not an arbiter

Cited surfaces facts: what a person recommended, in what source, at what moment, with what stated reasoning — alongside their credentials, any publicly known conflicts of interest, and what other credible voices said about the same thing.

We do not rank experts by credibility, assign credibility scores, or adjudicate who is right. We let the reader weigh. This is a deliberate choice: the missing thing in this space is a structured, factual index of what experts actually recommend — and an index is more defensible, and more useful, than a referee.

What we index

Recommendations from experts in public podcasts, talks, and long-form interviews. A statement is recorded when it's specific enough that a reader could act on it — whether it's affirmative ("I take X for Y"), against ("I wouldn't use X"), or conditional ("if you struggle with X, consider Y").

Expert voices only. On Q&A-format episodes a host often reads a listener's written-in question. That listener isn't an indexed expert, so their aside isn't recorded as a recommendation — only the expert's answer is. This rule is enforced in code before anything is saved.

What we don't index

• Vague gestures ("supplements can help with sleep") and off-the-cuff mentions without context.
• Anything the speaker retracts within the same source.
• Prescription drug dosing recorded as if it were guidance.
• Interventions for diagnosed psychiatric conditions.
• Pediatric-specific advice, advice involving activities illegal in our jurisdiction, and advice given to a named individual.

How we classify a recommendation

Every recommendation carries four structured dimensions, so we can represent the full nuance of what was said without judging the content of it:

• Stance — recommends in favor, advises against, or neutral discussion.
• Strength — from an offhand mention, to a clear recommendation, to an emphatic endorsement or warning.
• Certainty — an explicit declaration, a hedged "you might consider," or framed strictly as the speaker's own personal use.
• Type — an organic mention versus a sponsored read, a guest's recommendation, or a formal endorsement.

When the type is genuinely ambiguous — say, a host is sponsored by a brand and also uses the product unprompted — we default to the more conservative reading.

Conflicts of interest & sponsored reads

When a recommendation is part of a paid sponsor read, we mark it. Where a financial relationship is publicly knowable — sponsorships, advisory roles, equity in a company whose product is being recommended — we aim to surface it too.

We disclose what is public; we do not investigate or speculate. And we don't claim that a financial tie makes a recommendation biased — we state that the relationship exists, and let you weigh it. When in doubt, we disclose.

Quotes and sources

We paraphrase by default and keep direct quotes short, and every recommendation links back to its source with a timestamp. We don't republish transcripts — we point you to the original publisher so you can hear the full context for yourself.

Disagreement

When credentialed experts are substantively in tension on the same item — different doses, opposing conclusions — we surface that disagreement without declaring a winner. The absence of a flagged disagreement does not mean the experts agree; it means we haven't surfaced one.

Safety

We're neutral on opinions but active on safety. Where official guidance from a regulator or professional body (FDA, NIH, professional societies) is directly relevant to a recommendation, our aim is to present it alongside — not to adjudicate whether the expert agrees with it. We index additional sources; we don't referee them.

Accuracy & verification

Cited is in preview. Today, recommendations are auto-extracted by AI and are not yet human-verified. We do not claim the extraction is perfect — AI has known failure modes, including misattribution, misquotation, and transposed doses.

Two things follow from that. First, until a recommendation is human-verified it is labelled as auto-extracted, and human verification is the layer we're building next. Second, every card is designed so you don't have to trust us blindly — read the quote and follow the timestamp to the source. If you find an error, our corrections process is how we fix it.

What we deliberately don't do

• We don't assign credibility scores to people.
• We don't adjudicate truth claims.
• We don't summarize or characterize public criticism of a person — where it exists, we link to it.
• We don't declare consensus.

Our test: when we're tempted to take a position, we ask whether we can instead surface a source that already takes it. If we can, we index that source. If we can't, we probably shouldn't be making the claim.