How the index works

The score and its verdict

Every article carries one headline number, 0 to 100, and one plain-English verdict. The number answers a single question: how strongly do the experts agree on the benefits — and how good is the evidence behind that agreement? A high score does not mean a thing is good for you. It means the expert view is settled. A low score means it is not.

• Strong consensus

  Many independent sources — including high-tier research — point the same way.
• Moderate consensus

  Sources broadly agree, backed by some strong evidence or several senior experts.
• Mixed evidence

  The evidence exists, but it does not point firmly in one direction.
• Contested

  The best sources actively disagree on whether it works. We make the disagreement the story.
• Insufficient evidence

  Too few independent sources to draw a conclusion — so we say so, rather than guess.

Broadly, a score of 85 or above earns Strong consensus and 70 or above Moderate — but only when enough independent, high-tier sources are present to justify it. Contested and Insufficient evidence are not low scores — they are structural verdicts. An article is Contested when its strongest sources genuinely conflict, whatever the number says. It is Insufficient when there simply are not enough independent sources to assess. We would rather tell you the evidence is not there than manufacture a number.

How the score is computed

The score is not a vibe. It is computed by an advanced algorithm from a structured reading of every source — the same inputs always produce the same number. It happens in three steps.

1. 1
   An AI reads and classifies every source.

   For each one it records what kind of evidence it is, how authoritative the expert is, and which way it leans — supportive, mixed, or cautionary.
2. 2
   The algorithm combines them, on two separate tracks.

   First: do the experts agree? Each source’s lean is weighted by the authority of who is speaking and the strength of where it appeared. Second, and kept deliberately apart: how strong is the evidence? — how much of the source base is high-tier research rather than opinion or journalism. Strong agreement on a thin evidence base scores only moderately; agreement is not proof.
3. 3
   A named editor reviews and confirms it.

   The algorithm produces a draft score, the full derivation, and a list of things to check. Nothing is published until the responsible editor — Luke Welch, or the domain editor for that subject — has reviewed it.

One more rule, stated plainly: when several sources turn out to be the same underlying study or article, we count it once. A single popular paper cited ten times is one piece of evidence, not ten.

Evidence tiers

Not every source carries equal weight. A meta-analysis and a single-athlete anecdote are both worth knowing about — but they are not the same kind of evidence, and the index does not treat them as such. Every source is graded for the strength of its evidence.

• T1
  Aggregated evidence

  Meta-analyses, systematic reviews and formal position stands — the strongest tier, because they themselves aggregate prior research.
• T2
  Primary research

  Individual randomised controlled trials, large prospective cohort studies, and guidance from credentialed practitioner bodies.
• T3
  Lower-quality primary or secondary

  Smaller observational studies, narrative reviews, and well-cited content from credentialed practitioners.
• T4
  Expert opinion

  Interviews, podcasts and opinion pieces from credentialed authorities — real attribution, but not evidence.
• T5
  Anecdote and journalism

  First-hand accounts, single-athlete stories, and reporting without primary citations. Narrative weight; structurally weak as evidence.
• M
  Manufacturer

  Product pages and commercial or sponsored sources. Counted for context, never toward consensus.

Manufacturer sources are a separate channel, not a ranked tier. They appear in an article’s source list so nothing is hidden — but they never count toward the consensus score, and are never the sole citation for a claim about a product’s effect.

Expert authority

Where a source is published matters; so does who is speaking. A professor who has authored the field’s key reviews is not interchangeable with a practitioner new to the topic, and the index does not pretend otherwise. Every named expert is placed in one of four tiers.

• Leading
  A published researcher or practitioner widely cited as an authority in the field.
• Established
  A credentialed expert with clear domain expertise and a real track record.
• Qualified
  Holds relevant formal credentials, with a limited publication record.
• Uncredentialed
  A journalist, athlete or commentator without formal qualifications in the field — often worth reading, but carrying less weight toward consensus.

Where an expert is a published researcher, we do not guess. We look them up in OpenAlex — a free, public index of more than 250 million scholarly works — and grade their standing from their actual citation record. Where they cannot be matched to a research record — many excellent practitioners, and all journalists, will not be — the classification falls to the AI and our team. The score never treats “we could not find a research record” as “not credible.”

Product tiers

When an article recommends products, each one carries a plain buy-this verdict — a separate judgement from the evidence tiers above.

• Top pick
  Strong evidence, expert consensus, and no conflict of interest. The form we would choose first.
• Reasonable substitute
  Moderate evidence or consensus, or a higher conflict of interest. A sound choice with caveats.
• Best avoided here
  Better suited to a different goal, a lower-quality form, or carrying known issues for this use.

Is this right for you?

The percentage is global: it measures how strong the evidence is, the same for everyone. But the strongest evidence in the world can be beside the point for your goals. So beneath the score we add a short fit chip — “Strong fit for menopause”, “Limited fit for performance”, “Critical if you're deficient”. The fit chip is about relevance to you, not a re-judgement of the evidence. We never move the percentage to flatter you — the same magnesium scores the same whether it suits your goal or not; only the fit chip changes.

There is one exception, and we tell you when it applies. Occasionally a recommendation rests on a genuinely different body of research for one group than for another — creatine for muscle and bone through menopause is studied differently from creatine for raw power. In those rare cases the percentage itself changes, and the reason is spelled out right there in the article, in what the experts agree on and where they disagree. A moved number always comes with its explanation.

How we handle disagreement

When experts disagree, we show the disagreement. We never flatten it into a false consensus. Every article surfaces what the sources broadly agree on and, separately, where they diverge. When the highest-tier sources actively conflict on whether it works, the article is marked Contested — and the disagreement becomes the story rather than a footnote to it. A passing dispute about one detail does not make an article Contested; a real split between strong sources does.

Why we point you to communities

An index is defined by what it points at. We read every expert so you don’t have to — but we are not the only voice worth hearing, and pretending otherwise would be the exact closed-loop move that made health information so hard to trust in the first place. So some of our articles end by pointing you outward, to the peer communities where readers like you compare notes. These communities are not part of our evidence base: they sit downstream of the article as a place to continue the conversation, never upstream as a source we score. We won’t always agree with what gets said there, and we’ll tell you so plainly when that’s the case.

We don’t recommend a community unless it clears a rubric — and we re-check them every quarter, removing any that drift:

• Active and alive — a substantial membership and posts within the last day or two, not an abandoned forum.
• Moderated — a visible, responsive moderation team with clear rules, not a free-for-all.
• On-topic — the bulk of discussion is genuinely about the subject, not off-topic noise or product spam.
• Evidence-tolerant — people reason and cite, and the community is not structurally hostile to mainstream evidence.
• Clean — no record of vote manipulation, brigading, or obvious commercial astroturfing in its top posts.

We’ll also be straight about our own interest here: we read Reddit communities to understand what readers actually struggle with, and linking back to them is part of being a good participant in that ecosystem rather than only taking from it. That’s a real motive, so we’re stating it openly — but a link only ever appears when it’s a community our editor would genuinely point a friend to.

Worked examples

The clearest way to see the algorithm at work is to watch it land on real articles. These three are live in the index today.

• 82
  Moderate consensus

  Creatine for strength and recovery

  Creatine is one of the most-researched supplements there is — and the index cites that research directly: six high-tier studies, meta-analyses among them, sit alongside the expert commentary. High-tier evidence and credentialed experts point the same way, which lifts the score to Moderate consensus, just short of Strong.
• 96
  Strong consensus

  Beetroot juice for endurance

  Beetroot juice for endurance is backed by a wall of evidence — thirteen high-tier meta-analyses and systematic reviews, near-unanimous that dietary nitrate improves endurance performance. Strong research and strong agreement together earn Strong consensus.
• 45
  Contested

  Sodium bicarbonate for runners

  The clearest case of genuine disagreement — and adding more research did not resolve it. High-tier meta-analyses still split: little benefit for continuous running, real gains for short, repeated efforts. When the best sources conflict like this the verdict is Contested whatever the number — more evidence of a real split is still a split.

Who is responsible

Articles are synthesised by AI from expert sources and scored by the algorithm above, then reviewed and confirmed by a named human editor. The AI assembles and structures the evidence; a person is accountable for what is published — the scores, the tiers, the framing, and the choice of which sources to weigh. We disclose the AI’s role on every article: “Synthesised by AI from N expert sources, edited by [the responsible editor].”

Luke Welch is our editor-in-chief and is accountable for the index as a whole. Subjects with a dedicated domain editor are reviewed and signed off by them:

• Luke Welch — Editor-in-chief
• Jayne Welch — Menopause editor
• Brendan Welch — Endurance editor

A second name on some articles

Where a subject earns a registered professional’s eye, we ask one — a dietitian, a doctor, a sports scientist — to read the finished synthesis and confirm it reflects the evidence in their field. Their name, credential and registration number sit beside the editor’s, because you should be able to look the number up with the issuing body — the HCPC, the GMC, the SENr register — and check it yourself. The reviewer didn’t write the article and isn’t its author; they’re the person who put their own professional registration on the line to say it’s sound. We check each registration against the relevant body before we publish the name, and we pay reviewers a fixed fee per article — never a share of anything you buy through us.

Corrections

If something is wrong — a misquoted expert, a mis-tiered source, a finding overtaken by newer evidence — tell us. We read every message, and we correct errors openly rather than quietly. An index is only as trustworthy as its willingness to be wrong in public.

What we are still improving

The index is young, and we would rather be open about its limits than oversell it.

• The index scores the sources we have gathered, not the entire field. A topic where we have found more journalism than journal articles scores more cautiously than the underlying science might deserve — we are continually widening the source base.
• Matching an expert to their research record is not yet perfect: common names and incomplete public records mean some experts are graded conservatively. Our team reviews every case the system is unsure of.
• We are building an independent check — a panel of domain experts scoring a sample of articles by hand — so we can measure how closely the algorithm tracks expert judgement, and publish the result.

Why this exists

Knowing who to trust about your health has never been harder. The problem is documented, and it is getting worse:

• Ofcom finds one in four UK adults encounter false or misleading health information. Source
• The London School of Hygiene & Tropical Medicine warns that dangerous online health information risks a "lost generation". Source
• Mintel names "widespread confusion" as the defining barrier in the UK vitamins and supplements market. Source
• Which? finds supplements at risky doses sold widely across UK marketplaces and the high street. Source

The index exists to give that decision a clear, honest reference point — aggregated expert knowledge, delivered in a way that is actually worth reading.