How we rate evidence

The four ratings

A rating describes the state of the human research literature — not whether a study exists, and not how interesting the food is. Plenty of foods with fascinating laboratory research sit at limited or insufficient here, and that is correct.

Strong evidence

Multiple well-conducted human trials, usually summarised by systematic reviews or meta-analyses, point the same way. Remaining disagreement is about size of effect rather than whether an effect exists.

Moderate evidence

Human research exists and generally trends in one direction, but the body of work is smaller, the trials are shorter or smaller, results are mixed, or the reviews themselves flag meaningful heterogeneity.

Limited evidence

Findings are preliminary: few or small human trials, surrogate outcomes such as blood markers rather than health events, or research that is largely mechanistic, animal or in vitro. Interesting, not established.

Insufficient evidence

There is no dependable body of human research behind this association, or Know2Eat has not yet attached a source to it. Treat it as an open question, not a finding.

The ceiling rule

An editor proposes a rating. The software then caps it against the sources actually attached to the claim, and the cap can only ever lower the rating, never raise it. This is enforced in one function that every page and every admin save runs through, so an optimistic rating cannot survive to publication:

• No sources attached at all → insufficient
• Only animal, laboratory or unclassified research → limited
• At least one human study → moderate
• Two or more systematic reviews or meta-analyses → strong

Reaching the ceiling does not automatically award the rating. A claim with three meta-analyses behind it can still be rated insufficient if those reviews conclude the effect is not there — and several on this site are. The ceiling is a maximum, not a target.

The language rules

Claim copy is checked automatically before it can be published, in the seed pipeline and again in the editor. These phrases are rejected outright — not flagged for review, rejected:

• cures
• heals
• treats
• prevents
• detoxes / ify / ifies / ification
• flushes out toxins
• miracle
• superfood
• proven to
• guarantee[sd]
• boosts your immune system
• kills cancer
• fights off cancer / disease
• reverses
• alkalize / es / ing

What you will see instead: may support, has been studied for, contains compounds associated with, research suggests, evidence is limited for. The hedging is not legal throat-clearing. It is the accurate description of what nutrition research can and cannot establish.

What this site cannot do

• It is not medical advice. Nothing here can account for your history, your medications or your individual risk. Interactions listed on a food page are a prompt to ask a pharmacist or clinician, not a substitute for asking one.
• Ratings are about populations, not you. A meta-analysis reports an average effect. Yours may be larger, smaller, or absent.
• Extract ≠ food. A great deal of nutrition research uses concentrated extracts at doses far above culinary intake. Where that is the case, the claim says so — and it is usually the single most important caveat on the page.
• The database is incomplete. An empty section means we have not recorded anything dependable, not that nothing exists. Gaps are shown as gaps.