How a label gets its verdicts

From label to verdict

Six steps, in the order they run.

1. 01

   Paste the label

   Copy the text from the pack or the product page: the ingredient list with amounts, and any claims it makes. A single ingredient name, such as ashwagandha, also works and opens that ingredient’s profile.

2. 02

   Ingredients and claims are found

   Each ingredient is recognised with its amount and form. Each health claim is tied to the ingredients it is about, and claims that say the same thing are grouped. Wording that names nothing measurable, such as “supports a sense of wellness”, is marked “Too vague to check”.

3. 03

   Research is counted by kind of study

   For each ingredient and claim, papers in the Prism research corpus are counted and sorted by kind of study, most reliable first. Findings are grouped by how strong the study was, and the yearly trend is measured against all research published that year. These numbers come straight from the data.

   • Combine many studies. Meta-analyses and systematic reviews. The most reliable kind.
   • Tested in people. Randomised controlled trials, compared with a placebo or another group.
   • Watched groups of people. Observational studies. They can show a link but cannot prove cause.
   • Expert summaries. Narrative reviews of earlier research.
   • Animal or lab only. Preclinical studies. Not tested in people yet.
4. 04

   The most relevant studies are read

   The studies that best match the claim are read one at a time. Each is marked as finding a benefit, finding no benefit, or helping only in some cases. Every reading must quote the paper. A quote is shown only when it matches the paper word for word, and a reading whose quote cannot be found is not counted.

5. 05

   A verdict is given

   Fixed rules turn the counts and the readings into one step on the scale below. Strong study designs count for more than volume, so many weak papers cannot outweigh a few good trials. Because the rules are fixed, the same research gives the same verdict.

   • Likely to workGood studies agree that it helps.
   • Might workSome good studies say it helps, but there are not enough yet to be sure.
   • Mixed resultsSome studies say it helps and others say it does not.
   • Not provenThere are too few good studies to tell yet.
   • Unlikely to workGood studies did not find a benefit.
   • Too vague to checkThe wording does not name anything that studies can measure.
6. 06

   What users say, kept apart

   Separately, Prism social signals show what people report about each ingredient: the doses they mention, how it went and the side effects they name. This is labelled as self-reported experience, and it never changes the verdict.

What the check cannot tell you

Read every verdict with these limits in mind.

• It is not medical advice. It does not replace a doctor, pharmacist or dietitian, and it does not say what you should take.
• Studies test ingredients, not your product. The dose, form or mix in a product can differ from what was studied.
• It does not test what is in the pack. Purity, contamination and whether the amounts match the label are outside what it checks.
• It does not check interactions. Medicines, pregnancy and health conditions can change what is safe. Ask a clinician.
• New or rarely studied ingredients have little research. For them the honest answer is often “Not proven”.
• Readings cover the most relevant studies. Counts cover the whole corpus, but not every paper is read one by one.
• Data can be partly offline. When it is, counts can be lower than usual, and the page says so.
• What users say is anecdote. It shows experience, not proof, and it is never mixed into the verdict.

Informational only, not medical advice. Talk to a clinician, especially if pregnant, on medication, or with liver, thyroid or kidney conditions.