Where the studies agree, and where they don’t

We count the trials on each side and explain the difference from the rows themselves: who was studied, how many, for how long, and how each was designed. We never average them into one number.

Consensus finder 1,899 graded rows across 433 supplements No paid links on this page How it decides

680 questions

Agree: 258; Every row on one side
Mostly: 116; Two thirds or more
Split: 306; No clear side

Pick a question

Choose a supplement to open its largest outcome; each result lists that supplement's other outcomes with 2 or more rows. Below, an example: valerian root and insomnia.

Your question

Supplement

Valerian Root and insomnia

4 rows · 1,689 people

Split. 2 rows found an improvement and 2 found no difference.

No difference

Improved

Each block is one evidence row, sized by how many people it enrolled, with its grade inside.

Grade C× 3
No effect× 1

Rows: 4
Trials: 19
People: 1,689; Pooled reviews may count some twice

Where they differ

From the rows

Trial sizeStudy design

The best-graded row, grade C, is a meta-analysis of 1,093 people (2006) and found an improvement. 1 row is graded No effect, our grade for adequately powered trials that found no effect.

The 2 rows that found an improvement were larger: a median of 583 people, against a median of 262 people for the 2 rows that found no difference.

Only the 2 rows that found an improvement include pooled reviews of several trials (1 of 2); the 2 rows that found no difference are single studies.

The rows

2 improved · 2 no difference

Grade CSelf-rated sleep quality (improved or not) · Improved
Meta-analysis, 16 studies · n 1,093 · 2006 · Adults in 16 placebo-controlled trials
Varied across trials · Varied
Across the 6 trials reporting a yes/no outcome, valerian users were 1.8 times as likely to report improved sleep (95% CI 1.2–2.9), but there was evidence of publication bias and most studies had methodological problems.
Row on the Valerian Root page · PubMed 17145239 (opens PubMed)
Grade CSleep quality (PSQI) and actigraphy sleep latency · Improved
RCT · n 72 · 2024 · Indian adults aged 18 to 50 with sleep complaints (mild insomnia symptoms)
200 mg, 1 hour before bedtime · 8 weeks
PSQI total score fell more than with placebo on days 14, 28 and 56; wrist actigraphy showed shorter sleep latency and longer sleep time from day 3. No safety concerns were seen.
Row on the Valerian Root page · PubMed 37899385 (opens PubMed)
Grade CSleep quality (PSQI) during cancer treatment · No difference
RCT · n 119 · 2011 · People undergoing cancer treatment with sleep problems
450 mg 1 hour before bedtime · 8 weeks
No difference from placebo on the primary sleep score (P = 0.70). Exploratory analyses favoured valerian on some fatigue measures.
Row on the Valerian Root page · PubMed 21399726 (opens PubMed)
No effectMinimally important improvement in sleep quality · No difference
RCT · n 405 · 2007 · Norwegian adults aged 18–75 with insomnia recruited via a TV program
600 mg/day of extract (3 tablets nightly, per the PMC full text) · 2 weeks
29% improved on valerian vs 21% on placebo (P = 0.08), not significant. Slightly more valerian users rated their sleep better on a global question (5.5% difference, P = 0.04). No serious adverse events.
Row on the Valerian Root page · PubMed 17940604 (opens PubMed)

Read onRead the full evidenceThe full evidence chapter on its page, with every outcome and its sources.

Every question

680 supplement and outcome pairs with 2 or more graded rows, by health goal. 29 of them include a row that found a worse outcome.

By goal

12 goals

SleepAgree 8 · Mostly 2 · Split 1121 questions

Brain & focusAgree 15 · Mostly 6 · Split 2445 questions

EnergyAgree 7 · Mostly 6 · Split 1023 questions

StressAgree 10 · Mostly 1 · Split 2132 questions

Healthy agingAgree 29 · Mostly 14 · Split 2164 questions

HeartAgree 40 · Mostly 16 · Split 3793 questions

Bones & jointsAgree 22 · Mostly 11 · Split 2659 questions

GutAgree 29 · Mostly 14 · Split 2972 questions

ImmuneAgree 12 · Mostly 9 · Split 2748 questions

PerformanceAgree 23 · Mostly 17 · Split 3373 questions

Hair & skinAgree 16 · Mostly 1 · Split 926 questions

MetabolicAgree 47 · Mostly 19 · Split 58124 questions

How the finder decides

Fixed rules, the same for every question. Nothing is generated and nothing is averaged.

The rows.A question is one supplement and one outcome, as the supplement's own page groups its evidence (the “Outcome” filter over “Every graded row”). It needs 2 or more graded human-trial rows. Each row is one meta-analysis, systematic review or trial, with its grade, design, size, year, population, dose, length and PubMed record.

The verdict. Each row counts once, on one side, and the sides are counted, never pooled:

Improved: the outcome got better for participants. No difference: no effect against the comparison, or a row graded No effect. Worse: the outcome got worse. Mixed: the row reports inconsistent results.
Agree when every row is on the same side.
Mostly when one side (improved, no difference or worse) holds at least two thirds of the rows and also the most people. Mixed rows never lead.
Split otherwise, including a tie.
Any row that found a worse outcome is shown in red and named, whatever the verdict.

Where they differ. When rows sit on two sides, the two largest sides are compared on six fields, and a difference is reported only when it is clean: trial size (medians at least twice apart), design (pooled reviews on one side only), length (no overlap), who was studied (a group, such as women or older adults, named on one side only), when published (no overlap) and the best letter grade. When none of these separates the sides, the result says so; the dose and how each outcome was measured are in the rows.

The counts.People are the rows' own totals added up. A pooled review can include trials that another row also reports, so the total may count some people twice, and the result says when that can happen. Trials count a review's pooled studies when the row records them, else one.

The grades, word for word from our methodology:

Grade AAt least two concordant randomized controlled trials (RCTs), or a meta-analysis of RCTs, with a combined total of 300 or more participants.
Grade BAt least one adequately powered RCT, or a consistent set of smaller RCTs.
Grade CSmall or short RCTs, or trials with mixed results.
Grade DA single small trial, or uncontrolled human data (open-label, before-and-after).
No effectAdequately powered RCTs that found no effect on the outcome.

Where each result leads.One link, chosen by topic, never by who pays us: the product comparison for that outcome, or the supplement's Where to buy section, when the rows lead with an improvement; otherwise the supplement's full evidence chapter.