Sources and evidence standards

What counts as a source here, what does not, and how we check that a link says what our sentence says it does.

The short version

Every health statement in a guide points to a source you can open. We prefer the strongest kind of study that exists for the question. We say so when only weak evidence exists, and we count a maker's own page as a claim, not as proof.

Last reviewed September 19, 2026.

What we count as a source

Systematic reviews and meta-analyses of randomized trials
We use these first, to say whether an ingredient does what is claimed. Examples: Gholami et al., 2022 (melatonin and sleep quality), Hepsomali et al., 2020 (oral GABA) and Bulman et al., 2025 (L-theanine and sleep).
Other reviews and single trials
We use these for mechanisms, for safety, and for questions that no pooled analysis covers. Examples: Rao, Ozeki and Juneja, 2015 (L-theanine) and Maffei, 2020 (5-HTP toxicology).
Government health agencies
We use these for side effects, interactions, definitions and who should ask a provider first. Examples: the NCCIH fact sheet on melatonin, the NHLBI pages on insomnia and the NIGMS fact sheet on circadian rhythms.
Clinical practice guidelines
We use these to say what clinicians are advised to do and how strong the recommendation is. Examples: the American Academy of Sleep Medicine guideline on behavioral therapies for chronic insomnia and the American College of Physicians guideline on chronic insomnia.
Regulations and regulator guidance
We use these for what a label must show. Examples: 21 CFR 101.36 and the FDA Dietary Supplement Labeling Guide.
The maker's own website and label
We use these for what the maker claims, what the label says, and the price and terms. Each is dated when we read it. They tell us what the maker says. They do not tell us it is true.

What we do not count as evidence

  • Customer testimonials, star ratings and review counts. A count on a sales page is a claim we cannot check.
  • Press releases and paid placements, including ones that appear on well-known news sites.
  • Another site's review of the same product.
  • Forum posts, social media and influencer videos.
  • AI-generated answers and summaries. We open the study or page they point to. A claim that shows up only in a summary is left out.
  • Studies on animals or cells, when the claim is about people.

How we cite

  • A study gets its authors, title, journal, year and a link to its PubMed entry, where the abstract is free to read.
  • A web page gets its address and the date we read it, because pages change.
  • A regulation gets the section number.
  • Links to studies and agencies are never affiliate links and earn us nothing. The only affiliate link in a guide sits in its own labeled box, with the disclosure next to the button, and is marked as sponsored.

How we check a source

Before a guide is published, each sentence that cites a source is compared with the source itself:

  • The number of trials or people we give matches the source.
  • What we say was measured is what was measured.
  • The authors' own conclusion is reported as they worded it, including when they concluded that the evidence is limited.
  • When the label discloses a dose, we compare it with the dose the study used.
  • We look at who wrote and who paid for the source. When the authors work for a company that sells the ingredient, or the company funded the work, the guide says so.

When a maker lists studies of its own, they stay the maker's claims until we have read them. A guide that has not checked a reference says so.

When the evidence is weak or mixed

We do not pick the study that suits the story. When a good review finds little, we report that, even for an ingredient in a product we link to. The GABA section of our Yu Sleep review is an example. It reports a 2020 systematic review of 14 placebo-controlled human trials that found "limited evidence for stress and very limited evidence for sleep benefits," and it says GABA is not a reason to weigh either way.

We show the strength of the evidence with four words, defined on the research methodology page. Where the only source is the maker, the word is "Not verified."

What sources cannot tell you

  • A study describes a group of people. It does not tell you what will happen to you.
  • Our searches are not systematic reviews of their own. A study we did not find could change the picture.
  • We do not re-run anyone's analysis. We report what the authors report, and we can misread it.
  • Published research can be wrong, contested or out of date. Every guide shows the date we last reviewed it.
  • We work from sources published in English.

Corrections and suggestions

If a source does not say what we say it does, or you know of a better one, write to editors@sleepanchors.com. We correct the guide and change its review date.

Questions

Are the links to studies affiliate links?

No. They go to PubMed, to government agencies or to the regulation itself. The affiliate disclosure explains the one place a guide does carry an affiliate link.

Why link to PubMed and not to the journal?

PubMed is free and shows the abstract and the study's identifier. Some full texts sit behind a paywall.

Do you count a maker's cited studies as evidence?

Not until we have read them. Until then they are the maker's claims, and we label them that way.