How much evidence is there, really?

How this tool works · sources read October 3, 2026 · Grey Peptides Editorial Board

What the explorer counts

Every peptide is marketed with the same phrase: 'backed by research'. The Evidence Explorer puts a number on that phrase. For each of the 348 compounds in our encyclopedia it asks PubMed, the US National Library of Medicine's index of biomedical literature, how many records mention the compound, and then how many of those are indexed as randomized controlled trials, as clinical trials of any kind, as observational studies and as reviews. It also asks how many are indexed as involving humans and how many as involving animals but not humans. Separately, it asks ClinicalTrials.gov how many registered studies list the compound as an intervention, split by phase ^{1 2}.

The result is an evidence profile: not a judgment of whether a compound works, but a picture of what kind of research exists. A drug that has been through regulatory approval looks completely different from a research chemical, and the difference is visible at a glance.

What the contrast looks like

Compare semaglutide with BPC-157. Semaglutide has 5,724 PubMed records, 309 of them indexed as randomized controlled trials, and 182 registered phase 3 studies on ClinicalTrials.gov. BPC-157 has 224 records, 0 randomized trials, 105 records indexed as animal-only research and 3 registered studies of any phase. Both are called peptides with research behind them. One has been tested in thousands of people in controlled trials; the other has been tested mainly in rats.

Across all 348 compounds, 64 have no PubMed record indexed as a randomized controlled trial, 14 have no PubMed record at all under the search we used, 84 have no registered study on ClinicalTrials.gov, and 53 have more animal-only records than records indexed as involving humans.

How to read a profile

Read the bars from the top. Randomized trials are the strongest single kind of evidence that a treatment causes an effect, because chance decides who gets the treatment and who gets the comparison. Clinical trials is a wider PubMed category that includes randomized trials and also uncontrolled and early-phase studies, so it is always at least as large. Observational studies follow people who chose or were given a treatment, which can show associations but struggles to prove cause. Reviews summarise other studies; many reviews with few trials underneath them is a common sign of a field that talks more than it tests.

The human and animal only bars come from PubMed's indexing of who or what was studied. 'Human' covers any record indexed with the Humans heading, which includes trials but also case reports, surveys and some laboratory studies of human material, so it overstates how much has been tested in living people. 'Animal only' counts records indexed for animals and not humans: the mouse and rat studies that most research peptides rest on. The registered trials bars show how far a compound has travelled through clinical development: phase 1 tests safety in small groups, phase 2 looks for effect and dose, phase 3 is the large confirmatory trial that approval usually needs, and phase 4 studies follow an approved drug.

Why every query is printed

A count is only as good as its search. Under every profile, the explorer prints the exact PubMed query and the ClinicalTrials.gov intervention term it used, and the date it was run. Anyone can paste the query into PubMed and get the same number on the same day, or a slightly larger one later as new papers are indexed. Most queries search the compound's name and its aliases in titles and abstracts; for compounds whose names are common words or ambiguous abbreviations, the search is narrower, and the profile says which basis was used.

Printing the query also exposes its limits. Some searches catch more than the compound: the search for TB-500 includes 'thymosin beta 4', the full protein from which the TB-500 fragment was taken, so its {_f(T['pubmed']['rct'])} randomized trials are trials of thymosin beta-4, not of the fragment sold online. Our article on why TB-500 is not thymosin beta-4 explains the difference. Hormones with very large literatures, such as hCG, count research on the natural hormone as well as the drug. Where a name is shared or broad, the counts are an upper bound.

What a count cannot tell you

Counting is deliberately crude. It treats a large, well-run phase 3 trial and a small, poorly run one as one randomized trial each. It counts papers, so one trial reported in five publications counts five times. It depends on PubMed's indexing, which is done by people and lags behind publication, so recent papers may not yet carry a publication type. It counts registered trials whether or not they finished or reported results. And it cannot tell a positive result from a negative one: a compound with several randomized trials that all failed has a fuller profile than one never tested at all.

That is why each compound's encyclopedia entry, linked from its name, carries an evidence grade based on reading the studies, and study cards that say what each one found. Use the explorer to see the shape of the evidence and to spot claims that outrun it, and use the entry to see what the evidence actually says. Our methodology page sets out how grades are assigned.

Using it well

Three habits make the explorer useful. First, compare a research peptide with an approved drug in the same family, such as BPC-157 with a GLP-1 drug, or ipamorelin with tesamorelin, to calibrate what 'well studied' looks like. Second, look at the ratio of reviews to trials: a long list of reviews built on a handful of trials, or none, means the same few findings are being repeated. Third, check the animal-only bar before believing any claim about what a peptide does in people. When almost all of the research is in rodents, every human claim is an extrapolation.

The counts are refreshed when the encyclopedia is reviewed, and the read date on each profile tells you how current it is. If you find a compound whose search plainly misses its literature, the printed query is the place to start, and we would rather fix the query than keep a misleading number.

Sources

1. National Library of Medicine. PubMed, searched through the E-utilities API on October 3, 2026, with the queries printed under each profile.
2. ClinicalTrials.gov API v2, searched October 3, 2026, by intervention term and phase.

This explainer describes how the tool works and what its numbers rest on. It is not medical advice and does not recommend any dose or product.