This site has spent most of a summer arguing that a certificate of analysis reports several independent things, and that a single purity percentage answers only one of them. Identity, content, purity, counterion, water, endotoxin, sterility and folding are separate measurements, and a number on one line tells you nothing about a line that isn't there.
That argument has usually been made structurally, from what each instrument is physically capable of detecting. Over the past two years a small body of peer-reviewed work has made it empirically instead, by buying products off the unregulated market and running the full panel on them.
The findings are worth reading carefully, because the most instructive result is not that products failed. It is that on the same vials, different quality axes failed in different directions at once — which is exactly what the structural argument predicts and what a single headline number cannot express.
The one study that measured everything on the same samples
The anchor paper is Ashraf et al., Journal of Medical Internet Research 2024;26:e65440 (PMID 39509151), from the University of Pécs with collaborators at UC San Diego and the Hungarian National Police.
The design combined market surveillance with test purchases. From 1,080 search-engine links the authors identified 317 belonging to online pharmacies, of which 134 led to 59 unique illegal pharmacy sites; the top 30 affiliated domains drew over 4.7 million visits in a single quarter of 2023. Test purchases were placed with six sellers. Three vials arrived. Three prefilled pens never did — non-delivery scams rather than product failures.
The three delivered vials were then put through visual inspection, sterility testing, endotoxin testing and quantitative LC-MS. The results, reported in the abstract:
- Content exceeded the labeled amount by 28.56%–38.69%.
- Measured purity was 7.7%–14.37%, against 99% claimed on the labels.
- No peptide-like impurities were identified.
- No viable microorganisms were detected at the time of testing.
- Endotoxin was present in every sample, at 2.1645–8.9511 EU/mg.
Read those first two lines together. They look contradictory and are not. Content asks how much semaglutide is in the vial relative to the label; purity asks what fraction of the total material is semaglutide. A vial can be over-strength and grossly impure simultaneously, and these were. That is the net peptide content distinction, measured on real market samples rather than argued from first principles.
The third line narrows it further. If the non-peptide mass were failed synthesis — deletion sequences, truncations, oxidised variants — it would show as peptide-like impurities. It didn't. Whatever made up the other 85–92% of the material was something else entirely.
And the last two lines are the endotoxin argument in one sentence: sterile and endotoxin-free are not the same claim. Killing bacteria does not remove lipopolysaccharide, which is heat-stable and passes sterilising filters. Every sample here passed the microbiological test and failed the pyrogen one, at levels well above the 0.1 EU/mg figure that the cell-biology literature has associated with measurable immune-cell effects in culture.
Before purity: is there a peptide in the vial at all
Forensic laboratories, who have been doing this longer than anyone, do not start with purity. They start with presence.
Høj et al., Drug Testing and Analysis 2021;13(7):1457–1463 (PMID 33686802), describe the routine workflow at the Section of Forensic Chemistry in Copenhagen: a colorimetric Bradford assay to screen for any peptide or protein at all, then LC-UV on a C4 column using ratios between peak areas at 220, 254 and 280 nm, then LC-TOF-MS against reference standards, with an immunoassay to confirm HCG.
Of 36 samples received in 2019, 15 contained a listed doping substance, 12 contained substances not on the Danish list, and nine contained no peptides or proteins whatsoever.
Two things transfer. The sequencing of that workflow is a statement about which question comes first, and it is not "how pure." And the LC-UV step reading 220, 254 and 280 nm at once is the practical form of a point made here about detection wavelength: a peptide with no tryptophan or tyrosine is nearly invisible at 280 nm, so the ratio between wavelengths is itself identifying information.
The broader forensic picture is similar. Fabresse et al., Forensic Science International 2021;322:110771 (PMID 33838562), analysed 110 products seized from the French bodybuilding black market between 2016 and 2019. Of 75 pharmaceuticals, 33% were substandard on content, 32% were counterfeit on qualitative formulation, and 19% matched their label.
Wrong molecule is a different failure from wrong amount
The most serious published outcome in this literature involves neither purity nor potency.
Stammers et al., Nederlands Tijdschrift voor Geneeskunde 2026;170:D9116 (PMID 42390457), report two patients without diabetes or obesity who obtained semaglutide outside the regulated system and presented in coma with severe hypoglycaemia, hypokalaemia and hypothermia. The cause was intoxication with synthetic insulin. As the authors note, semaglutide monotherapy is not associated with significant hypoglycaemia in healthy individuals — the presentation did not match the labelled molecule, because the labelled molecule was not what was in the vial.
No purity percentage addresses this. Neither does a matching mass, unless somebody checked which mass. Identity is the first question, and it is answered by a spectrum tied to a named peak, not by a number in a summary table.
Independently, a pharmacovigilance analysis found the matching signal at population scale. Zinzi et al., Frontiers in Pharmacology 2026;17:1805842 (PMID 42137313), retrieved 234 individual case safety reports relating to potential counterfeit semaglutide from EudraVigilance across 2018–2025. 89.3% were classified serious, and disproportionality analysis showed higher reporting frequency of hypoglycaemia, use in an unapproved indication, malaise and drug ineffectiveness compared with non-counterfeit semaglutide.
Two methods, two continents, one converging signal: hypoglycaemia where the labelled pharmacology does not predict it.
What the safety databases add, and what they cannot
The pharmacovigilance layer is genuinely useful and routinely over-read, so it is worth stating the limits plainly.
McCall et al., Expert Opinion on Drug Safety 2026;25(3):581–588 (PMID 40285721), analysed FAERS from 2018 to 2024. Of 81,078 GLP-1 receptor agonist reports, 707 involved compounded products. Compounded formulations carried higher reporting odds ratios for contamination (19.00), compounding or manufacturing issues (8.51), preparation errors (48.92) and hospitalisation (2.35).
Those are striking numbers, and they are reporting odds ratios, not incidence rates. A spontaneous-reporting database records what people chose to report; it has no denominator, and the decision to file a report is itself influenced by knowing a product was compounded. The authors say so. This is the same distinction the site drew between observational and randomised evidence: a disproportionality signal is a hypothesis about a population, generated by a system that was never sampled at random.
The correct reading is that FAERS, EudraVigilance and the Brazilian regulatory surveillance study of Hurtado et al. (Diabetes, Obesity and Metabolism 2026;28(5):4185–4193, PMID 41796096) locate problems worth analysing chemically. They do not quantify them. The chemistry does that.
The literature is about GLP-1 products, and that is itself a finding
Almost everything above concerns semaglutide and tirzepatide. That is not a coincidence — it is where the money, the shortages and the regulatory attention are.
For the rest of the research-peptide shelf, there is no equivalent published survey. The two studies that come closest are the Copenhagen and French forensic papers above, both from 2021, both framed around doping rather than research supply, and neither designed as a systematic market sample.
The 2026 review literature acknowledges the gap rather than filling it. Hailu et al., Cureus 2026;18(6):e110657 (PMID 42437212), survey unregulated peptide use as a digitally mediated phenomenon and name the central issue as "products of uncertain identity, purity, potency, sterility, and safety" — an accurate list, offered as a description of what is not known. DiStefano et al. (Journal of Pharmaceutical Policy and Practice 2024;18(1):2441220, PMID 39776466) mapped 93 direct-to-consumer websites in a single US state and characterised what was advertised, including one product compounded with BPC-157, which the authors note the FDA has determined unsuitable for compounding; that study measured advertising, not vials.
So the honest summary is narrow: the analytical record covers one corner of the market well and the rest of it barely at all. An absent survey is not evidence of a clean shelf, in the same way an absent COA line is not evidence of a passing result — it means the question was not asked.
What the analytical record actually supports
Four transfers, none of which require accepting any claim about any particular seller.
One. The axes are independent and can fail in opposite directions on the same material. Over-strength and low-purity co-occurred; sterile and endotoxin-positive co-occurred.
Two. Identity precedes purity. The Copenhagen workflow screens for presence first; the Dutch case reports describe the consequence when identity is wrong. A purity figure presupposes an answer to a question it does not ask.
Three. Labels were measured against results and diverged by an order of magnitude — 99% claimed, 7.7–14.37% found. A printed specification is a claim, and what a document is a claim about — which lot, sampled how, tested by whom, on what date — is the part that makes it checkable.
Four. The published record is small, recent, geographically clustered and dominated by one drug class. Cite it for what it measured. Statements about the market as a whole run well ahead of it — and checking what a citation actually says is the same skill applied to a different document.
Short FAQ
Does this literature show that most research peptides are substandard? No, and no published study supports that generalisation. The chemical analyses covered small numbers of deliberately selected samples from illegal online pharmacies and seizure caseloads, which are not random samples of anything. What they establish is that these specific failure modes occur and are measurable — not their prevalence.
Why did purity and content disagree so sharply in the Ashraf samples? Because they measure different things. Content is mass of the named peptide against the label; purity is the named peptide as a fraction of everything present. With no peptide-like impurities detected, the balance of the material was not related synthesis by-products. The two figures are multiplicative, not interchangeable — the point argued at length in the net peptide content piece.
Is a sterility result enough on an injectable-format research compound? The measurements here answer that directly: every tested sample was free of viable microorganisms and every tested sample carried detectable endotoxin. They are separate tests with separate methods, and one does not imply the other.
The compounds named here are covered individually in the library; the analytical concepts behind these measurements are set out under quality.
This article is educational and for the laboratory research community. Trulogic Labs products are sold for laboratory and research use only and are not for human consumption.