Do Peptides Actually Work?

Some do, some have never been tested in a human being, and the word "peptides" covers both — which is why the question cannot be answered as asked. A peptide is a short chain of amino acids. That is a structural description, not a category of effect, and it takes in insulin alongside compounds whose entire published record is a handful of rodent experiments. Asking whether peptides work is close to asking whether tablets work.

The version of the question that has an answer is narrower: for this specific compound, for this specific purpose, what has been demonstrated in humans under controlled conditions? For a small number of peptides the answer runs to thousands of trial participants. For most of the compounds sold for laboratory research, the answer is that nobody has looked.

Which peptides are proven to work?

A number of peptide medicines are proven in the ordinary sense of the word: large randomised controlled trials in humans have shown they do what they claim. Insulin is the oldest and least disputed example — a peptide hormone replaced therapeutically, in clinical use for a century, with outcomes nobody argues about.

The GLP-1 receptor agonists are the most heavily studied recent case. In a 68-week randomised, double-blind, placebo-controlled trial in 1,961 adults with overweight or obesity, semaglutide produced a mean change in body weight of −14.9% against −2.4% in the placebo group 4. That is the shape of evidence the word "proven" should be held back for: pre-registered, placebo-controlled, conducted in humans, large enough that chance is not a plausible explanation, and published somewhere other researchers can attack it.

Approved peptide medicines are not a short or exotic list. Broad classes in routine clinical use include:

• Insulin and its engineered analogues, in diabetes care.
• GLP-1 receptor agonists, licensed for type 2 diabetes and for weight management.
• Gonadotropin-releasing hormone analogues, used in oncology and fertility medicine.
• Somatostatin analogues, used in acromegaly and certain neuroendocrine tumours.
• Parathyroid hormone fragments and calcitonin, used in bone medicine.
• Desmopressin, a vasopressin analogue used in diabetes insipidus.

What these share is not chemistry — structurally they have little in common — but process. Each was taken through phased human trials by a sponsor able to fund them, examined by a regulator, and licensed for a stated indication with a known adverse-effect profile. The peptide bond certifies nothing. The trial programme does.

Which peptides are not proven?

Most compounds supplied as research peptides are not proven, in the precise sense that no controlled human trial has been published on them. Their evidence base is usually some mixture of cell-culture work, rodent studies, and a scattering of uncontrolled human observations with no comparison group.

This is not a fringe minority of the catalogue. For a great many of these compounds the whole literature can be read in an afternoon, much of it from one small group of investigators. That is a normal stage for early research to be in. It is not the stage a licensed medicine is at.

It does not follow that nothing is known about them. Receptor binding is often well characterised, mechanisms are plausible, and animal findings can be genuinely striking. What is missing is the step that turns a plausible mechanism into a demonstrated clinical effect in a person — expensive, slow, and the only one that settles anything.

Why do so many peptides have animal data but no human data?

There are two reasons, and only one of them is scientific. The scientific reason is that a large share of promising animal findings do not survive contact with a human trial, so many compounds are tested, fail, and quietly stop being developed. The commercial reason is that trials cost money and somebody has to advance it.

A randomised trial large enough to answer a clinical question is enormously expensive, and that money is normally put up by a company expecting to recover it through a period of exclusive sales. That expectation is what makes the arithmetic work. Remove it and the trial is never commissioned.

A peptide sequence first described decades ago, or one that cannot be protected by a patent, has no sponsor with anything to recover — no matter how interesting the preclinical data look. So the absence of human evidence tells you rather little about the compound and rather a lot about how clinical research is financed.

This cuts both ways. An unpatentable peptide with striking animal results is exactly the kind of thing that stays untested indefinitely rather than being disproved, and exactly the kind of thing that attracts confident claims, because no trial result exists to contradict them. "Never tested" and "suppressed" look identical from the outside, and only one of them is usually true.

If it works in animals, doesn't it probably work in people?

No — animal results predict human results considerably less reliably than most people assume, and this has been measured rather than guessed at. When researchers followed up 76 highly cited animal studies, roughly 37% were replicated in subsequent human randomised trials, about 18% were contradicted by them, and around 45% were never tested in humans at all; only 8 of the 76 interventions went on to be approved for patient use 3.

A systematic review in the BMJ took a different route to the same conclusion, comparing animal experiments with the human trials of the same six interventions. The animal data agreed with the human result for three of them and disagreed for the other three 2. "Coin toss" overstates it slightly, but not by as much as anyone would like.

The reasons are mundane rather than mysterious:

• Laboratory animals are genetically uniform, young, healthy and housed identically. Patients are none of those things.
• The animal "disease" is an induced approximation — a lesion, a knockout, a chemically provoked state — resembling the human condition in some respects and not others.
• Animal experiments often use exposures chosen to make an effect visible, not to model how a treatment would be given to a person.
• Many are small, and randomisation and blinding are applied less consistently than in registered clinical trials.
• The endpoints differ. A biomarker moving in a mouse is not a clinical outcome improving in a patient.

None of this makes animal work useless. It is the correct first step, and almost every approved medicine passed through it. It is simply a filter with a poor positive predictive value: much better at ruling things out than at ruling them in. Treating a rodent result as a preview of a human result inverts what the method is good for.

Why do I keep seeing studies cited then?

Because studies exist for almost everything, and a study existing is not the same as a question being settled. A literature search will return papers for very nearly any compound. What it will not show you is how many similar experiments were run and never written up, or how small and how loosely controlled the published ones were.

The structural version of this argument is well known: research findings are least likely to be true where studies are small, effect sizes are modest, many relationships are tested, analytical choices are flexible, and the field is competitive and fashionable 1. That is close to a description of early-stage peptide research. A small exploratory study in a busy field is exactly the setting in which a published positive result is most likely to be a false positive.

Publication bias sharpens the effect. Experiments that find nothing are less likely to be written up, submitted or accepted, so the visible literature is a filtered sample of the work actually done — filtered towards positive findings. Counting papers tells you almost nothing about the weight of evidence.

Then there is the citation chain. A claim is repeated in a review, the review is cited by a summary, the summary by a blog post, and by the fourth repetition the hedges have fallen off. Follow the references back and you often arrive at one small study, a conference abstract, or a review citing another review.

Does "no evidence it works" mean "evidence it doesn't work"?

No, and conflating the two is the most common single error in this whole area. "No evidence of effect" means the experiment has not been done, or has not been done well enough to tell. "Evidence of no effect" means it was done properly and the compound did not beat its comparison. The first is an empty column; the second is a result.

The practical difference is what could happen next. A compound with no evidence might work or might not, and could still turn out either way if somebody funds the trial. A compound with good evidence of no effect has been asked and has answered. Reading the first as the second dismisses things too early; reading it as a hidden endorsement accepts them far too readily.

The same distinction applies to harm, and this is the half people forget. No published reports of adverse effects is not a demonstration of safety. Safety data come from the same trials that generate efficacy data, and where those trials do not exist there is no safety record either. An untested compound has no record — not a clean one.

How can I tell which category a compound is in?

Go to the human trial literature yourself instead of reading a summary of it — it takes a few minutes and it is usually decisive. In order:

1. Search a biomedical database for the compound's generic name and its synonyms, filtered to randomised controlled trials in humans. Zero results settles it immediately.
2. Where trials exist, check size and control first: how many participants, was allocation randomised, was there a placebo or active comparator, was it blinded.
3. Look at what was measured. A change in a laboratory marker is not a change in an outcome that matters to a patient.
4. Check whether the trial was registered before it started. A pre-registered protocol makes it much harder to pick a favourable analysis after seeing the data.
5. Note who funded and ran it, and whether an independent group has reproduced the result. One unreplicated trial is a lead, not a conclusion.
6. Trace any strong claim back to its primary source. If the trail stops at a review, an abstract or a press release, the claim is unsupported however often it appears.
7. Check regulatory status with a national medicines regulator. Approval is not proof of everything, but it means somebody independent read a full trial dossier.

Most compounds resolve at step one, which is why it goes first. If a search restricted to human randomised trials returns nothing, everything else you have read about that compound rests on animal work, mechanism or anecdote.

So do peptides actually work?

Some do, demonstrably and at scale; for most, nobody knows. That is the honest position, and it is not a cynical one — it is simply what the evidence base looks like when you stop treating a chemical class as though it were a single product with a single answer.

The distinction worth carrying away is between three states rather than two: demonstrated in humans, never tested, and tested and found wanting. Most public argument about peptides collapses the middle state into one of the outer two, depending on which conclusion the speaker prefers. It belongs in neither.

Nothing here says the untested compounds are worthless. Some are the subject of serious ongoing work, and the reason many will never be tested is a funding structure rather than a judgement against them. But "we don't know" remains the accurate description, and a field willing to say so plainly is in far better health than one that fills the gap with confidence it has not earned.