How do you check whether a cited clinical trial exists?

A page selling the peptide KPV carries a table of clinical trials. It has the shape you would want:

Two Phase 1 dose-escalation studies and one Phase 2 efficacy trial, 184 participants in total. The page goes further than the table. It reports placebo-arm side effect rates to the percentage point — nausea in 14% of participants against 16% on placebo, headache 8% against 11%. It reports oral bioavailability of 1.8%, a half-life of 2.1 hours, an immune sub-study in 45 participants with T-cell counts and standardised punch biopsies, endoscopy findings, and bleeding times measured at baseline and week 12.

None of it is in any record we can find.

In its May 2026 review, FDA found no study in which KPV was administered to a person, no clinical pharmacokinetic study, no case report, and no human exposure record by any route. That is the answer this publication already reported in August. The question here is different and more practical: when something cites a trial at you, how do you find out in a couple of minutes whether it happened?

Start with the journal, because that is the checkable part

The third row is the one that settles it, and it settles it fastest.

The page attributes that trial to Inflammatory Bowel Diseases. The journal is real. It has published exactly one paper on KPV. It appeared in 2008, and it is a study of mice — Kannengiesser and colleagues, indexed under Animals, Mice and Disease Models, Animal.

This is the general lesson, not a quirk of this case. A real journal name is the most credible part of a citation, and it is also the part a reader can test in one search. Nothing else on the page can be checked as quickly: you cannot verify a dropout percentage or a plasma concentration from outside, but you can always ask whether the paper is where it is said to be.

Then the registries, and check the right one

A trial has an administrative footprint before it has a result. ClinicalTrials.gov returns zero studies for KPV.

One registry is not enough, and the reason is worth knowing. ClinicalTrials.gov is the largest register and lists studies from most of the world, but no single registry has complete coverage. National and regional registers exist alongside it, and a study can appear in one without appearing in another. A single zero is therefore suggestive, not conclusive.

The second stop is the one that makes this workable for any claim, not just this one. The World Health Organization runs the International Clinical Trials Registry Platform, which aggregates the national and regional registers into a single search. That is how you check a country whose register you could not name — and it is why “search the registry” is bad advice while “search the registers” is good.

It returns No results were found for: KPV.

It took two attempts. The first hit an outage — “temporarily unavailable or there has been an error” — which is worth saying rather than quietly skipping, because a reader repeating this check may meet the same page. Retrying the same day gave the clean negative above.

The third stop is the specific national register, when you already know which one applies. Two of the three claimed trials here are Australian, which makes the choice obvious: the Australian New Zealand Clinical Trials Registry returns Number of records found: 0.

So the shape of the check is three destinations, in order of effort: the largest register, then the WHO portal across all of them, then the national register if the claim names a country. Three registers holding nothing is a different quality of answer from one, and it closes the gap a defender of the page would otherwise aim for.

The most persuasive part of the page is its caution

Here is why none of this is obvious to a reader in a hurry.

The page does not read like marketing. Roughly three-quarters of the way down it carries a section titled The Honest Truth About KPV Safety Evidence. In it, it concedes that the research is narrow in scope. It states that three trials, 184 participants and twelve weeks of maximum duration are the entirety of the published human evidence. It lists who has not been studied: no pregnant women, no children, no adults over seventy, nobody with severe liver or kidney impairment. It warns the reader that anyone describing KPV as a proven therapy is overstating what the research shows.

Every number it responsibly qualifies is unsupported.

That is what makes it work. Hedging is the signal most of us have learned to read as honesty, and a source volunteering its own limitations is performing the behaviour of a careful source. The performance survives even when there is nothing underneath it.

Which produces a rule that feels wrong and is not: a source that admits its limits has not earned anything yet. Check a figure anyway.

Why the check matters more than it used to

A page like this used to have to win a search result to reach anyone. This one does not win them. On 27 August 2026 it was absent from Google’s first ten results for KPV safety studies, a query it is written to answer, and FDA’s own compounding page ranked third. Bing both indexes the page and summarised it.

It reached readers anyway, because summaries now stand between the reader and the pages.

Asked on 27 August 2026 how many people had joined human safety trials of KPV, Bing’s Copilot Search returned the three trials above under the heading Evidence Summary, with the same participant counts. It added a figure that appears in none of its sources — “fewer than 500 cumulative human participants” — and closed with a section headed Clarifying Caveats noting that no Phase III or long-term trials have been conducted. That last statement is true, and it is the most misleading sentence in the answer, because it implies Phase I and Phase II trials were.

The hedge travelled with the numbers.

Two things stop this being a story about unreliable machines. The first is that Bing listed its sources, and at least four of the nine state, in the snippets Bing itself displayed, that no human trial of KPV exists at all: “Strong Mouse Data, No Human Trials”; “No registered clinical trials exist specifically for KPV”; “no registered interventional clinical trials”; “No human clinical trials of any phase”. A seventh source supplied numbers too, and they were different numbers — 185 research papers and 50 clinical trials. The correct information was not missing. It was listed alongside.

The second is that the other surface got it right. Asked the same question in the same words the same day, Google’s AI Mode answered that there are zero participants in official human safety trials for KPV, cited FDA, and asked the reader where they had seen the trial mentioned.

Both had the same web. One treated a commercial page as its evidence base and one did not.

Two limits on that comparison. The Google reading was taken once; the Bing result was reproduced on two days under two different wordings. And no other assistant was tested, so nothing here says where any of the rest would land.

Why the unsupported version tends to be the more detailed one

There is a structural reason the wrong answer looked more like an answer, and it explains why this problem will not go away on its own.

A source stating that no human trials exist offers a summariser almost nothing. It is a single negative fact. There is no table to build from it, no figure to place in a sentence, no structure to reproduce.

A source stating 32 plus 62 plus 90 offers everything: institutions, years, phases, routes, durations, dose ranges, dropout percentages, adverse event rates. It fills the shape of an answer. When the task is to summarise what is known, a page behaving as though a great deal is known has more to contribute than a page reporting that nothing is.

An accurate answer here is an absence, and absences have no details.

This is an exception, and it should be read as one

It would be easy to file this under machines being unreliable. Our own measurements do not support that.

On the same day we checked seven peptide entities on the same surface against the primary source behind each of our articles. Six were right. The distinctions held were the ones this field gets wrong most often: an analogue separated from the compound it derives from, a fragment from the full protein, a mechanism from an efficacy claim, an advisory committee vote from an approval.

One in seven. The subject here is not a search engine. It is a single commercial page carrying a clinical record that the regulator, the registers we searched and the journal it cites do not contain — and the fact that a reader now needs a way to catch that before it reaches them wearing someone else’s authority.

The check, in order

1. Look up the journal. If a claim names one, the paper is either there or it is not, and that takes one search. A real journal name is the most credible part of a false citation and the easiest part to test.
2. Search the registers, plural. No single one is complete. Start with ClinicalTrials.gov, then the WHO’s ICTRP portal, which searches the national registers together so you do not have to know which one applies. If the claim names a country, finish at that country’s own register — here, ANZCTR for the two Australian trials. One clean-looking negative proves less than it looks.
3. Ask what the number implies. A Phase 2 trial with 90 participants needs ethics approval and a sponsor, and in most countries a registration before the first participant joins. If a figure that specific has no paperwork anywhere you look, the specificity is the problem, not the evidence.
4. Treat volunteered caution as neutral. It is a writing style, not a credential.

Primary sources

• Kannengiesser et al., the only KPV paper in Inflammatory Bowel Diseases, 2008, murine models
• FDA evaluation of KPV free base and KPV acetate
• FDA, Category 2 bulk substances: no human exposure data identified
• ClinicalTrials.gov search for KPV
• WHO ICTRP search portal, KPV
• ANZCTR search for KPV
• LifeScore Edge, Does KPV have human evidence?

The commercial page is described rather than named. This article makes no claim about whether its figures were invented deliberately or generated carelessly, only that the regulator, the registries and the journal it cites do not contain them. Both AI answers were captured on 27 August 2026 and are dated observations of products that change.

Independent editorial summary. The authors and organizations are not affiliated with LifeScore.