The sloppification of peptides
Anyone who’s not been living under a rock knows that we’re in full peptide mania.
For the uninitiated, “peptides” are drugs which are made of short chains of amino acids, unlike most drugs which are small molecules. The difference is mostly academic except that you can’t take peptides orally for the most part, as they’re just protein chains, and your stomach is very good at digesting protein and turning it back into amino acids which your body has other uses for!
So peptides have to be either sprayed into your nose, or injected. Hence the (probably apocryphal) stories of “peptide parties” in SF where people go around injecting each other.
Anyway. Trying to buy gray market peptides is now a strange and difficult experience. With the advent of agentic AI coding tools, anyone can make a very pretty, convincing-looking website. Here’s a popular one, Qingdao Sigma, a brand you’ll see recommended on a lot of websites:
They have zillions of products and at incredibly low prices as well. Apparently all lab-tested, they take credit card (usually a good sign, although it’s understandable that companies selling gray market peptides might not want to), and good reviews on TrustPilot. Let’s click on that green banner.
Huh, those reviews do look pretty good. There’s even one right there from someone in Germany, which is great – I’m in Berlin at the moment. (I had hoped they were using geotargeting to dynamically show the user’s country, but it’s actually just a review about Germany.)
But wait a second. I’m not on TrustPilot; I’m still on the QSC website, just on a page that looks like TrustPilot. These guys don’t actually have a page on TrustPilot at all. And these reviews are all AI-generated:
No worries, you think – I’ll just have a look at a review website to see what suppliers have a good track record. Here’s CompoundTalk, a review site I came across while researching for this piece. Looks pretty modern:
But look closer. The trained eye, weary of vibe coded designs, will see Claude Code’s aesthetic fingerprint all over this. First of all, the text is way too small! Every time I’m working with Claude I have to tell it to make the text bigger. Anthropic’s models obviously have pin-sharp eyesight. Some of the text here is 8pt!
A lot of the text here is AI-generated – long-winded pages about GLP-1 side effects and so on – although that’s not uncommon, and those aren’t the bits we care about anyway.
The forum is where the real action is. There’s a “Verified Vendors” subforum, which should have some good information on places to buy peptides. Here’s one thread on QSC, a popular peptide supplier:
Every single post is AI-generated. None of it is real. The whole thing, convincing as it might look, is a mirage.
I asked Claude to do a deeper analysis. It found:
the site is hosted by AlexHost in Moldova, a known offshore/DMCA-ignoring provider
20 vendor pages, all of which are Chinese companies selling raw peptides
every forum thread has exactly 5 posts
985 threads have a date that’s before the domain was registered, 10% of posts were made before their “author” joined
every page links out to an “independent” peptide analysis company called PeptideMeter - but compoundtalk.com and peptidemeter.com were registered 13 minutes apart, at the same registrar, and have the same pair of Cloudflare DNS servers (which get assigned randomly when you sign up, so they’re likely on the same account)
robots.txt which explicitly welcomes LLM crawlers
The robots.txt is the real story. The site itself is just window dressing; it was never really meant to be consumed by humans. It’s slop that’s designed to be ingested by OpenAI, Anthropic, Google for incorporation into their LLMs so that it gets recommended to unwitting users asking about peptides. They advertise a package for peptide providers to get featured, although naturally “sponsorship never affects moderation, Buyer Beware, or review outcomes.”
I was actually pretty impressed by this. The whole thing is a Potemkin village, the creation of which is only possible with LLMs able to pump out huge amounts of convincing-looking garbage.
For now, it’s a huge boon that AI-generated text is pretty obvious to the somewhat-trained eye. It’s a sense worth training, because it’s going to become increasingly important to know if you’re interacting with a human or an AI. AISI’s recent loss of control incident saw a model try to merge malicious code into an open-source repository, faking its identity in the process – a human reviewer caught it. To my eye, you can discern which comments on the pull request were written by bots11. See Security Incident INC-2026-07-28-01, page 11..
Pangram, the AI detector I used extensively here, is also excellent, and going to increasingly become a necessary part of one’s arsenal when interacting with the web, just like an ad blocker.
Stay strapped.