1. What this site is
PeptidesDB is an evidence-based reference library for research peptides. It aggregates studies from peer-reviewed databases, attaches AI-generated summaries with explicit source citations, and overlays community-contributed effect data — all in one open, free interface.
It is not a clinic, a pharmacy, or a marketplace. Peptides discussed here are research compounds; nothing on this site is a recommendation for personal use, diagnosis, or treatment.
2. How content is sourced
The Librarian — an automated pipeline running on this site — pulls new study metadata daily from PubMed (NCBI E-Utilities), Semantic Scholar (paper search API), and ClinicalTrials.gov. Each ingested record is deduplicated, classified by evidence type (human / animal / cellular / review), and stored with its original publication date and the date it was added to the library.
3. Role of AI
AI reads the literature here; it does not decide what is true. Machine-written summaries carry an AI analyzed label and a confidence figure wherever they appear.
Not every job gets the same model, and the split is deliberate. Condensing an abstract is a task a small, cheap model does about as well as a large one. Deciding whether a paper is even about the compound it was filed under is not — that one turns on knowing that DSIP means Delta Sleep-Inducing Peptide and not “domain-specific information preservation”, which is a real mistake this library has made. So the work is sorted into two tiers:
- Routine reading (OpenAI GPT-4o mini) — High-volume work where the answer is already in the text: condensing an abstract into a few sentences, pulling a title and journal off a record, drafting prose that a person or a later check will review.
- Judgement calls (OpenAI GPT-4o) — Low-volume decisions that remove or reshape what is published — chiefly deciding whether an indexed paper is really about the compound it was filed under, which is the one call that can unpublish a study.
The stronger model costs roughly sixteen times more per word, which is the whole reason for the split: spending it on the few hundred calls a month where a wrong answer changes a page, rather than on the thousands where it would not.
AI does not author original primary content or replace the original study link — every AI summary is shown alongside a direct citation to the source. Effect scores are AI-derived from the indexed corpus and are never moved by votes or unverified reports.
Recent AI activity is publicly auditable on the Librarian activity feed. A daily and a rolling weekly spend cap gate the budget, checked before each call against the model that call is about to use — so promoting a job to the stronger tier cannot quietly outrun the limit.
4. How we grade evidence
Every peptide carries a letter grade from A to D describing the strength of the evidence base indexed here, and nothing else. A grade is not a recommendation, a safety rating, or a measure of popularity — a compound can be graded A and still be entirely unsuitable for you.
| Grade | Reads as | Requires |
|---|---|---|
| A | Human clinical | An indexed human randomized trial or meta-analysis, or a study registered at trial phase 2, 3, or 4 |
| B | Human evidence | At least one indexed human study, but nothing yet at phase 2+ and no randomized trial or meta-analysis |
| C | Preclinical | Indexed studies exist, but none in humans — animal, cell-based, review, or uncategorized work only |
| D | Emerging | No indexed studies yet — treat the claim as anecdotal |
The grade is computed, not voted on. It is derived from the study rows we hold for that peptide — their evidence type, study type, and registered trial phase — and recomputed automatically as the Librarian ingests new work, so a grade moves the moment the underlying literature does. Nothing about a peptide's popularity, vendor availability, or community sentiment feeds into it.
5. What “summary depth” means
Every study we index carries a number between 0 and 1 that we label summary depth. It is the single most misreadable figure on this site, so it is worth being exact about: it describes how much of the paper we were able to read, not how good the paper is.
When the Librarian summarises a study it is asked how confident it is that its own summary reflects the source. A high figure means a full abstract was available and the summary tracks it closely. A low figure usually means the feed handed us little more than a title, so our summary is thin and you should read the source before relying on it.
| Reads as | Figure | What it tells you |
|---|---|---|
| Full text read | 0.70 – 1.00 | A clear abstract was available; our summary closely tracks the paper. |
| Partial text | 0.40 – 0.69 | Some abstract detail. Check the source for anything decisive. |
| Title only | 0.00 – 0.39 | Little beyond the title reached us. Our summary is thin — the study may be excellent. |
A low figure is a statement about our pipeline, not about the research. Plenty of strong papers — including randomised trials in major journals — arrive with no machine-readable abstract and score at the bottom of this scale. For that reason we do not hide a study because its summary depth is low. It is shown with the label above so you know how much of it we actually read, and the link to the source is always there. Deciding what to publish is done on relevance, evidence type and journal quality instead — the things that describe the study rather than our reading of it.
6. Editorial policy
- Cite the source. Every study card links to its original journal / database record. Cite the primary source in your own work, not PeptidesDB.
- No medical advice. Content is research-reference only. Peptides discussed here are research compounds; nothing on this site is a recommendation for personal use, diagnosis, or treatment.
- No paid placement. Vendors are listed for transparency. Inclusion is not endorsement. We do not accept paid rankings.
- Corrections. If a study summary misrepresents the underlying paper, open an issue on GitHub or email the address below.
7. Contact
Corrections get read first and answered fastest. Email us with the page and, if you have it, the contradicting source.
If you are reporting an error, tell us the specific claim and — where you can — the source that contradicts it. That is the fastest possible route to a fix.
This is not a channel for medical advice, and PeptidesDB does not sell, source, or recommend any compound. For anything about your own health, speak to a clinician who can examine you.