Research-literacy siteEducational evidence reviews only — not medical advice, not dosing guidance, not a protocol for human or animal use. Medical disclaimer.

PeptideStacks

AI Use Disclosure

We use AI tools in our editorial workflow. This page explains what we use them for, what we do not use them for, how we verify outputs, and how we keep AI from substituting for primary sources.

Where AI is used

  • Drafting and structuring. Producing first drafts of new pages, suggesting heading structures, summarising long documents, and proposing internal-link patterns that an editor then validates.
  • Citation discovery. Helping locate relevant PubMed entries for an editor to verify against the actual paper. AI is a search aid here, not a source.
  • Risk-keyword review. Identifying risky language ("recommended dose", "before/after", "best stack") across our content corpus for human editorial review — see our editorial policy.
  • Internal workflow. Link checking, metadata generation, sitemap construction, accessibility checks, and orphan-page detection.
  • Adversarial self-review. Running our own content through AI red-teaming prompts to surface claims that outrun the cited evidence.

Where AI is not used

  • AI is not used as a substitute for primary sources. Every citation must be verified against the actual paper or regulatory document before publication.
  • AI is not used to manufacture, fabricate, or hallucinate citations. Citations that cannot be verified are removed; we do not publish "AI-discovered" papers we have not seen.
  • AI-generated claims are not published without human editorial review.
  • AI is not used to generate personalised dosing advice or self-administration instructions — we do not publish that content at all (see editorial policy and responsible information policy).
  • AI is not used to generate evidence-grade assignments for individual stacks or peptides without editor sign-off. Grades drive our A–X methodology and must be assigned by a person, not a model.

Human review workflow

Every page on PeptideStacks is reviewed by a human editor before publication. The process is the same regardless of whether the first draft was AI-assisted:

  1. Source verification. The editor opens every PubMed PMID and DOI link cited on the page and confirms that the paper exists, that the claim attributed to it actually appears in the paper, and that the species / model / sample size are reported accurately.
  2. Regulatory framing check. Any statement about UK regulatory status, MHRA position, or POM advertising is checked against MHRA guidance documents (principally the Blue Guide, the relevant Human Medicines Regulations 2012 provisions, and current MHRA enforcement statements).
  3. Evidence grade pass. The editor reviews the evidence dashboard against the cited material and downgrades — never upgrades — where the underlying evidence does not support the default.
  4. Risk-language scan. The page is run through our risky-keyword scanner (npm run scan:risky) and any HIGH-severity hits are resolved before publication.
  5. Internal-link check. Every internal href is verified to resolve to a live page.

Citation verification cadence

Citations are verified at the point of authoring and re-verified when a page is materially updated. Material updates are logged in our evidence changelog. Where an editor cannot re-verify a citation (e.g. the cited paper is no longer findable), the claim it supports is removed or downgraded rather than left to drift.

When AI-suggested claims fail review

Where an editor is uncertain about an AI-suggested claim, the default is to remove the claim or downgrade it until verified. Common AI failure modes we see and act on:

  • Hallucinated citations. A PubMed PMID that does not exist, or that points to an unrelated paper.
  • Misattributed findings. A real paper cited for a claim it does not support.
  • Animal-to-human elision. Reading a rodent finding as if it were human data without flagging the translation gap.
  • Over-confident regulatory phrasing. Stating a compound is "approved" without specifying jurisdiction or indication.
  • Synthesised numbers. Plausible-looking effect sizes or sample sizes that do not appear in the cited paper.

Why we disclose

Many sites that cover research peptides use AI without disclosure. We disclose because:

  • Readers deserve to know when and how AI is involved.
  • AI hallucination — particularly of citations — is a serious risk in YMYL content (see YMYL).
  • Disclosure makes our editorial process auditable.
  • It anchors a higher standard: if we are open about AI use we cannot quietly relax verification.

Reporting AI errors

If you spot a hallucinated citation, fabricated quote, or AI-generated claim that does not check out against primary sources, please use our corrections policy. We treat these as priority corrections and log resolved cases publicly in the evidence changelog.