AI disclosure

That paragraph is the whole disclosure. The rest of this page is detail for anyone who wants it.

What AI is used for

• Research and drafting

  AI tools are used to survey literature, draft explanations of mechanisms, and structure entries. No claim reaches a page on the strength of a model having asserted it.

• Illustration

  The plates and diagrams on this site are AI-generated images. They are decorative and explanatory; they are never evidence, and no verdict depends on one.

• Code

  The site itself — the templates, the search, the checks — is largely AI-written. This has no bearing on whether the health content is correct, and is mentioned only for completeness.

What a human does, specifically

The check that matters is the sourcing one. For every mechanism claim on this site, a named human traced the claim to a primary source, opened that source, and confirmed it says what the entry says it says. Where no such source exists, the entry says the evidence is thin instead of asserting something plausible.

That check is backed by a machine check rather than left as a promise: a committed script re-queries the PubMed API for every PubMed ID on the site and fails if a record stops resolving, or if its title, journal or year drifts from what is printed here. Every citation passed on 15 August 2026. The script is run before publishing rather than automatically on every deploy — stating that precisely, because overstating a verification claim would be the same category of error this page exists to prevent. Language models are known to produce citations that look correct and do not exist. Treating that as a live risk to be tested for, rather than a thing that happens to other people, is the single most important safeguard on this site.

What nobody checks

“AI wrote it” is not a defence

This disclosure exists so you know how the site is made. It does not transfer responsibility for anything on it. If a health claim here is wrong, it is wrong because a person published it, and that person is named in the editorial policy. This page sits next to the review policy, not instead of it.

Labelling AI-generated images, and the law

The illustrations on this site are AI-generated and are labelled as such wherever they appear. We would do this regardless of the legal position, because it is cheap and it is true — but the legal position was checked rather than assumed, and it is worth setting out accurately.

The EU AI Act’s transparency rules — Article 50 of Regulation (EU) 2024/1689 — became applicable on 2 August 2026 and are current law rather than a future obligation. Two parts are relevant to a site like this one:

• —Marking synthetic output is an obligation on the providers of the generative systems, not on a site that uses them. It is their machine-readable marking, not ours.
• —Disclosing AI-generated text published to inform the public on matters of public interest is an obligation on deployers — but it carries an exemption where the content has undergone human review and a named person holds editorial responsibility. On this site a named person does hold that responsibility, so the exemption very likely applies and this disclosure is very likely not required of us. We publish it anyway. A reader deciding whether to trust health information should not have to rely on us having qualified for an exemption.

That is our reading of the regulation as it applies to this site, and it is not legal advice. Where the rules and this page disagree, the rules win and this page gets corrected.

Why this is published rather than minimised

Search engines do not penalise AI-assisted content; they penalise unhelpful content, and they always have. The reason to disclose is not search-related. It is that this site’s entire proposition is that it is the one page on these questions not written by someone selling the answer — and a site making that claim while being vague about how it is produced would be making it dishonestly.

Last updated 15 August 2026. Changes to this policy are recorded in the changelog.