AI Use Policy

Language models are used in producing this site. Rather than leave that to be inferred, every guide carries a content-method note stating how it was made, and this page explains what those notes mean.

The labels used on pages

Each guide is marked with one of four methods:

• Human-written. Drafted by the named author without machine assistance.
• Human-reviewed, AI-assisted. Machine-assisted draft, checked line by line by the named author against its sources.
• Editorial synthesis. Assembled from identified public sources, with the sources selected and verified deliberately. This is the label on the current guides.
• First-hand testing. Based on something the author did or measured directly.

The schema behind the site enforces part of this: a page cannot be published as an editorial synthesis without at least one source, and any method other than human-written requires a written disclosure. These are build-time constraints, not conventions.

What is currently true of these guides

The published guides are editorial syntheses of public trade, customs and product-safety guidance. Their content-method notes say that, and they add that the page is pending named-author review.

That notice is accurate rather than decorative. It means:

• The sources were chosen deliberately and each was loaded and confirmed before use
• The structure, the judgements and the omissions are editorial decisions
• The named author has not yet checked each page line by line against those sources

It stays on a page until that review happens. Removing it earlier would make the byline claim more than the process has delivered, which is the specific failure this policy exists to prevent.

What disclosure does not do

Disclosing a method does not make the content correct. A page that is transparent about being machine-assisted and is also wrong is still wrong.

Disclosure is therefore paired with things that do affect quality: primary sources checked before publication, claims matched to the source that supports them, reasoning marked as reasoning, and a corrections process that anyone can use.

Nor does disclosure license volume. Pages here are opened only where they answer a question no existing page answers, and near-duplicates are merged rather than published — a constraint that matters more, not less, when drafting is cheap.

What is never machine-generated

• Author identity and credentials. Experience is stated as supplied by the author and nothing is invented to fill a gap. Where a field cannot be filled truthfully it is left empty.
• Sources. Citations are not produced from memory. Each is loaded and confirmed to resolve to the page being cited before it is used.
• First-hand claims. Nothing on this site claims direct experience of a supplier, transaction or inspection that did not happen.

AI crawlers

The two kinds of AI crawling are treated differently in robots.txt, and the distinction is intentional:

• Search and answer crawlers are allowed. If a page here answers a question inside an AI search result, that is the site working as intended, and citation sends readers back.
• Training crawlers are disallowed.

This is a publishing decision rather than a technical one, and it can be revisited. It is recorded here so the reasoning is visible rather than buried in a configuration file.

Reporting a problem

If a page reads as though it claims experience it does not have, or a source does not support the sentence it sits beside, that is a reportable error under the fact-checking policy.