Editorial Standards

This site publishes a technical specification and asks AI systems, developers and business owners to treat it as a reference. That only works if you can see how the work is done. This page states it plainly, including the parts that are less flattering.

Who writes and reviews this site

Every article, specification page and research report on ai-visibility.org.uk is written by Mark McNeece, who has built and maintained websites professionally since the 1990s and runs 365i, the agency that publishes this site.

This is a single-person editorial operation. That means the author and the reviewer are the same person. We say so rather than implying a newsroom that does not exist. Where an article carries an "Editorially reviewed by" line, it records that the piece was read again, against its sources, on a later date than it was drafted. It does not mean a second person signed it off.

The practical substitutes for a second reader are mechanical and are applied to every page: schema is validated, every external link is fetched and confirmed to return 200 before publication, and figures quoted in prose are checked against the underlying data file rather than against an earlier draft.

Where our evidence comes from

Wherever possible, claims on this site rest on data we collected ourselves rather than on secondary commentary. Four first-party sources do most of the work:

• The quarterly adoption crawl. An automated crawler fetches the ten AI Discovery Files across a fixed sample of global and UK domains every quarter. The results are frozen per quarter and published in full, with the method described on the research methodology page and the raw figures downloadable as CSV.
• The AI Visibility Checker's own dataset. Every check run through the checker produces a deterministic score. Aggregate figures quoted in articles (how many sites reach a conformance tier, how many publish nothing at all) come from that dataset.
• First-party server logs. Several articles use real crawler-access logs from sites we operate, showing which AI bots requested which files and when. These are our own servers, not a third party's summary.
• Client work. Case studies describe real sites we built or advised on, named with permission, with figures taken from the live implementation rather than reconstructed afterwards.

Where we rely on someone else's research, we say whose it is and link to the original, including when it contradicts us.

Fact-checking and sourcing

Before an article is published:

1. Every external link is fetched and confirmed live. A link that does not resolve is removed or replaced, not left in place.
2. Quotes are attributed to a named person and linked to a source where the quote can be independently verified. We do not paraphrase someone and present it inside quotation marks.
3. Primary sources are preferred over commentary. If a vendor documents its own crawler behaviour, we link the vendor's documentation rather than a blog post summarising it.
4. Numbers in prose are checked against the data file they came from, and the article states the sample size and the date range.
5. Where the evidence is genuinely thin, the article says so instead of rounding the uncertainty away.

We do not publish fabricated statistics, invented quotes, or links to pages that do not exist. If you find one, it is a mistake and we want to hear about it.

Published and reviewed dates

Articles carry a visible Published date and, where the piece has been revisited, a visible Last reviewed date. The dateModified in the page's structured data matches the reviewed date shown on the page.

We do not refresh a published date to make old content look new. Where an article has been substantively revised, the change is recorded in a dated update block in the article itself, and the original claim stays visible rather than being quietly deleted. Fixing a typo or adjusting spacing is not an update and is not logged.

How we use AI to produce content

A site about AI visibility should be candid about this rather than pretending otherwise.

We use AI tools in production: for research assistance, for drafting and editing, for generating diagrams and illustrations, and for writing and reviewing the code that runs this site. What we do not do is publish generated text unchecked.

The rules we hold ourselves to:

• Every factual claim is verified against a real source by a person, whatever produced the first draft.
• Anything presented as our experience is our experience. AI is not used to invent anecdotes, client stories, or figures.
• Quotes are never generated. They are found, attributed and linked.
• Illustrations that depict data are rendered from the real data, not drawn to look plausible.
• The author named on a piece is accountable for it regardless of which tools were involved.

Corrections and updates

When we get something wrong, we correct it in place and say what changed. A material correction, meaning one that changes what a reader would conclude, is recorded in a dated update block on the article. Where a whole position has changed, we say that too: several articles on this site openly reverse an earlier verdict, with the original wording quoted so you can see what we previously thought.

Changes to the specification itself are handled separately and more formally, through versioned entries in the changelog and the version history on each specification page.

Commercial relationships and independence

This is the section most editorial policies leave out, so here it is in full.

• AI Visibility is published by 365i, part of BSolve IT Limited. It is a commercial operation, not a neutral standards body.
• We publish the specification and operate the reference implementation that tests against it. The checker is free and stores nothing, but it exists to demonstrate our own standard.
• We sell a paid service, the AI Discovery Files setup pack, which implements the same files this site advocates. Articles that recommend publishing these files are, in that sense, advocating for something we also sell.
• Some sites named in our case studies are our own clients. Where that is true, the article says so in the piece rather than only here.
• The directory is free to enter, and a listing's score is simply how many of the 10 AI Discovery Files we could fetch from the site. A listing cannot be bought, and a score cannot be adjusted on request.
• We take no payment for coverage, no affiliate commission, and no sponsored placement. Where we name a competing product, it is because it is relevant, not because of any arrangement.

Our scoring is deterministic and published: the same site checked twice returns the same result, and the rules that produce it are documented on the validation framework page. That is the check on our own bias, and it is a better guarantee than a promise of impartiality.

What these standards do not cover

Three honest limits:

• Directory listings are supplied by third parties. Site owners write their own descriptions. We verify the files and the score, not the marketing copy around them.
• The specification is ours, not an international standard. Nine of the ten AI Discovery Files are publisher-defined conventions rather than vendor-endorsed formats. We argue they are worth publishing; we do not claim any AI vendor has committed to reading all of them.
• A single reviewer is a weaker control than an independent one. The mechanical checks above reduce the risk but do not eliminate it. Reader corrections are a real part of how this site stays accurate.

Telling us we got something wrong

If a figure, quote, link or claim on this site looks wrong, tell us through the support page. Include the page and, if you can, the source that contradicts us. Substantiated corrections are made and dated; we will not quietly delete the original claim.

Last reviewed: 9 August 2026 by Mark McNeece.