Editorial policy

SEO and AI-search writing has a credibility problem: confident claims, no sources, no dates, and no way to tell a published standard from something someone tried once. This page sets out the rules we write to, so you can check whether an article follows them.

The same evidence discipline governs the product itself. What the software reports and what this site publishes are held to one standard, described on the methodology page.

Research methodology

An article starts from primary sources, not from other articles about them. Before a factual claim about a search engine, an AI provider, a crawler, a specification or WordPress is written, the current first-party documentation for it is fetched and read. If the documentation does not say the thing, the thing is not written — or it is written as an inference and labelled as one.

Nothing is quoted at length. Sources are paraphrased and then analysed, because reproducing a provider's own wording adds nothing a reader could not get from the provider.

Where sources conflict, the conflict is reported rather than resolved by picking the more convenient one. Where a question has no public answer, the article says the answer is not public.

Source hierarchy

Sources are ranked, and a lower tier never overrides a higher one:

1. Published specifications and web standards — RFCs, W3C recommendations, the Robots Exclusion Protocol, Schema.org vocabulary.
2. Current first-party provider documentation — Google Search Central, Microsoft Bing Webmaster documentation, OpenAI, Anthropic and Perplexity developer and crawler documentation, the WordPress developer handbooks.
3. Reproducible published measurement from a credible source, where the method is described well enough to be repeated and the results are consistent across studies.
4. Field practice — widely adopted techniques with a plausible mechanism but no provider confirmation.
5. Our own reasoning, clearly identified as reasoning.

Marketing material from tool vendors, including material about competing tools, is not a source for a factual claim. Neither is another blog's summary of a provider's documentation; if a claim matters, we go to the documentation.

The six evidence tiers

Every recommendation on this site carries a tier, in words or in a table:

An EXPERIMENTAL or HYPOTHESIS technique is never described as a ranking factor, a signal, or a requirement. Google states there are no additional requirements or special optimisations for its AI features beyond ordinary SEO, and we do not contradict that without citing a source that does.

Certain distinctions survive every article regardless of how much cleaner the sentence would read without them: crawled is not retrieved, retrieved is not cited, cited is not recommended. An llms.txt file existing is not evidence of ingestion. Structured data is not a citation. An API response is not what a person sees in a consumer chat product. One query is not a ranking. The phrases "LLM ranking" and "position" are not used for generative answers, because generative answers do not have positions.

Verification and dating

Every article carries two dates in its front matter and on the page:

• reviewed — when a person last read the article end to end and confirmed it still says what it should.
• sources_verified — when the external sources it relies on were last fetched and checked against what the article claims.

Those are separate on purpose. An article can be well written and internally consistent while resting on documentation a provider quietly changed. The sources-verified date tells you how stale the risk is.

Every article ends with an official sources list — the actual URLs, with the organisation named, so you can check the claim rather than trusting the summary. A URL is never written from memory. If we did not fetch it, it does not appear.

Corrections

We correct rather than quietly edit.

If a factual claim is wrong, it is fixed as soon as it is confirmed. A correction that changes the meaning of a recommendation is noted in the article, saying what was wrong and what it now says. Typographical and clarity edits are made without a note. Nothing is deleted to make a past mistake disappear, and no correction is disguised as a routine update.

If you find an error, tell us. The contact page has the channels; include the article, the specific sentence, and the source that shows it is wrong. A correction supported by a primary source will be made quickly.

Updates and review

This subject changes underneath its own documentation. Provider crawler documentation is revised without notice, model behaviour shifts, and proposals like llms.txt gain and lose adoption.

Articles are reviewed on a schedule and out of schedule when something they depend on changes — a crawler user-agent being renamed, a provider publishing new guidance, a specification moving. A review either refreshes the dates or rewrites the affected section. Where a claim can no longer be supported, it is removed rather than softened into something vague.

An article whose sources-verified date is old is still shown with that date rather than hidden, because an honest date is more useful to you than a silently refreshed one.

AI-assisted content policy

This is stated plainly because the alternative is a disclosure nobody believes.

Content on this site is produced with AI assistance, under human research and human review. What that means in practice:

• Claims are researched against primary sources, fetched and read, not recalled by a model. A model’s memory of a provider’s documentation is not a source and is never treated as one.
• Every article is reviewed by a person before publication, and carries a last-reviewed date and a sources-verified date recording when that happened.
• Every article names a real, configured author — a person who exists and who is accountable for the content. We do not invent an author, a headshot, a job title, a credential or a biography. There is no fictional expert persona on this site.
• No fabricated trust signals of any kind: no invented statistics, no made-up case studies, no testimonials, no customer counts, no awards, no citations to sources that do not exist.
• No content produced at scale to fill a keyword map. Google's spam policies define scaled content abuse as generating many pages without adding value, and an article here exists because it says something a reader cannot get from the provider’s own documentation.

If an article contains a number, we either observed it and say how, or we cite who did. There is no third option.