How this desk decides what counts as a fact

Search is the worst-documented subject this desk has worked on, and website-builder search is the worst-documented corner of it. The platforms describe their own tooling in marketing language. The independent coverage is mostly a decade of recycled observations with the dates filed off. And the one party who knows how any of it is weighted publishes almost nothing about weights.

So this site does the part that can be done properly, and says out loud that it is not doing the rest.

The evidence standard

Three tiers, applied to every capability claim on the site.

A claim is verified when it appears in the vendor's own documentation, or — for a claim about how Google behaves — in Google's own documentation or an on-record statement from a named Google employee with a date attached.

A claim is corroborated when the vendor is silent but two independent sources say the same thing without appearing to have copied each other.

A claim is softened when one source says it and nobody else does. In the copy that reads "reportedly", or "independent testers have found", or "at the time of checking". It never reads as a flat statement of fact, and where the single-source claim is consequential the page says which claim it is and invites a correction.

Everything in the first two tiers carries a source in the site's fact register, along with the date it was last read. Everything on this site was re-checked on 19 September 2026.

The absence rule

When a vendor does not document a capability, the audit records not documented rather than absent. These are not the same thing and treating them as the same is how a lot of builder comparisons quietly become fiction.

It matters most at the bottom of a ranking, where it reverses the usual direction of unfairness: the platforms scoring lowest here are the ones whose documentation is thinnest, and part of what a low score measures is how hard the vendor has made it to find out. That is said plainly on the control audit rather than buried here.

What this desk will not publish

No ranking predictions. No claim that a platform "ranks better", because no method exists to show it.

No hands-on performance numbers. This desk does not run load tests on other people's sites, and it does not reprint somebody else's as if it had. Where page speed comes up, it is discussed as a question of which levers the platform gives you.

No invented case studies. No "we moved a client from X to Y and traffic rose 340%". The genre is full of these and virtually none are reproducible.

No advice written to be quoted by an AI system rather than read by a person. Text aimed at steering generative answers is spam under Google's own policy, and quite apart from that it makes for terrible writing.

Corrections

Every page here is provisional, because the products change. Builders move settings, rename panels, add controls that did not exist last quarter and silently retire ones that did. A page that was accurate in September will be partly wrong by March, and the only defence is a short correction loop.

If something on this site no longer matches what the vendor documents, tell the desk and it gets re-checked and re-dated. Corrections go to the front of the queue, including — especially — the ones that make a scored platform look better than it does now.