The standard, published before the studies are.

Every benchmark here leads with the ninetieth percentile and reports the median underneath it. That ordering is the whole methodology in one decision. A four-minute median tells you nothing about whether a business can be reached, because the failures are not in the middle of the distribution — they are in the enquiry that arrived on Friday evening, the channel nobody monitors, the form that routes to a mailbox somebody left. Those live in the tail, and an average is designed to hide them.

Non-responses are counted separately and kept out of the percentiles entirely. A subject that never replied has no duration, and folding it in as an arbitrarily large number would let the choice of that number decide the result. It is reported as what it is: a count, next to the sample it came from.

Cohorts are only reported where there are enough subjects in them to mean anything, and that floor is enforced by the build rather than by judgement. A thin cohort produces the most quotable line in any study and the least defensible one.

Only surfaces a business publishes to the world are used — the contact form on the site, the number on the listing, the address on the invoice. Nothing behind a login, a credential or a paywall is touched, whatever the finding would have been. Measuring a company's public funnel is research; getting behind its login is not, and the distinction does not bend for an interesting result.

No measured business is named and no subject is identifiable from what is published. Figures are aggregated before publication rather than after, which is a different thing: the identifying data does not survive into the file the site reads, so there is no version of the study in which a name could be published by accident.

The burden on a subject is one ordinary enquiry of the kind they receive every day, submitted once. That is the entire imposition, and it is stated on every study rather than assumed — the party best placed to notice an ethical shortcut is never the one being measured.

Generative engines suppress redundancy. A page restating a consensus already in the model's training data gives it no reason to cite anything, which is why original measurement is the only durable form of visibility left — you cannot be synthesised out of a result nobody else has.

It also does the commercial work of four assets at once. It is the proof an operator without a client list cannot otherwise generate. It supplies the calculators defaults that are ours rather than borrowed. It gives the industry hubs a data floor. And it is the only outbound worth sending to somebody who did not ask, because it tells them something true about their own market rather than something flattering about us.