The comparison we refuse to make on our own proof page, and the house rule that enforces it

There is a sentence we are not allowed to write. Our proof page carries two Google Search Console exports, both taken on 2026-10-14, and the house rule says they are published as two separate case studies and never compared: not in one sentence, not in one table row, not in one pair of bullets. That rule removes the most persuasive line available to us, which would be the two results read against each other as though they were a sequence. It is written down precisely because nobody would break it on purpose and several people would break it by accident.

The rule, as it is actually written

One domain per passage. Write about one case or the other. The two differ in domain age, in link profile, in window length and in the claim each one supports, so neither is evidence for the other and a figure drawn across both would describe a site that does not exist.

The page says it in its own words too, above the second case: different domains, different ages, different claims, neither is evidence for the other. Putting the rule on the public page rather than only in the style guide is part of the enforcement, because a reader who knows the rule can see when it is kept.

The first case, on its own

clawlaw.in covers 2026-07-13 to 2026-10-12, 92 days with data. It records 18,537 clicks and 929,405 impressions, with daily clicks moving from 59 to 349 across the window. That domain had history and existing pages before the window opened, so the figures are growth on a working site.

The second case, on its own

aiknowsus.com covers 2026-10-04 to 2026-10-12, 9 days with data. It records 5,965 impressions and 3 clicks, and the 5,965 is a discovery result rather than traffic. That domain was registered days before the window opened: no history, no links, no age.

What the rule costs us, stated plainly

It costs the narrative. Two cases presented in sequence imply a journey, and a journey is what a buyer wants to be shown. Written that way the page would read as one story with a beginning and an end, which is more memorable than two self contained measurements sitting in separate boxes.

It also costs the bigger number. A sentence that reaches across both cases can quote whichever figure is largest and attach it to whichever starting point is smallest, and every word of it would be individually true. That construction is available, it is not a lie, and it is banned.

Why a rule rather than care

Care does not survive volume. We have found this in our own publishing more than once, and the mechanism is always the same: nobody decides to write the bad sentence, the structure of the work produces it.

The clearest case was a closing paragraph. A brief told a large number of writers to end every page the same way, and they independently arrived at nearly the same sentence, so an audit found one phrase sitting word for word on 82 live pages and another on 46. Each writer had written it once. The site had published it a hundred and twenty eight times.

A cross case comparison would go the same way. Each writer, reaching for one vivid line, would reach for the same one, because it is the best line available. The rule is cheaper than catching it afterwards, and a machine check that fails a file carrying both domain names in one paragraph is cheaper still.

The other refusal already on this site

This is not the only stated refusal here, and the older one is instructive because it also costs us something a reader asks for. Two live benchmark pages decline to reproduce any tool's prices, our competitors' included, and call it a deliberate refusal rather than an omission. It rests on a finding from 17 September 2026 that an engine discounted accurate but undated competitor prices.

A refusal with its reason attached is itself the kind of disclosed method the engines keep asking for. 99 of 102 build spec conversations run on ChatGPT and Gemini on 15 October 2026 asked for a disclosed methodology, and 64 of 102 asked for the commercial interest to be declared on the page by name. We sell AI visibility measurement, so both of those apply to us before they apply to anybody we quote them at.

The same family of rule governs how a single figure is reported. A share of a total means nothing until the report states what it divided by, and the three legitimate denominators behind one of our own figures shows the same count honestly producing three different percentages.

Writing a rule like this into your own house style

A rule that lives in somebody's head is a preference. Four things turn it into a rule.

• Write it as a prohibition, not a principle. Never both in one sentence is checkable. Be careful when comparing is not.
• Say what it costs. A rule whose price is unstated gets quietly dropped the first time somebody wants the line it forbids.
• Put it where the work happens. In the brief every writer reads, not in a document nobody opens twice.
• Make something fail. A check that refuses the file is the only version that survives a busy week.

If your organisation needs this as a document rather than a habit, a marketing policy you can copy section by section is the right shape to hang it on, because a claims rule belongs beside the approval route that enforces it rather than on its own.

Questions about the one domain per passage rule

Is the comparison it forbids actually misleading?

It is unsupported rather than false. Every figure would be accurate and the sentence would still imply a relationship between two domains that differ in age, links and window length. The rule removes the implication rather than correcting a number.

Does this mean the two cases are weak?

Each is as strong as its own window makes it, and both are published with the window, the day count and the export date. What the rule removes is a claim that neither export contains, which is a different thing from a weakness in either.

Could you publish a third case that bridges them?

Only by measuring one, with its own window and its own export date, and it would then be a third separate case under the same rule. A bridge built out of arithmetic rather than measurement is the thing the rule exists to refuse.

How do you stop the rule being broken in a table?

A table row is the easiest place to break it, because a row invites a side by side reading without a sentence being written. The check fails on any row carrying both domain names, which is why each case has its own table on the page rather than sharing one.

Why publish the rule instead of just following it?

Because a reader cannot tell restraint from an oversight. Publishing the forbidden sentence and the reason it is forbidden turns a gap in the page into a disclosed method, and 98 of those 102 conversations asked for the same kind of explicitness about engine names for the same reason.