The third domain that is not on our proof page

There is a third domain in the same export, and it is not published. The source note on our proof page settles it in one line, recording that a third domain shares the export and stays off the page because its figures will not carry a claim. That sentence is the most useful line on the page for a sceptical reader, because every other figure on it was selected by the same hand, and knowing what the selection rejected is the only way to judge what it accepted.

Why the omission is the interesting part

Any published result is the survivor of a selection, whether or not anybody admits a selection happened. A supplier with several clients and several domains has a set of candidate case studies and publishes some of them. The published ones are not evidence about the method. They are evidence about the publishing decision, and the only way to recover anything about the method is to know the size of the set the published ones came out of.

So one sentence naming the rejected candidate converts a case study from an unknown selection into a stated one. It does not make the published figures better. It makes them interpretable, which they were not before.

What a reader may conclude from it

• That the published cases came from a small set rather than a large one, because a third candidate existed and was named.
• That the rejection test was the numbers rather than the client, since the reason given is the figures.
• That a case study on that third domain would have existed if the figures had cooperated, which is a statement about what we would have been willing to publish.

What a reader may not conclude is anything about the third domain itself, because nothing about it is published beyond its existence. That is the right boundary. Naming a site in order to say it did badly would be a worse disclosure than the one being fixed.

Selective reporting, and the two versions of it

Version	What the reader sees	What they can work out
Silent selection	Two strong case studies	Nothing about how many there were
Stated selection	Two case studies and a named exclusion	The size of the candidate set and the test applied

Both pages look almost identical. One sentence separates them, and it is the sentence that decides whether the rest of the document is evidence or advertising.

The asymmetry is worth noticing. Adding the sentence costs a supplier almost nothing, because the exclusion already happened and nobody is being asked to publish a bad result in detail. Leaving it out costs a careful reader everything, because without it there is no way to distinguish a method that works most of the time from a method that worked twice. The cheapest sentence on the page is the one carrying the most weight, which is usually a sign that the omission is habit rather than strategy.

There is measured demand for the negative half

This is not only a matter of taste. When ChatGPT and Gemini were asked on 15 October 2026 what would earn a page a citation, 68 of the 102 conversations put failures and negative results on the list. Set that beside what the same programme counted a day later: of 20,143 citations recorded on 16 October 2026, 82.0% went to a page belonging to the company selling the thing being recommended. In a field weighted that heavily towards sellers, a result published against yourself is the one move almost nobody else is making.

We have published harder versions of the same thing. Our own citation count in the 15 October run was 3 readings out of 597, with none in the United States on either engine, and that figure is on the site under its own heading rather than inside a footnote.

How to do this in your own reporting

The mechanism is cheap and it survives contact with a legal review, which matters more than the principle.

• Count your candidate case studies before you choose. The count is the number that belongs on the page.
• State the test you applied. Ours was whether the figures support a claim, which is a test a reader can imagine failing.
• Name no client who did not agree to be named, including the excluded ones.
• Say it in the source note rather than the headline. It is a scope statement, not a confession.
• Keep the sentence true. If the excluded case later produces figures worth publishing, publish them and say the exclusion ended.

Judging this on somebody else's proposal is a related skill, and the ten line scorecard in how to read the evidence an agency shows you is built from the same 102 conversations, with the failure line marked as the one most worth watching.

Questions about what a case study leaves out

Is it not bad marketing to admit a case did not work?

It costs one slide and buys the credibility of the others. The alternative is a page where a careful buyer has no way to size the selection, and careful buyers are the ones worth winning.

Should I name the excluded client?

No, unless they have asked to be named for that purpose, which is rare. The disclosure that matters is the existence of the exclusion and the test, not the identity.

How many exclusions should a page admit to?

However many there were. A number is better than a word, which is why ours says a third domain rather than some domains.

What if every case study you have is good?

Then say how many you have and that none was excluded, which is a stronger sentence than silence and is just as cheap. A reader's suspicion is about the unknown denominator rather than about failure.

Does an engine actually reward this?

68 of the 102 build spec conversations on ChatGPT and Gemini asked for it by name on 15 October 2026. That is what two engines said they wanted, on one date, and it is the closest thing to a specification this category has. We sell the measurement and the pages behind it, so we are judged by the same list.

One sentence, and what it is doing there

The line about the third domain is not an apology and it is not modesty. It is a scope statement doing a specific job: telling a reader how wide the net was before the catch was displayed. Any published set of results has a shape like that behind it. Most pages leave the reader to assume it, a few leave the reader to suspect it, and the cheapest fix is to write the number down.