Four louder headlines the same export would have carried, and why none of them is on the page

The claim we publish from the aiknowsus.com export is that the pages got found this fast, from nothing. The file behind it is on our proof page: 2026-10-04 to 2026-10-12, 9 days with data, exported 2026-10-14, 5,965 impressions and 3 clicks, on a domain registered days before the window opened. That same file would have carried at least four louder headlines, every one of them built from numbers in the file rather than invented. Each is set out below with the exact thing it takes that the export does not give. Choosing between them is an editorial act and it happens before the page is written, which is why it is worth showing.

Headline one: nearly six thousand people found us in nine days

Two faults in one sentence, and they compound. The first is the rounding. The figure is 5,965, and a number given as nearly six thousand cannot be checked against anything, because the reader no longer knows what the file said.

The second is the unit. An impression means a page was in the results for something somebody typed. It does not mean anybody arrived. The clicks over the same nine days total 3. Converting impressions into people is the single most common move in this category and it is the one that would make our own page a falsehood about our own data.

Headline two: 2,685 in a single day

This one is true and it is still wrong, because of what sits either side of it. 11 October 2026 was the best day in the window at 2,685 impressions. The day after recorded 2,055. The full series across the nine days reads 5, 4, 6, 7, 19, 84, 1,100, 2,685 and 2,055.

A headline built on the peak invites the reader to treat the peak as the level. Publishing the whole series instead costs one line and removes the question, and it is why the proof page prints every day rather than the best one.

Headline three: a growth multiple from the first day to the best

The arithmetic is available to anybody. The first day of the window recorded 5 impressions and the best day recorded 2,685, so a multiple can be computed and it would be a very large number.

It would also be information about the base rather than about the work. A multiple off a first day of 5 is a statement that the starting figure was tiny, dressed as a statement that the finish was large. The honest form of the same fact is the two endpoints, printed, so a reader who wants the multiple can produce it and immediately see what it was divided by.

Headline four: this is what the method does for a new site

This is the most tempting one, because it is the sentence a buyer wants, and it is the furthest from the file. The export covers one domain over nine days. There is no second domain in it, no control, and no repeat.

80 of 102 build spec conversations run on ChatGPT and Gemini on 15 October 2026 asked for the sample size N, and 83 of 102 asked for a repeat count, a variance or a confidence interval. A method claim from a single nine day window fails both of those, and it fails them in public, on a page whose whole purpose is to be checkable.

The sentence that survived, and the test it had to pass

What is left is narrower and it is the only one with nothing borrowed: a domain with no history, no links and no age was being shown in Google results 5,965 times within nine days of 4 October 2026, and took 3 clicks in the same period.

The test each candidate had to pass is one question. If a reader pulled the same export and read every row, would they arrive at this sentence on their own? Headlines one through four all require the reader to accept something that is not in the rows. The published one does not.

Why the rejected versions are published rather than discarded

Because a reader cannot tell a careful claim from a cautious one without seeing what was available. A page that only shows the final sentence looks modest. A page that shows the four it declined shows the standard, and the standard is the thing being sold.

That is also how the rest of this site handles its own bad numbers. The audit page reporting that our own domain was cited 3 times in 597 readings is published under that heading rather than buried, and 68 of those 102 build spec conversations asked for exactly that: failures and negative results shown as well as wins.

We sell the measurement and the work that moves it, which is the reason the restraint has to be visible. A company with no commercial interest in a figure can afford to be vague about it. We cannot.

For the general version of this discipline, applied before a number exists rather than after, the before and after protocol sets out how to fix what counts as working while the result is still unknown.

Questions about the claim we chose and the ones we did not

Would the louder headlines have been false?

One of them would. Calling impressions people is a false statement about what the column measures. The other three are not false, they are unsupported: each needs the reader to supply something the export does not contain, and the page would be relying on them not noticing.

Why not publish the multiple and the base together?

Because the two endpoints already contain the multiple and nothing else contains the endpoints. Printing 5 and 2,685 lets a reader compute any ratio they like. Printing a ratio and hoping somebody asks for the base is the version that gets quoted without its base.

Is three clicks worth publishing at all?

It is the most useful figure on that case. A click column reading 3 beside an impression column reading 5,965 is what stops the page being read as traffic, and it is the number that would be missing from any version of this written to flatter us.

How would somebody disprove the published claim?

By pulling the same performance export for aiknowsus.com over 2026-10-04 to 2026-10-12 and finding different rows. The window, the day count and the export date are all printed so that the check is one screen of work rather than a request.

When does a single nine day window become a method claim?

When it has been repeated on other domains, with the repeat count and the dates published for each. Until then the right sentence names one domain, one window and one export date, which is what ours does.