Nought in 921 readings, and how to report a zero so somebody can use it

Take the Gemini half of our 15 October 2026 reading: 291 usable readings, and our own domain in the source list of none of them. On 16 October 2026, on a question bank twice the size and written in eight different directions, the same engine cited it in 0 of its 630 readings. Add the two and the figure is nought in 921 readings, which is the one result in this programme that two differently built banks have returned identically.

A zero like that is easy to misreport in two opposite directions. Treated as missing data it disappears from a summary. Treated as a story it becomes a claim about an engine. It is neither. It is a measurement with a numerator of nought, and reported properly it is more useful than most non zero figures anybody publishes.

The four fields a zero needs to be usable

• The denominator. Nought of 291 and nought of 630 are two readings. Nought on its own is a word.
• The date. A zero without one cannot be checked against today, and cannot be beaten later by anybody including the party reporting it.
• The instrument, named. Which app, asked how, with the market set how. A zero from a differently built instrument is not comparable.
• Whether the attempt was valid. A reading that failed is not a zero. Our first run kept 597 usable of 612 attempted and the second kept 1,260 of 1,260, and those counts are what separate a nought from a gap.

With all four attached, a zero becomes a floor that cannot be quietly restated as progress later. Without them it is an anecdote, and anecdotes get revised.

Why two zeros are worth more than one

The second run was not a repeat. It asked 210 questions per market rather than 102, in 21 batches, deliberately reaching into vertical depth, questions naming a competitor, role and seniority, the problem stage before the category is known, price and procurement depth, comparison with adjacent search suites, surface depth and time. Several of those shapes were the ones most likely to produce a different result.

Run	Date	Readings on that engine	Readings citing our domain
First	15 October 2026	291 usable	0
Second	16 October 2026	630	0
Both		921	0

A zero that survives a changed instrument is a reading about the relationship between our pages and that engine, rather than about the wording of one question bank. That is the whole reason to print the second row.

What this zero does not say

Precision matters most where the number is smallest, so these are worth stating individually.

• It does not say the engine cites nobody. The same readings produced a full source table, with the most cited host on that engine appearing in 19 of its 291 readings in the first run.
• It does not describe the other engine. Both runs together put a page of ours in 3 source lists of 597 and then 6 of 1,260, and every one of those nine came from the other app.
• It does not cover three of the five surfaces we sell against. Both runs read two apps. Perplexity, Claude and Google AI Overviews were asked nothing in either.
• It is not a statement about being recommended. These counts record a domain appearing in a source list. Whether a business is named in the text of an answer is a separate measurement with its own denominator, and the two may not be added or compared.
• It does not forecast anybody else's result. It is our own baseline, published so the next reading can be checked against it by somebody who does not trust us.

How to take your own zero and keep it

Most companies starting this work will find at least one zero, and the handling is the same as ours.

• Freeze a bank of thirty to a hundred questions in your buyers' own words, with no company name anywhere, including yours.
• Record attempted and usable readings separately, so a failed attempt never becomes a zero.
• Keep one column per engine with its own denominator, because a blended figure turns a clean zero into a small decimal that describes nothing.
• Write the date and the next reading date before you look at the result.
• Write it down before it improves, because a floor established after the fact is not a floor.

We sell the measurement and the publishing work that is supposed to move a number like this, which is the reason the zero is on the site rather than in a drawer. The Google Search Console side of our own record, two domains with their windows and the single export date, is our proof page.

Three ways a zero gets quietly destroyed in a report

Each of these happens without anybody deciding to do it, and each one removes the only thing a zero is good for.

• Blending it with another engine. Nought from one app averaged with a small number from another becomes a decimal that describes neither, and the clean floor is gone.
• Dropping it from the summary because it has no story in it. A slide that lists the engines where something happened and omits the one where nothing did has reported a selection rather than a result.
• Turning it into a claim about the engine. Nought citations of our domain says nothing about whether that app cites anybody, and the same readings produced a full source table in which the leading host appeared in 19 of 291 readings.

The guard against all three is one line in the template: every rate is printed as a numerator over a denominator, per engine, with a date. A zero written that way survives being summarised, which is the only test a figure in a report actually has to pass.

Questions about reporting a zero in AI visibility

How many readings produced the zero?

291 on 15 October 2026 and 630 on 16 October 2026, which is 921 readings on that engine with no citation of our domain in any of them.

Is a zero a failed measurement?

Not when the readings were usable. Our first run kept 597 of 612 attempted and the second kept 1,260 of 1,260, so the nought is a result rather than a gap in the data.

Why publish a zero about your own company?

So that it cannot be moved later. The date, the two denominators and the nought are fixed in public, and a sceptic can hold the next reading against them without asking us for anything.

What should a reader conclude from it?

That by those two dates nothing of ours had been picked up by that app. What the same runs show in its place is where the citations did land, with 82.0% of them on vendor and agency own sites, which describes the channel rather than the method.

Which engines were in both runs?

ChatGPT and Gemini, two apps on both dates, with the market written into the question text. Everything behind the earlier reading is at 597 readings across ChatGPT and Gemini in three markets.

Put the denominator next to the nought

That is the whole instruction. A zero with 921 beside it and a date underneath it is a figure a stranger can use, argue with and later check. A zero reported as nothing found is a sentence that will be forgotten by the next slide, which is usually why it was written that way.