Simulator

One check or twelve: see the difference

Set how often ChatGPT really names a business, then set how many times you ask. Each ask is a fresh chat that names the business or does not, at that rate. Watch what a report built on those asks says, and how far it can sit from the truth.

Simulation, not a measurement of any business. The real numbers

How often ChatGPT really names you30%

The true rate. No single check can see it.

How many times you ask1 ask

Each ask is a fresh chat, the same question.

Report 1: your asks

NamedNot named

Ten reports, nothing changed between them

What the report saysIts 95% rangeThe true rate

The 95% range is where the true rate probably sits. About 1 report in 20 misses it, so a bar can miss the dashed line.

The arithmetic

What each number of asks can tell you

The same 95% interval the simulator draws, worked out for a few numbers of asks. A margin of plus or minus 24.6 points means a report of 50% could sit anywhere from about 25% to 75%.

Margin of error and chance of a complete miss, by number of asks
Asks	Margin in points, true rate 50%	Margin in points, true rate 10%	All asks miss a 10% brand
1	±44.5	±41.5	90.0%
3	±37.5	±31.8	72.9%
6	±31.2	±24.4	53.1%
12	±24.6	±17.7	28.2%
24	±18.6	±12.4	8.0%
40	±14.8	±9.5	1.5%

Wilson 95% interval, with the report equal to the true rate. At 12 asks and 50% it gives the plus or minus 24.6 points published in One AI check against twelve. The published plus or minus 17.0 is for an observed 1 of 12. The 10% column gives 17.7 because 10% of 12 asks is 1.2.

To bring one question's margin under plus or minus 15 points takes 17 to 39 asks for a true rate between 10% and 90%.^[1] That is why the Stack reports a margin of error on every rate and a real or noise verdict on every change, instead of one number.

Published

What the real test found

45.9%

Where 12 ChatGPT runs named a brand at least once, the share of single runs that named it. One check missed more often than it found.^[1]

21.7%

How often a single ChatGPT check gave the less common answer about whether a brand was named.^[1]

How the Stack measures a month

Limits

What the simulation assumes

Every ask is independent. That fits ChatGPT in the published test. Perplexity's repeats were not independent, so this simulator is about ChatGPT only.^[1]
The true rate holds still. In real life a model update can move it overnight, which is one more reason every number carries a date and a model.
One question. A monthly report pools many questions, and pooling narrows the margin.

Measure the real number

The baseline asks each of your questions 12 times, on ChatGPT and on Perplexity, in each language. What the baseline covers

Book a discovery call

Delivered by EWR Digital, Houston

Sources

The Best SEO Podcast, One AI check against twelve. Figures dated 2026-10-25, from answers collected 12 to 26 September and 1 to 5 October 2026.

Cite this page

"One check or twelve: a simulator." llmvisibilitystack.com, updated October 28, 2026. https://llmvisibilitystack.com/measurement/simulator

To cite the method itself, cite the book: Bertram, M. (2026). LLM Visibility: A Decision-Grade System For Winning AI-Mediated Discovery.