Same Question Twice
The same question, twice in one day. 4 names in 10 changed.
Two people asked this in the comments of my last launch here: if an AI names your business today, does it name you again tomorrow? Nobody selling AI-visibility tools publishes an answer, so I measured it.
In 24 US cities I asked Google's AI four plain questions - "What's a good restaurant in Ann Arbor, MI?" and three like it - then asked the same four questions again, word for word, in a separate session on the same day. Nothing else changed.
It named 292 businesses. 126 of them appeared in only one of the two runs. That is about 4.3 out of every ten names, on the same day, with the same question.
The data is at growl-ai.com/guides/same-question-twice - Fazier's Visit button only accepts a domain, so it lands on the site root.
Every name is on that page, city by city, both runs, so you can check it instead of believing me. The four questions and the searches the model ran are there too. So is the arithmetic: 166 + 126 = 292.
Before publishing it I recounted all 313 names by hand and found that three of my own rows were not businesses at all - headings the model wrote that I had recorded as names. They are still in the table, marked, with the strict recount next to them: 121 of 288, or 4.2 in ten. I would rather show you my bad rows than a rounder number.
One engine only: Google AI (Gemini), grounded. Not ChatGPT, not Perplexity, not AI Overviews - I did not ask them, so I am not going to say anything about them.
I sell a tool that runs this measurement, and the honest thing to put next to it is how much the answer moves on its own. The page is free, there is nothing to sign up for, and there is no email box on it.
Features
Every business name published - both runs, two trades, 26 cities measured in all
The headline uses the 24 cities where both runs used identical wording; the other 2 are marked
The four questions word for word, and the Google searches the model actually ran
The addition is shown on the page: 166 + 126 = 292, so the count can be checked
A strict recount that removes my own bad rows: 121 of 288, printed next to the headline
"What this does not tell you" is listed before any conclusion is drawn
One measurement day, one engine, stated plainly instead of implied
Use Cases
You are being sold "AI visibility" and want to know how noisy a single reading is
You track your brand in AI answers and need a baseline for run-to-run variance
You write about AI search and want a primary source rather than a vendor's claim
You are building something similar and want numbers to compare your own against
Comments
I built the tool first and the dataset second, and the second one changed how I sell the first. I pointed my own AI-visibility tool at my own website. The page scores 94 for AI-readability. The AI named me in 0 of 10 questions. I published that instead of hiding it - and then it started bothering me that I was about to tell other people their single reading meant something, when I had never checked whether a single reading is stable. So I ran the same four questions twice on the same day, in 24 cities. 4 names in 10 moved. That is an uncomfortable number to publish when you sell the measurement, but it is the honest thing to put next to it. If you run any kind of brand-in-AI check: do you see the same size of drift? And if you find an error in my table, tell me - I already found three of my own and left them in, marked.
Premium Products
Sponsors
Buy
Makers
Makers
Comments
I built the tool first and the dataset second, and the second one changed how I sell the first. I pointed my own AI-visibility tool at my own website. The page scores 94 for AI-readability. The AI named me in 0 of 10 questions. I published that instead of hiding it - and then it started bothering me that I was about to tell other people their single reading meant something, when I had never checked whether a single reading is stable. So I ran the same four questions twice on the same day, in 24 cities. 4 names in 10 moved. That is an uncomfortable number to publish when you sell the measurement, but it is the honest thing to put next to it. If you run any kind of brand-in-AI check: do you see the same size of drift? And if you find an error in my table, tell me - I already found three of my own and left them in, marked.
Premium Products
New to Fazier?
Find your next favorite product or submit your own. Made by @FalakDigital.
Copyright ©2025. All Rights Reserved