How to check what ChatGPT, Gemini and other AI assistants say about your business

To see what AI assistants say about your business, ask them the questions your customers ask, not your business name. Write 10 such questions, put each one to ChatGPT, Gemini, Perplexity, Claude and Grok, and log four things: whether you are named, who is named instead, which sources are cited and what facts are wrong. Repeat the same set every week. The log tells you which source to fix, because each assistant reads different data.

Why check at all

Customers now ask assistants where to eat, which salon to book or which electrician to call. In BrightLocal's February 2026 survey, 45% of US adults said they use AI tools for local recommendations, against 6% a year earlier, while the share who read reviews on Google fell from 83% to 71% (BrightLocal Local Consumer Review Survey 2026). These are survey answers, so they describe what people say they do, not measured behavior.

Being named is not automatic. SOCi looked at more than 350,000 locations of 2,751 chain brands and found that ChatGPT recommended 1.2% of them, while 35.9% showed up in the top three of Google Maps (SOCi 2026 Local Visibility Index). That report covers chains and comes from a company that sells local marketing software, but it makes the point: you cannot assume an assistant knows you because Google Maps does.

You also cannot fix what you have not seen. An assistant can name you with last year's hours, a closed second location or a service you stopped offering. The only way to find out is to ask.

Step 1: write 10 questions your customers ask

Do not type your business name. People who already know your name are not the ones you are trying to reach. Write the questions a new customer would ask before they know you exist. A good set mixes three kinds:

• Category and place: "best Thai restaurant in Silver Lake".
• Need or detail: "where can I get a gel manicure on a Sunday in Burbank".
• Comparison or trust: "is [competitor] or [another competitor] better for a first visit".

Use the neighborhood or city your customers use, and the words they use, which may not be your own menu or service names. Your booking notes, phone calls and review texts are a good place to find them.

Example question sets

Pick your 10 and freeze them. The value of the check comes from asking the exact same questions every week.

Step 2: ask five assistants

Ask each question in ChatGPT, Gemini, Perplexity, Claude and Grok. They do not share one database, so being right in one says little about the others. A few practical rules keep the results comparable:

1. Start a new chat for every question, so earlier questions do not shape the answer.
2. Use the same accounts, the same device and the same physical location each week. Note it in the sheet.
3. Turn on web search where the assistant offers a choice, since that is what most people asking about a local business will get.
4. Copy the full answer and the list of cited links, or take a screenshot. Memory is not a log.
5. Do not argue with the assistant or correct it in the chat. You are measuring, not training.

Step 3: log the same four things every time

For every question and assistant, record whether you were named, where in the answer, who was named instead, which sources were cited and which facts were wrong. Here is a template you can paste into Google Sheets or Excel (tab-separated):

Week	Date	Location	Assistant	Question	Named (Y/N)	Position	Named instead	Sources cited	Wrong facts	Notes
W1	2026-11-09	Santa Monica, CA	ChatGPT	best brunch in Santa Monica with outdoor seating	N	-	Cafe A; Bistro B; Diner C	yelp.com; cafea.com; eater.com	-	Our patio not mentioned
W1	2026-11-09	Santa Monica, CA	Gemini	best brunch in Santa Monica with outdoor seating	Y	2 of 5		Google Maps	Hours: says closes 3 pm (we close 4 pm)	Check special hours
W1	2026-11-09	Santa Monica, CA	Perplexity	best brunch in Santa Monica with outdoor seating	Y	4 of 6		yelp.com; tripadvisor.com	Old address (moved 2025)	Update Yelp address

The rows above are made-up examples showing how to fill the columns. Once a few weeks are in, a summary table makes the pattern visible:

The "named instead" column is often the most useful. If the same two competitors appear every week, look at what sources the assistant cites for them and compare those pages with yours.

Why answers change between runs and places

Expect the results to move even when nothing about your business has changed. Assistants generate each answer fresh rather than looking it up in a fixed list, the web results they read can differ from one run to the next, and local questions are answered for the place the question appears to come from. The same question can name you on Monday and leave you out on Tuesday, or name you from a phone in Santa Monica and not from a laptop in Downtown.

That is why one check proves little in either direction. Being named once is not a win, and being left out once is not a loss. What counts is the share of answers over several weeks, asked the same way from the same place. If you need to know how you look from another neighborhood, add it as a separate location in the sheet instead of mixing the results.

Step 4: fix the source that assistant reads

Each assistant leans on different data, so the place to fix a wrong answer depends on which assistant gave it. Sources change quickly, so the dates below are the dates of the reports.

We did not find a primary source describing where Grok gets local business data, so for Grok use the "sources cited" column of your own log to see which pages it relies on.

Reading the log

• Wrong hours in Gemini: check regular and special hours on your Google Business Profile.
• Old address or phone in Perplexity: check Yelp and Tripadvisor; one stale listing can outlive a move by years.
• Not named by ChatGPT, competitors are: look at the sites ChatGPT cites for them. If they are the competitors' own pages, compare how clearly those pages state what, where and for whom. If they are directories or "best of" lists you are missing from, that is your gap.
• Wrong service list everywhere: the error is probably on your own website, which most assistants read. In Yext's study of 6.8 million AI citations, 44% pointed to businesses' own websites and 42% to listings (Yext, 9 Oct 2025). Yext sells listing management, so read that as an industry observation.

Change one source at a time when you can, and write the date of the change in the Notes column. Then keep asking the same questions. Some sources update in days and others take weeks to be picked up, and only the weekly log will show you which.

What this check can and cannot tell you

The log shows how often you are named for the questions you chose, by which assistants, from which sources and with which errors. It does not show how many customers asked those questions, and it cannot guarantee that a fix will get you named. No one outside these companies controls what an assistant says. What you can control is whether the sources they read describe your business correctly and completely, and whether you notice when they stop doing so. That is the work AIReputa does for clients, and it is work you can start on your own with a spreadsheet and an hour a week.