If you want to know what ChatGPT says about your business, the natural first move is to type the business name and read the reply. Most owners do exactly that, feel relieved or annoyed for a minute, and close the tab.
That reply is the least useful one you can get. A buyer who already has your name is nearly done deciding. The buyer you pay to reach types something else: a service plus a city, a problem in plain words, a question about price. Your name appears in none of those. Whether the assistant brings you up anyway, and what it says when it does, is the thing to check.
Start with the questions buyers type
Write your list of questions before you open any assistant. Around 15 is a workable size, and it is the count NG used for the industrial cleaning client described below. Build them from the way customers reach you: the phrases on your contact forms, the first thing people say when they call, the questions you answer again and again. Mix a few kinds:
- Service plus city. Say, emergency plumber in Burnaby, or bookkeeper for contractors in Kelowna.
- A problem in plain words, the way a stressed person types it late at night.
- A shortlist request: best, top, who to call.
- A price question: how much does it cost, in your city.
Add two or three questions that do include your business name, and keep them in their own group. They tell you what each assistant says when asked directly. The other questions tell you whether a stranger ever hears about you.
Ask four assistants, not one
ChatGPT, Gemini, Perplexity and Claude can answer the same question differently. One may name a few businesses, another names a different set, a third names nobody and points to a directory. If you check only ChatGPT, you see one view out of four, and the buyer may be using any of them.
Open a fresh chat for each question, and turn off memory in the settings first. ChatGPT can reference saved memories and past chats, and that can shape a reply. Run each question more than once, because the wording of an answer can shift from one run to the next. Then put the results in a sheet with columns for the question, the assistant, the businesses named in order, the sites cited, and whether you appear.
The first run takes the longest, since you are writing the questions as you go. Later runs reuse them.
Read the sources before you read the answer
The answer text gets the attention. The citations explain it. ChatGPT and Perplexity show links to their sources beside the reply when they search the web. Where an assistant shows none, note that in your sheet. Sort what you find into three piles: directories, review sites, and the business's own pages.
Say a competitor is named first and the answer cites a directory where that competitor has a full listing and yours is half empty. That is a fix you can make yourself. If the answer cites a competitor's service page that answers your exact question, you need a page that answers it better. If it cites your own pages, write down which ones. Those are doing the work, so leave them alone.
Three results and what each means
You are missing from the answers. The assistant names other businesses and never yours. Look at what the cited pages have that yours lack: a page for that service in that city, clear prices, a listing with the same details as your site.
You are named with wrong details. An old address, a service you dropped, a phone number that belongs to someone else. The assistant may be drawing on an outdated source, or it may simply have it wrong. Where the answer shows its sources, start there.
You are named and cited, but only on one assistant. That is common, and it is worth knowing before you assume the whole picture looks the same.
Watch for the wrong company
A business with a common name can be merged with a different company in the answer. Say you run Pacific Roofing in the Lower Mainland. An assistant may blend you with another Pacific Roofing in a different city and hand your buyer their services or their phone number.
Read each answer that names you and check what it says about the city, the services, the years in business and the contact details. If another company has crept in, make your own details unmissable: the same name, address and phone number on your site and every listing, and a home page that states the city and the services in its first lines.
What a better result looks like
NG runs this check every month for clients, across ChatGPT, Gemini, Perplexity and Claude, and records which sites the answers cite. For an industrial cleaning company in BC, the check meant buyer questions, several assistants, and a count of how many answers cite the client's site. On September 24, 2026, the site was cited in 76% of the answers to 15 buyer questions across three assistants. On July 7 the figure was 47%. The full numbers are on the industrial cleaning case page.
The point of counting is that one percentage can be compared month to month. A feeling that you probably show up cannot.
How often to check, and what to do with it
Answers shift, so a single check is a snapshot. Monthly is the pace NG uses for clients, and it is enough to see a trend without turning the sheet into a second job.
To understand what the assistants weigh when they pick a local business, read how AI recommends local businesses. If your check turned up a description of your company that is wrong or thin, your AI salesforce is already describing your business to customers covers why that happens before a buyer ever reaches you.
When the sheet shows gaps, the work lands in SEO and AEO: pages that answer the buyer's question, listings that match your site, and sources the assistants can read. NG's SEO & AEO plans run $800, $1,200 and $1,800 a month (CAD). SEO compounds over months, with first moves in 6 to 10 weeks and real traffic in 3 to 6 months.
You can run the first check yourself this week with nothing more than a list of questions and a spreadsheet. If you would rather have someone run it every month and fix what it finds, request a quote.