Ask ChatGPT, Perplexity and Google's AI who builds websites for a plumber in your town and you can get three different lists, because the three engines don't pick a local business the same way.
Here is how ChatGPT, Perplexity and Google AI pick a local business, measured on our own trade. In our October scans, OpenAI's search relied almost entirely on businesses' own web pages: 137 of its 138 citations went to business sites and one to a directory. Perplexity used business sites too (483 citations) but also leaned on LinkedIn company pages (27 citations) and agency directories (50 directory citations in all). Gemini, Google's model, cited business sites 345 times and directories 7 times, and favoured sites with many city and service pages. So the same business can be named by one engine and ignored by another, and the fix is different for each.
A citation is the page an engine links to as its source; counting them shows what each engine reads.
How we measured it
We wrote 40 buyer questions for three North Texas towns, the kind a customer types when looking for a web designer or a chatbot builder, and we keep the list fixed. Each question was run once through OpenAI's model with web search, once through Perplexity and once through Gemini with Google Search grounding, and a second model pass pulled out every business named and every page cited. We count how many of the 120 answers name a business, never where it sits in the list, and we ran the same set three times in October; the method and the free check are at /aeo.
ChatGPT: your own web page is almost the whole signal
What it cited: in the latest of our October scans, OpenAI's search made 138 citations and 137 of them went to businesses' own websites. The two earlier runs looked the same (129 business sites and 2 directories, then 138 and none). It cited our three city pages three times each, and our domain was its most-cited domain in that run, with 12 citations.
What that means: ChatGPT runs a web search, reads the pages that come back and writes a few names into its answer. If no page on your site says, in plain words, that you do this service in this city, it has nothing to quote. Our own count on this engine was 2 of 40 answers in the first two scans and 12 of 40 in the third, after our city pages had been indexed. One scan is not a trend, but the pages it cited were exactly the city pages.
The one fix that matters most: a page on your own site for each service you want to be named for, written for the city you serve, with the service and the city in the heading and the first sentence. Our guide to writing a service page an AI can quote shows the layout.
Perplexity: your page, plus LinkedIn and the directories
What it cited: 483 business-site citations and 50 directory or profile citations in the latest scan. LinkedIn company pages alone accounted for 27, and the two agency directories it used most were cited 9 and 6 times. That directory layer was there in all three scans (53, 50 and 50 citations), and LinkedIn was its most-cited domain every time.
What that means: Perplexity reads the same business websites as ChatGPT, then cross-checks them against places where someone else describes the business. In its answers about us it called our evidence "company-published" and asked for independent reviews. Our count on this engine moved least of the three (2, 2, then 3 of 40), and we are on LinkedIn but not yet in the directories it cites.
The one fix that matters most: a complete LinkedIn company page with the same one-paragraph description as your site, then full profiles on the two or three directories your trade already uses, then reviews on a site you don't control.
Google's AI: many pages beat one good page
What it cited: 345 business-site citations and 7 directory citations in the latest scan, close to the earlier runs (340 and 4, then 351 and 7). The four sites it cited most were each cited 7 to 10 times, and each has a separate page for many cities and services.
What that means: every extra page is another chance to be the one Gemini pulls in for a question. Our count here went 2, 5, then 2 of 40. In the second scan it named us in three answers for one town, using the wording from that town's page; in the third, none of the three repeated. We measured Gemini with Google Search grounding, the closest we can get by API to Google's AI Overviews and AI Mode, which we did not scan directly.
The one fix that matters most: more real pages, not a longer home page. One page per city you serve and one per service, each with its own answer.
Why ChatGPT, Perplexity and Google AI name different businesses on different days
The same question can name different businesses on different days, on every engine. Between our first and second scans, four days apart, the business named most often in the first scan (17 of 120 answers) was named 8 times in the second, and only three businesses were named by all three engines in both runs.
We were named in 6, then 9, then 17 of 120 answers across the three scans, but the engines didn't move together: Gemini went from 5 to 2 of 40 between the second and third scan while OpenAI went from 2 to 12. One screenshot, good or bad, proves nothing. The honest measure is the count over months: a fixed list of questions, the same engines, and the number of answers naming you, written down each time. Our ten-minute test shows how to run the first round yourself.
The engines don't share a method, so visibility has to be earned one engine at a time.
One question, three engines, two scan days
Take one question from the set: "Who builds AI assistants for businesses in Prosper, Texas?"
In the second scan, Gemini named us, citing our site and describing us as located next door. OpenAI said it couldn't find a verified builder with an office in the town and listed regional firms from their own sites; we weren't on the list. Perplexity put a Prosper agency at the top, citing its LinkedIn page and its site, and we weren't there either.
Six days later, the same question came back different on all three. OpenAI gave a short list with one page per business, and we were on it, cited from our page for that town. Perplexity named the same agency again and now us as well, citing our home, about and services pages. Gemini gave the longest list, drawn from sites with many location pages, and we had dropped off it. Same question, same week, three different answers, each explained by what that engine reads.
What to do, by engine
- For ChatGPT: write the page. One page per service per city on your own site, service and city in the heading and first sentence. Nothing else moved its citations in our scans.
- For Perplexity: a complete LinkedIn company page that matches your site word for word, profiles on two or three directories for your trade, and independent reviews.
- For Google's AI: more pages, one for every city you serve and every service you offer, each with a real answer.
- For all three: the same business name, description and phone number everywhere, and a monthly count of a fixed question list so you can see which engine moved. Our post on the five sources AI engines trust gives one action per source.
How Deca Digital helps
We build the city and service pages as part of a custom website, set up the LinkedIn page and the directories so they match the site, and run the monthly count across the three engines so you can see which one is naming you. We're based in Celina, Texas; the full AI visibility service, with a free first check, is at /aeo.
To find out which businesses the engines name for your trade today, book a free discovery call and we'll run the check before we talk.
Frequently asked questions
Does ChatGPT use directories or reviews to pick a local business?
Not much, in what we measured. In our October scans, 137 of OpenAI's 138 citations in the latest run went to businesses' own websites and one went to a directory. It does weigh evidence when it describes a business: asked about us by name, it said we looked worth talking to but were not yet a proven, strongly reviewed agency. The page gets you named; reviews shape how you're described.
Why does Perplexity name businesses that ChatGPT doesn't?
Because it reads more kinds of sources. In our October scans Perplexity made 483 citations to business websites and 50 to directories and profiles, including 27 to LinkedIn company pages, while OpenAI's search made one directory citation in total. A business with a complete LinkedIn page and a couple of directory profiles can be named by Perplexity and never appear in ChatGPT, and the reverse happens just as often.
Is one check enough to know whether AI engines recommend my business?
No. The same question can name different businesses on different days. In our October scans our own count on one engine went from 5 of 40 answers to 2 of 40 in six days while another went from 2 to 12. Run a fixed list of questions through the same engines on a schedule, write down how many answers name you, and judge the count over months rather than any single answer.