How to check whether ChatGPT recommends your business

The short answer

Ask ChatGPT the questions your customers ask, the way they ask them, and write down what comes back. Do it with search turned on, more than once, and on more than one engine. Record three things separately: whether your business is named, whether your website is cited as a source, and whether you are actually recommended.

That gets you a first look. It does not get you a measurement. ChatGPT's answer to a question is one sample, and the answer your customer sees can differ from the one you see. To know where you stand you need the same questions asked the same way, across the engines, over weeks. That is what our free audit starts.

Step one: write the questions your customers ask

Start from real conversations, not from your keyword list. Homeowners ask assistants full questions, and they ask them in several shapes:

• Service and place. Who is the best roofer in Chandler?
• Problem first. My AC is blowing warm air in a Mesa house, who should I call?
• Comparison. Is it worth replacing single-pane windows in Phoenix, and who does it well?
• Named check. Is this company any good? (with your business name)

Write a handful of each for your main services and the cities you want work in. Give the problem-first questions extra attention: the homeowner has not picked a trade or a company yet, and no business name is in the question at all.

Step two: ask them properly

Use ChatGPT with search on, since that is how it looks up local businesses. OpenAI documents that ChatGPT's search features surface websites through its own crawler, OAI-SearchBot¹, so an answer that shows sources is the one that tells you the most.

Start a new chat for each question so earlier questions do not steer later answers. If you can, ask from a device and an account that are not yours, so the test is closer to what a stranger would see.

Ask each question more than once. One answer is an anecdote. If your business appears in one answer and not in the next, you have learned something about how stable your position is.

Then repeat the exercise in Gemini, Perplexity, and Claude. The engines read different sources and can name different businesses, which we measure in why the engines recommend different contractors.

Step three: record what you see

For every answer, note:

• Named. Does the answer mention your business at all?
• Cited. Is a page from your website among the linked sources?
• Recommended. Is your business presented as a choice to hire, rather than just mentioned?
• Who else. Which competitors are named, and in what order?
• Which sources. Which websites did the answer cite? Directories, review sites, and articles on that list are the pages ChatGPT read before it answered.
• What it got wrong. An old phone number, a closed location, a service you do not offer. Wrong facts come from somewhere, often a page you can find and fix.

If your business is never named, look at the sources, and check whether the competitors who are named are described on those sites. That is the most useful thing a manual check produces: a list of pages that matter, and whether you are on them.

Step four: rule out the technical failures

If ChatGPT never cites your website, even for a question that names your business, check that it can reach the site. OpenAI's help center says that to be eligible for ChatGPT search, you must allow OAI-Searchbot to crawl the site², and your host or CDN must allow traffic from OpenAI's published search-bot IP addresses. A robots.txt rule or a firewall setting can block that without anyone noticing. Our guide to AI crawler access shows how to check.

Why a manual check runs out of road

A spreadsheet of answers is a good start and a bad measurement. Three problems show up fast.

Scale. A meaningful sample is large. Our Phoenix window and door index, in its newest weekly run, was built from 86³ answers from Claude and Gemini, and those answers named or cited 11⁴ different businesses. That is one trade and two engines; ChatGPT and Perplexity will be added to the run, and each one adds its own answers to read. Few owners will read that many answers by hand, every week, and compare them.

Change. The answers move. A check in March says little about June. What you want is a trend: your share of the answers, week by week, and whether the work you did moved it.

Selectivity. Being recommended is harder than being listed. SOCi found that ChatGPT recommended 1.2%⁵ of the locations of the multi-location brands it studied. That is a different kind of business from a local contractor, but it shows how narrow the list at the top of an AI answer can be.

This is also where a manual check can mislead. A question that names your company can come back with a good description of it. That is not the same as being recommended to a homeowner who has never heard of you, and it is the second one that brings in work.

What this means for an owner

Run the manual check. It takes an afternoon and it will show you whether you are in the conversation at all, and which competitors and sources are.

When you want the technical side checked properly, use our free AI visibility audit. It crawls your site the way the AI crawlers do, checks what ChatGPT, Claude, Perplexity, and Gemini rely on, from crawler access and structured data to whether your name, address, and phone agree across the web, and emails you a score and the findings that matter most, alongside where your trade stands in our weekly measurements. Clients then get the weekly version: share of voice per engine, citations tracked, and listings fixed. We report what moved. We never promise a ranking.