How to track whether ChatGPT mentions your business over time, and across the other engines

The short answer

Pick a fixed list of the questions your customers ask, ask each one several times in ChatGPT with search on, and log whether your business is named, whether your site is cited, and who else appears. Repeat the same list on a schedule, and on Gemini, Perplexity, and Claude too. The number to track is your share of the answers, not whether one answer mentioned you.

A spreadsheet works for a first month. After that, most businesses either use AI visibility software or a service that runs the questions for them. Whichever you choose, the method matters more than the tool: the same questions, asked the same way, repeated, across engines, over weeks.

Why one check is not tracking

Our guide to checking whether ChatGPT recommends your business covers the one-afternoon version. Tracking is different, for three reasons.

Answers vary. Ask ChatGPT the same question twice and you can get a different list of businesses. A mention in one answer and not the next tells you your position is unstable. It does not tell you whether you are winning.

Engines disagree. ChatGPT, Gemini, Perplexity, and Claude read different sources and name different businesses. Our Phoenix measurement so far covers window and door questions and two engines, Claude and Gemini. On 100%¹ of the 8² questions we compared, they named the same business most, yet they did not name the same list: of the businesses Google shows that we could compare, the share Claude never named was 33%³, and for Gemini it was 0%⁴. They also cited different websites most, ecowatch.com⁵ for Claude and energyshieldwindowsanddoors.com⁶ for Gemini. Tracking one engine tells you about one engine. The reasons are in why AI engines recommend different contractors.

Mentions and citations are different things. An answer can name your business without linking to any source, and it can cite your website without recommending you. Similarweb found that roughly 6.8%⁷ of US ChatGPT prompts produced answers with web citations by May 2026, across all kinds of prompts. Whatever the rate for your questions, record names and links separately.

A manual tracking log that holds up

If you want to start by hand, set it up so the results can be compared from week to week.

Fix the question list. Write the questions your customers ask, in their words. Include service-and-city questions ("best roofer in Chandler"), problem-first questions ("my AC is blowing warm air, who should I call in Mesa"), and comparison questions. Do not change the wording between rounds. A reworded question is a new question.

Fix the conditions. Use search on. Start a new chat for each question so earlier questions do not steer the answer. Use the same account or, better, a logged-out session or one that is not yours, so your history does not tilt the result.

Repeat each question. Ask each question several times in a round. A business named in most repetitions has a stronger position than one named once.

Log the same fields every time:

• Date, engine, question, and repetition
• Whether your business is named, and where in the answer
• Whether a page on your site is cited
• Whether you are recommended as someone to hire, or just mentioned
• The competitors named
• The websites cited

Compute a share. For each engine, divide the answers that named you by the answers you collected. That is your share of voice for the round. Track it round over round, and note what you changed in between (a listing fixed, a page published, a crawler unblocked), so you can see what moved it.

Where manual tracking runs out

The log gets large quickly. Our Phoenix window and door index, for one trade and two engines, was built from 86⁸ answers in its newest weekly run. A small business tracking even a short list of questions across four engines, with repetitions, every week, can easily be reading hundreds of answers a month. The reading is not the hard part; keeping the conditions identical for months is.

There is also a limit no spreadsheet fixes. A chat window shows you one answer to one person. The engines' APIs, with search enabled, are closer to a controlled measurement: the same question, the same settings, no personal history, repeated. That is how panel-style measurement works.

How panel-style measurement works

A prompt panel is a fixed set of questions for a market, sent to every engine on a schedule, with each question asked more than once. Every answer is stored and parsed for which businesses it names, which it recommends, and which URLs it cites. From that you get:

• Share of voice per engine: the share of answers that name each business, week by week.
• Citation sources: which websites the engines read for your market. In our Phoenix study the engines cited 154⁹ different websites. The ones that recur for your questions are where your business needs to be described correctly.
• Trend and cause: when a number moves, you can line it up against the work done that week.

The kinds of tools that do it

You do not have to build this. There are three broad options, and each has a place.

AI visibility platforms. Dedicated tools such as Profound, Peec AI, Otterly.AI, Scrunch, and AthenaHQ run prompts across AI engines and report mentions, competitors, and cited sources. They suit a business with someone to read the reports and act on them.

SEO suites with AI tracking. Semrush and Ahrefs have added AI visibility tracking to their existing platforms. If you already pay for one of them, this may be the easiest place to start.

Services that track and act. Some firms run the measurement and do the fixing. CiteMy is one: every week we ask ChatGPT, Gemini, Perplexity, and Claude the questions homeowners in your market ask, several times each, and report which businesses each answer names, cites, and recommends, and which sources it cites, alongside the work done that week.

When you compare options, ask how each one chooses its questions, whether it asks each question more than once, which engines it covers with search enabled, and whether you can see the raw answers. A share of voice without the questions behind it is a number you cannot check.

What tracking cannot tell you

Tracking tells you whether you are named. It does not guarantee you will be. The engines decide what they say, and no tool or service can promise a position in an answer. What good tracking gives you is honest feedback: whether the work you did changed how often the engines name you.

What this means for an owner

Start with a manual log for a few weeks. You will learn which questions matter, which competitors keep appearing, and which websites the engines lean on. Then decide whether you want software you run yourself or a service that runs it for you.

If you want a starting point, our free AI visibility audit checks the technical side (crawler access, structured data, and whether your name, address, and phone agree across the web) and emails you the findings, alongside where your trade stands in our weekly measurements.