To track whether ChatGPT recommends your business, ask it the questions your customers ask, the same way each time, on a schedule. From every answer, log three things: whether you're named, whether you're actually recommended, and who is named instead. One answer proves little, because ChatGPT's answers vary from run to run. The number to watch is the share of checks that name you, over several weeks.
What counts as a recommendation?
ChatGPT can mention your business in several ways, and they aren't worth the same. Recommended means the answer suggests you: "For emergency repairs, Summit Plumbing is a strong choice." Named means you appear, perhaps in a list, without being singled out. Cited means one of your pages is among the sources the answer drew on, whether or not your name made it into the text. Absent means no trace of you at all.
Log the strongest of these that each answer reaches. A month of "named but never recommended" tells you something different from a month of "absent", and the fix is different too. More on these levels of visibility.
Which questions should you track?
Start with eight to ten questions a customer might ask before calling you. Mix three kinds:
- Who to call. "Who's the best roofer in Lexington?" or "Which plumber in Daphne handles emergency calls?"
- A problem. "My water heater is leaking. Who can fix it today in Daphne?"
- A decision. "Should I repair or replace a 15-year-old roof in Lexington, and who does both?"
Write them the way customers talk, and put your town in every question. Local answers depend on place, and a question without one can return businesses from anywhere. Keep the wording fixed from week to week, or you end up measuring your phrasing instead of ChatGPT. How to choose your money questions goes deeper on picking them.
How do you ask fairly?
Your own ChatGPT account remembers what you've told it, and past conversations about your business can color its answers. Run your checks in a session that carries none of that, such as a signed-out browser window, so you see something closer to what a new customer sees.
For local questions, ChatGPT often searches the web before it answers, and it can only show pages its search crawler is allowed to read. OpenAI's crawler documentation says sites that block OAI-SearchBot won't appear in ChatGPT search answers. So check your robots.txt and firewall before you start counting. Our crawler guide shows how.
What should you write down?
A spreadsheet is enough. Each time you ask a question, record:
- the date and the exact question;
- whether you were recommended, named, cited or absent;
- where you appeared among the businesses named, if you did;
- which of your pages, if any, appeared as sources;
- every other business named, so your competitors show up in the data too.
Save the full answer text as well. When the result changes, the wording usually tells you why: a competitor praised for reviews you don't have, or a service you offer that the answer says nobody nearby does.
How do you read the results?
Answers vary between runs even when nothing has changed, so don't react to one. Look at the share of checks where you're named or recommended, question by question, over four weeks or more. Treat a change as real when it holds. If you were named for weeks and then absent in two of the last three checks, that's a slip worth looking into, not noise.
Then read the pattern question by question. Being absent everywhere points to access, or to how your business is described across the web. Being named for "who to call" questions but not for problem questions usually means your pages don't answer those problems directly. What shapes ChatGPT's picks covers what to change.
Can you see ChatGPT's recommendations in your analytics?
Partly. When someone clicks a link in a ChatGPT answer, the visit reaches your site. On speaklocal.com most of those visits arrive with utm_source=chatgpt.com on the link, and some show chatgpt.com as the referrer. In Google Analytics, look for either one under session source.
Those visits only show the recommendations someone clicked. Some people will act on an answer without ever visiting your site. So treat analytics as a floor, and the checks above as the real measure.
When is a tool worth it?
Ten questions a week is forty ChatGPT checks a month, and the same again for each other assistant you care about. By hand, that's two or three hours a month of asking, copying and logging, and it's the first habit to slip when work gets busy.
That's the job tracking tools do. We make one: Speak Local asks your questions to ChatGPT, Perplexity and Gemini on a weekly or monthly schedule, keeps every answer word for word, and records each one as named, cited, read or absent, with competitors marked beside you. Other tools handle parts of this differently, and we compared the main options, including where ours falls short. If you're choosing one, here's what to check before you pay.