How to track whether AI recommends you over time
Updated October 15, 2026
Pick a fixed set of buyer questions, ask all four assistants each one on a schedule, and write down every business named, not only whether you were. Compare the count from run to run. Answers vary, so one run proves little; the trend across several runs is what tells you whether your work is landing.
Which questions should I track?
The ones your customers ask, in their words. Start with the main jobs you want to win, then phrase each one a few ways: a short search, a conversational question, and a comparison. Each phrasing pulls back a different list.
We use 12 buying intents phrased four ways each, which gives 48 questions per business. You can start smaller by hand. Just keep the list fixed, or you'll be comparing different tests.
How do I run the questions fairly?
Use a fresh chat each time, with history and memory turned off, or have someone outside the business run them. Your own account knows who you are and where you are, and it can tilt the answer toward you.
Put the city in the question. Run each question in ChatGPT, Claude, Gemini and Perplexity, because they disagree far more often than they agree.
What should I record?
A spreadsheet is enough. One row per question per assistant per run, with:
- the date, the assistant and the exact question
- every business named, in order
- whether you were named
- any sources the answer cited
How often should I check?
Monthly is enough to see a trend if you're not making changes. Weekly is better while you're working through fixes, because you see sooner whether a change moved anything.
Don't react to one run. A drop from 5 questions to 3 in a week can be noise. A climb from 3 to 9 over two months probably isn't.
How do I read the trend?
Look at the count across all questions and all four assistants, and at which questions you gained or lost. A new page for a service should show up as gains on the questions about that service, not everywhere at once.
Track the businesses named instead of you as well. If the same two competitors win every question about a service, look at what their pages and listings have that yours don't.
Can I have it done for me?
Yes. Our free plan asks all four assistants 48 questions for your city every month and emails you the score. Starter at $29 a month checks weekly and shows the questions you gained and lost each week. Pro at $49 a month adds 3 competitors tracked week by week and up to 10 questions of your own.
Check your business for free
Enter your website. We ask ChatGPT, Claude, Gemini and Perplexity the questions your customers ask and email you who they recommend.
Common questions
How many questions do I need to track?
Enough that one odd answer doesn't swing the result. A dozen is a reasonable start by hand. We use 48 because it covers the main jobs in a trade with several phrasings each.
Why did my score drop when I didn't change anything?
Answers vary between runs, a competitor may have changed something, or an assistant may have been updated. Look at the trend over several runs before you read anything into one drop.
Should I track Google AI Overviews too?
If your buyers use Google heavily, it's worth a look, but keep it separate from the four assistants so you're comparing like with like.
Is there a tool that tracks AI recommendations automatically?
Yes, several, including ours. Whatever you use, check that it asks all four assistants, uses buyer questions with your city in them, and shows you the saved answers rather than only a score.