TLDR
You cannot manage AI visibility without measuring it. The measurement is asking the assistants what your customers ask, on a schedule, and logging who gets named.
The system needs three parts: a fixed question set, a monthly protocol, and a scorecard that shows the trend.
A spreadsheet and one hour a month cover it. Discipline beats tooling here.
What is an AI mention?
Any time an assistant names your business in an answer a customer could plausibly ask. That includes direct recommendations like "who is a reliable electrician in your town," comparisons between local companies, and answers that cite your website as a source. Each one is a small referral happening outside your analytics.
Mentions are the output metric of GEO. Everything else, from schema to reviews, exists to move this number.
How do you build the question set?
Write ten to fifteen questions that mirror real buying moments in your market. Cover three intents: the direct ask ("best HVAC company near [your town]"), the problem ask ("who fixes an AC blowing warm air in [area]"), and the research ask ("how much does duct cleaning cost in [city]").
Use your town names the way locals say them, and freeze the wording. The set only works as a measuring stick if the questions stay identical month to month.
What is the monthly protocol?
Same questions, same assistants, same week each month. Run the set through ChatGPT, Gemini, and the AI features in Google search. For each question log whether you were named, which competitors appeared, and any source the answer cited.
Take screenshots as you go. They become your before-and-after evidence, and they settle any doubt about what an answer actually said.
What does the scorecard track?
Three numbers tell the story. Mention rate is the share of questions where you were named. Average position is where you land when named. Competitor share shows who owns the answers you do not. Chart all three monthly and the trend line becomes your GEO report card.
Pair the scorecard with your business numbers, especially calls and branded searches, since mentions that never move the phone are trivia. The connection usually shows up a month or two behind the visibility.
How do you act on what you find?
Losses are assignments. When assistants name competitors for a service you offer, study what the machines can read about them. Look for stronger pages, better profiles, or cleaner structured data. If answers cite a directory instead of any business, that directory is a listing you should hold and polish.
Then feed the gaps into your content plan. Every unclaimed question in your log is a page waiting to be written to the people-first standard that earns citations.
When does the trend start moving?
Expect the first movement within weeks on assistants that read the live web, and slower shifts where answers lean on trained knowledge. The pattern that matters is direction over quarters, not noise between months. Businesses that run this loop honestly for two or three quarters almost always watch their mention rate climb, because most competitors are not even keeping score.
Frequently asked questions
Do answers change between runs even when nothing changed?
Yes, some variance is normal because these systems are probabilistic. That is why the scorecard tracks a fixed set monthly and reads quarterly trends, which smooth the noise into signal.
Should I use incognito or a logged-in account?
Keep it consistent, whichever you pick. A clean browser profile reduces personalization effects, and consistency month to month matters more than any single setup choice.
Is there a tool that does this automatically?
Tools for AI mention tracking exist and mature quickly, and FoundLocal runs this tracking for clients as part of the program. The manual spreadsheet version remains the best free start, because it forces you to read the actual answers.
Rette leads FoundLocal and has spent the better part of a decade getting US home service businesses found online. FoundLocal is where the full playbook lives.