AI visibility without measurement is just hope.
A business can spend months improving schema, cleaning directories and writing answer-shaped content — and never know whether any of it changed what an AI assistant says. Measurement closes that loop. It tells you where you started, what improved, what did not, and what needs work next.
It also tells you about your competitive position. If you query ChatGPT for "best [your service] in [your city]" and a competitor's name appears consistently while yours does not, that is a data point — not speculation. You know exactly what the model's current answer is, and you know what signals drive that answer. The measurement becomes the brief for the fix.
This matters more every month. Buyers are increasingly asking an assistant first and visiting a search results page second, if at all. The businesses appearing in those answers are not there by accident. They have built the right entity signals, whether deliberately or not. Tracking your visibility lets you close the gap deliberately — see AI visibility audit explained for the full diagnostic approach.
How to run a repeatable AI visibility check.
There is no official tool for this yet. The method is structured, repeatable querying across the main assistants, scored consistently each time you run it.
Build a query set.
Write 10 to 15 queries that represent what your customers actually ask: category + city, specific service + neighbourhood, comparison questions ("who is the best X in Y"), problem-led questions ("where can I get X fixed in Y"). Include both broad and specific variations. These stay fixed across every measurement cycle.
Query each assistant fresh.
Run each query in ChatGPT (with and without web search enabled if you have access), Perplexity, and Gemini. Use a fresh session for each to avoid personalisation effects. For Google AI Overviews, run the queries directly in Google Search and capture the overview panel when it appears.
Score each response.
For every query-assistant combination, score four things: Are you mentioned at all? Is your business name correct? Is your category or service described accurately? Is your location correct? A simple four-point score per response adds up to a visibility index you can track over time. Record the names of any competitors that appear instead of you.
Document and date-stamp everything.
Copy each response to a structured log with the query, assistant, date and score. Screenshots are useful backup. The log becomes the trend line — after three to four cycles you can see whether GEO improvements are moving the scores or whether a different lever needs pulling.
Run on a regular cadence.
Monthly is a reasonable starting cadence for most businesses. Models re-index and update frequently enough that a monthly check captures meaningful movement. If you are in an active GEO improvement period, bi-weekly gives more responsive feedback.
Three metrics worth tracking across every cycle.
Not all AI visibility is equal. A mention with a wrong description is worth less than no mention at all in some cases. These three metrics capture the quality of visibility, not just its presence.
How often does your name appear at all?
Expressed as a percentage across your query set and assistants. If you run 15 queries across 3 assistants (45 total responses) and appear in 9, your mention rate is 20%. This is the headline number and the clearest proxy for whether your GEO work is gaining ground. Track it query by query to see which types of question you win and which you lose.
When you are mentioned, is it correct?
A model can name you while describing the wrong service, the wrong location, or with outdated information. These inaccurate mentions can actively harm you — a buyer who hears you are in the wrong neighbourhood may not follow up. Score each mention for name accuracy, service accuracy and location accuracy. Inaccuracies point directly to entity clarity gaps.
Who is appearing where you are not?
For every query where you are not named, record who is. This competitive picture is as valuable as your own scores — it shows you which businesses the model has decided are the strongest entities for your category, and lets you audit their signals to understand the gap. Often the leaders have simply done the entity work more thoroughly, which is a solvable problem.
What tools exist and what they can't do yet.
Several tools have emerged claiming to automate AI visibility tracking. It is worth knowing what they do and where they fall short.
Tools like Semrush's AI tracking features, BrightLocal's AI Search Grader and various newer GEO-specific platforms attempt to automate querying and scoring. They are useful for scale — running hundreds of queries across many locations — but they are not reliable for fine-grained accuracy scoring, they do not capture the full nuance of how different assistants phrase answers, and they lag behind the model updates that change what assistants say week to week.
For most local and regional businesses, manual structured querying is still the most honest method. It takes two to three hours per cycle once you have a fixed query set, and it gives you verbatim responses you can analyze rather than aggregated scores from a black box. As the tooling matures, a hybrid approach — automated mention detection plus manual accuracy scoring for flagged responses — will likely become the standard. For now, build the habit of looking directly at what the assistants say on the queries that matter to your business.
Related, and what to do next.
Questions, answered.
How do I measure AI visibility for my business?
Build a set of 10 to 15 queries representing how your customers ask about your category and location. Query ChatGPT, Perplexity and Gemini with each, in fresh sessions. Score each response: are you named, is the description accurate, is the location right? Track this monthly. The mention rate and accuracy score across that query set is your AI visibility baseline.
Is there a tool that tracks AI visibility automatically?
Several tools now attempt automated AI visibility tracking, including Semrush's AI features and newer GEO-specific platforms. They are useful at scale but not yet reliable for nuanced accuracy scoring. For most businesses, structured manual querying gives more actionable data than automated tools do today.
How often should I check what AI assistants say about my business?
Monthly is a reasonable baseline cadence. AI models update frequently enough that monthly checks capture real movement. If you are actively doing GEO work, bi-weekly checks give more responsive feedback on whether specific improvements are taking effect.
What counts as good AI visibility?
There is no universal benchmark, but a useful target is being named accurately in response to at least half of your core category and location queries across the main assistants. The more specific the query, the higher the mention rate should be — if someone asks "best dental clinic in Scarborough" and you are a dental clinic in Scarborough, not appearing is a clear gap.
Why do different AI assistants give different answers about my business?
Each assistant uses different data sources, indexes different content, and weights signals differently. ChatGPT with web search, Perplexity and Gemini all retrieve and rank sources differently. This is why measuring across multiple assistants matters — you want visibility on the platforms your customers actually use, not just the one you checked.
What should I do if AI assistants describe my business incorrectly?
Inaccurate descriptions usually trace to a specific source the model read. Check your Google Business Profile description, your website's About or Services pages, and your schema markup. Find the version that matches what the model said, update it to be accurate, and add the correct information more prominently in multiple sources. Re-check in four to six weeks.