THE SHORT ANSWER
Measure four things. Citation share, meaning how often you are named across a fixed list of buyer prompts run monthly against each assistant. Referral sessions from assistant domains in your analytics. Answer accuracy, meaning whether what the assistant says about your prices and services is correct. And assisted conversions, since assistant traffic converts differently from organic. There is no rank position to track, and any tool claiming to provide one is inventing it.
The first instinct is to look for the AI equivalent of a rank tracker. It does not exist, because there is no ranking. Answers are generated per query, per session, sometimes per user, and two people asking the same question ten seconds apart can get different citations.
That sounds like it makes measurement impossible. It does not. It makes it statistical rather than positional, which is a familiar problem with a familiar solution: fix the sample, repeat on a schedule, and watch the trend rather than the point.
The numbers, at a glance
Prompt set size: twenty to fifty buyer questions, fixed and unchanged between runs
Run frequency: monthly is enough; weekly produces noise you will misread as signal
Citation share: the share of runs where you are named, tracked as a percentage over time
Referral share: assistant sessions typically one to five percent of total for service businesses
Metric one: citation share against a fixed prompt set
Write down twenty to fifty questions a real buyer would ask. Not keywords, questions, in the phrasing a person actually uses. Include the commercial ones you care about most and the comparison ones where you would want to be mentioned alongside competitors.
Run the whole set against each assistant on the same day each month, from a clean session with no personalisation. Record, per prompt: were you named, were you linked, were you described accurately, and who else appeared. Citation share is the percentage of prompts where you were named. It is a blunt number, and it is the only one that moves in a way you can attribute to your own work.
The discipline that matters most is not changing the prompt set. The moment you add prompts you have made this month incomparable with last month, which is how most teams destroy their own baseline in the second quarter.
Metric two: referral sessions, and why the number looks disappointing
Assistant referrals appear in analytics under identifiable hostnames: chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, claude.ai. Segment them into one channel group and track sessions, pages per session and conversion rate separately from organic.
Expect a small number. One to five percent of sessions is normal for a service business, and anyone reporting thirty percent is either tiny or miscounting. Judge the channel on conversion rate rather than volume, because the whole argument for GEO is composition, not scale. If assistant traffic is not converting better than organic, the problem is usually that the assistant is describing you inaccurately, which is metric three.
Metric three: answer accuracy, the one most teams skip
Ask each assistant directly what your company does, what it charges, where it operates and who it serves. Score each answer as correct, partially correct or wrong, and log the specific error.
This matters more than citation share for anyone with published pricing, because a confidently wrong price does measurable commercial damage. It is also the fastest thing to fix: wrong answers almost always trace back to an outdated page, an inconsistent description on a directory, or a price that exists only inside an image or a PDF where no crawler can read it.
Metric four: assisted conversions and the attribution problem
Assistant influence is systematically under-counted, because a common pattern is: ask assistant, read answer, then search the brand name directly and arrive as organic or direct. The assistant gets no credit and branded search gets a mysterious lift.
Two practical counters. Add a plain "how did you hear about us" field to your enquiry form with an explicit option for AI assistant, which is crude but catches the direct-navigation path. And watch branded search volume against citation share over several months, because the two move together when GEO is working, with the branded lift trailing by roughly a month.
A measurement setup you can stand up in an afternoon
Write the fixed prompt set and store it in a sheet with one row per prompt.
Add columns per assistant per month: named, linked, accurate, competitors present.
Create an analytics channel group for the five assistant hostnames.
Add an attribution question to your enquiry form with an AI assistant option.
Diary the run for the same date each month and do not change the prompts.
Want leads like this in your pipeline?
Flock runs the campaigns, screens the enquiries and hands you only the ones that match your service area, job size and capacity. You pay per lead, not per month.
Book a 15-minute fit check | See lead package pricing
Related answers
Frequently asked questions
Are the paid AI visibility tools worth it?
They automate the prompt runs, which is the tedious part, and that is a legitimate saving at scale. Be sceptical of any tool reporting a position or a share-of-voice score, since neither concept exists in a generated answer; the underlying data is still just did-you-appear, counted.
Why do I get different answers to the same question?
Generation is probabilistic, retrieval is live, and personalisation varies by account and location. This is exactly why the method is a fixed sample repeated over time rather than a single check.
How long before the numbers move?
Citation share on browsing-capable assistants can shift within weeks of publishing. Anything dependent on model training data moves on the model release cycle, so allow two to three quarters before judging.
What is a good citation share?
There is no benchmark, because it depends entirely on how commercial and how contested your prompt set is. Judge yourself against your own first measurement, not against a number from someone else's market.
NEED A CLEARER PLAN?
Let’s turn your next move into momentum.
Talk to us →