Which numbers actually tell you whether AI tools know your organization — and which are noise. A plain-language reference to the metrics behind AI visibility, built for a small team measuring by hand.
By Matt·Updated July 11, 2026
Quick answer
What are GEO metrics?
GEO metrics measure how often, how prominently, and how accurately AI tools like ChatGPT, Gemini, and Google AI Overviews mention or cite your organization in their answers. For most small teams, four are worth tracking: mention rate, citation rate, share of voice, and accuracy — all measurable with a fixed list of monthly test questions and a spreadsheet, no paid tools required.
Unlike classic SEO metrics, which track people clicking from a results page to your website, GEO metrics look at what happens inside the AI answer itself.
Why traditional tracking is no longer enough
Your analytics show visits to your website. AI tools increasingly answer people before they ever get there.
Classic measurement — keyword rankings, clicks, organic traffic — assumes people travel from a search results page to your site. AI answer engines break that assumption by answering directly. A 2025 Pew Research Center study of real browsing behavior found that when a Google search produced an AI summary, people clicked a traditional result only 8% of the time — roughly half the rate of searches without one. Someone can learn about your food pantry's hours, your association's certification, or your clinic's sliding-scale program entirely inside an answer, and your analytics will never see them.
That creates an odd but common situation: your website traffic can fall while your actual reach grows, because AI tools are recommending you in answers you can't see. GEO metrics exist to close that gap. Two things make them different from the numbers you're used to. First, AI answers vary from run to run — asking the same question twice can produce different responses — so these metrics describe frequency across many checks, not a fixed position like "ranking #3." Second, they're organization-centered: they track how strongly AI systems associate your name with your topics, whether or not a link to your site appears. (For the full measurement routine, start with how to measure AI visibility.)
The metrics, defined
Six terms you'll meet again and again. The first four are the ones most small teams should actually track.
Mention rate
Of your monthly test questions, how many produced an answer that named your organization — with or without a link? This is the baseline: proof the engines know you exist. (More in understanding AI mentions.)
Citation rate
How often an answer links to your website as a source. Citations are the closest thing generative search has to "ranking," and they're what send visitors your way. (Tracked in depth in AI citation tracking.)
Share of voice
When AI tools answer questions in your space, how often do they name you versus your closest peers? If answers about services in your county name three organizations and you're never one of them, that's the finding.
Accuracy & sentiment
Is what the engines say about you true and current — right programs, right hours, right leadership — and how is it framed? High visibility with wrong information is worse than no visibility at all.
AI referral traffic
Visitors who arrive at your site by clicking a link inside an AI answer. Usually a small number, but a telling one. (How to find it: AI referral traffic.)
Impact measures
Whether visibility turns into real-world results — donations, registrations, program sign-ups, applications — plus lifts in people searching your name directly after encountering you in an AI answer.
The four metric tiers
Each tier answers a different question. Start at presence and work right — the later tiers mean little until the earlier ones exist.
AI companies don't publish what people ask in private chats, so every presence and quality metric comes from your own sampling. The routine takes one to two hours a month.
1
Write a fixed list of 10–15 real questions
Questions your audience actually asks: about your topic ("where can families get food assistance in [county]?"), your services ("does the library have free meeting rooms?"), and your organization by name. Freeze the wording — consistency is what makes the numbers comparable month to month. The full method is in how to measure AI visibility.
2
Run them in each AI tool, monthly
ChatGPT, Google (watching for AI Overviews), Gemini, and Perplexity cover the landscape; free tiers are fine. Use fresh or logged-out sessions so your own history doesn't color the answers. Paid monitoring platforms can automate this at scale and may make sense for large organizations tracking hundreds of queries — but they're an upgrade, not a requirement.
3
Record four things per answer
Mentioned (yes/no), accurate (yes/partly/no), website cited (yes/no), and any peer organizations named instead. One spreadsheet row per question per tool, dated. The AI visibility scorecard gives you a ready-made structure.
4
Turn tallies into the four numbers
Simple division. Mention rate: answers naming you over total answers. Citation rate: answers linking your site over total. Share of voice: your mentions over all organization mentions in your topic questions. Accuracy: the share of your mentions that were fully correct. Nothing fancier than a percentage.
5
Act on what you find, then repeat
Fix factual errors at the source pages immediately, note the questions where peers appear and you don't, and feed both into your content plan. Then run the same questions next month and watch the trend — a single month's numbers mean little on their own.
What the metrics look like in practice
Two representative scenarios showing why the tiers are read together, not separately.
The food bank whose traffic fell while its reach grew
What the numbers showed
A regional food bank sees visits to its assistance pages decline and assumes it's losing ground. Its first monthly check tells a different story: most AI answers about food assistance in its counties name the food bank and summarize its programs correctly — people are simply getting what they need inside the answer.
What the team did
Instead of chasing the lost clicks, the team made the click worth taking: the pre-screening form and current distribution calendar moved to the top of every program page — things an AI summary can point to but can't replace. Mention rate and accuracy became the numbers leadership sees, with traffic as context rather than the headline.
The lesson: presence metrics reframe falling traffic. Without them, this team would have "fixed" pages that were quietly doing their best work ever.
The accounting firm with tiny-but-golden AI referrals
What the numbers showed
A 20-person accounting firm finds AI referrals are a sliver of its GA4 traffic — easy to dismiss. But reading further, those visitors behave differently: they arrive having already read a summary of the firm's specialties, go straight to service pages, and request consultations far more readily than search visitors.
What the team did
The firm treated the engagement tier as its signal and invested in what earns citations: real client questions answered plainly on service pages, and structured data so engines can parse who the firm serves. The goal wasn't more AI traffic — it was more of the same kind.
The lesson: judge AI referral traffic by quality, not volume. A small, pre-informed audience can matter more than a large casual one.
What these metrics can't tell you
Four honest limits to keep in mind before you put any of these numbers in a report.
There's no first-party data. AI companies don't share what people ask in private conversations, so your presence and quality numbers come from your own sample of test questions — a deliberate simulation, not a census. That's still useful; it's just worth being honest about in anything you send up the chain (see reporting results to leadership).
Analytics undercounts AI visits. Some AI tools pass a referral source GA4 can see; others send nothing, and those visits land in "Direct." Your measured AI referral traffic is a floor, not a total. Search Console has a related blind spot: it counts AI Overviews activity inside its normal reports without separating it out.
Answers vary from run to run. The same question can produce different responses on different days — that's how these systems work, not a flaw in your method. Read trends across months, never a single check.
Visibility and accuracy can point in opposite directions. A high mention rate built on an outdated news story or old reviews is a liability, not a win. This is why the quality tier exists — and why accuracy is the one number that demands action the moment it slips.
Start tracking this week
Five steps from zero to a working set of GEO metrics.
One evidence note for the skeptics on your team: the content changes these metrics reward are researched, not folklore. The original academic work on generative engine optimization — Aggarwal et al., presented at KDD 2024 — found that adding citations, quotations, and statistics improved a source's visibility in AI-generated answers by up to 40% in benchmark testing (more on applying that in create citation-worthy content). And if you'd rather have the tracking set up and run for you, tell me about your situation.
Common questions
What is the difference between an AI mention and an AI citation?
A mention is when an AI tool names your organization inside its answer. A citation is when the answer links to your website as a source. Both matter: mentions build awareness even when nobody clicks, while citations send interested visitors to your site. Track them as separate numbers, because they move for different reasons.
How do I see traffic from AI tools in Google Analytics 4?
Some AI tools pass a referral source GA4 can see — visits from ChatGPT and Perplexity usually appear under Referrals as chatgpt.com and perplexity.ai. Others send nothing, so those visits land in Direct. Check your referral report for known AI domains, and treat whatever you find as a floor, not a complete count.
Does ranking well in Google mean AI tools will cite me?
Not automatically. Strong, trustworthy pages help with both, but AI tools favor content that answers questions directly and states facts plainly. A page that ranks well in classic search can still be passed over if its answer is buried under introductions — and a modest site with clear, well-structured answers can out-cite a bigger one.
How often should a small team check these metrics?
Monthly. That is frequent enough to catch shifts when AI models update, and light enough — one to two hours with a fixed question list — for one person to sustain. Checking more often mostly measures random variation in the answers, not real change.
Do different AI tools need different tracking?
The routine is the same for all of them — same questions, one column per tool. As of mid-2026, Perplexity and Google AI Overviews lean on live web results, so they reward fresh, clearly structured pages; ChatGPT blends built-in knowledge with live search, so consistent information about your organization across the wider web matters more there. Tracking each tool separately shows you these differences in your own numbers.