GEO Reference
GEO metrics, defined.
The working vocabulary of GEO: the four metrics that tell you whether AI search is working for you, and the engine, measurement, and technical terms around them, each defined with examples.
Reference · Last updated
This is the working vocabulary of Generative Engine Optimization: the four metrics Resonate Labs uses to measure GEO outcomes, plus the engine, measurement, and technical terms that buyers, marketers, and engineers keep running into. The terms blur together easily, and a program can move one while another sits flat, so here's each, precisely, with an example.
The four outcome metrics
The metrics that answer the only question that matters: when your buyers ask AI which vendor to choose, do you show up, and are you the answer? Each reads separately across Gemini, Claude, ChatGPT, and Perplexity, because every engine has its own source diet.
Visibility rate
The share of tracked queries where an AI engine mentions you at all. It's the floor. Being visible means you made it into the answer, not that you were recommended. A vendor can be highly visible and rarely chosen.
Example: you appear in 40 of 150 tracked queries, a 27% visibility rate.
Citation rate
The share of responses where an engine references one of your pages as its source. This is distinct from visibility. Visibility means you were named; citation means your content was the thing the engine pulled from. Citation is what on-page content work most directly moves.
Example: an engine names you in an answer but links a third-party article as its source. That's visibility without citation.
Share of voice
Your portion of all vendor mentions across the full query set: presence measured against the competitors buyers are comparing you to, not in isolation. Rising share of voice means you're taking conversational space from alternatives.
Example: across all queries, 200 vendor mentions occur and 30 are you, a 15% share of voice.
Win rate
Of the queries where you're visible, the share where you're the primary recommendation: the vendor the engine surfaces first, or tells the buyer to evaluate first. Win rate isolates quality of position from breadth of presence.
Example: you're visible in 40 queries and the lead recommendation in 10, a 25% win rate.
GEO, and how the engines answer
The terms for the systems the metrics run against, and the mechanics that decide whether you get named.
Generative Engine Optimization (GEO)
The practice of improving how a brand is represented, cited, and recommended inside AI answer engines. Where SEO optimizes for a ranked list of links, GEO optimizes for the generated answer itself: whether the engine names you, sources your pages, and puts you on the shortlist. What is GEO? is the full explainer.
Answer engine
An AI system that responds to a question with a synthesized answer rather than a list of links. The four that matter for B2B buying are ChatGPT, Claude, Google's AI Overviews and AI Mode, and Perplexity. Also called a generative engine or AI search.
Large language model (LLM)
The model underneath an answer engine (GPT, Claude, Gemini, and others) that generates the text of the answer. Whether it names you depends both on what it learned in training and on what it retrieves at answer time.
Retrieval-augmented generation (RAG)
The step where an engine fetches live web pages and grounds its answer in them, rather than relying only on training memory. Also called grounding. It's why fresh, well-structured pages can be cited within days, and why on-page content work moves citation at all.
Source diet
The set of sources a given engine tends to pull from. Each engine has its own: Perplexity leans on the live web, Google on its own index, and the others weight differently. This is why a page cited heavily in one engine can be invisible in another, and why every metric is read per engine.
Query fan-out
When an engine silently expands one buyer question into several sub-queries, retrieves for each, and synthesizes across them. Google's AI Mode does this. It means you're competing on questions the buyer never typed, only implied.
AI Overview and AI Mode
Google's generative answers. AI Overviews is the summary that appears above the classic blue links; AI Mode is the fuller conversational surface. Both can name and cite vendors, and both are measured separately from ChatGPT, Claude, and Perplexity.
Zero-click
When the buyer gets their answer inside the engine and never visits a website. Zero-click is why being named and cited in the answer matters more than ranking a page the buyer never clicks.
How your brand appears
Being in the answer isn't one thing. These terms separate the ways an engine can name, frame, or misdescribe you.
Brand mention
Any time an engine names your company in an answer. A mention drives visibility but is not the same as a citation: the engine can name you while sourcing a third-party page. Unlinked mentions still shape the buyer's shortlist.
Entity
Whether an engine recognizes your brand as a distinct, well-defined thing, a company with a category, products, and people, rather than a loose string of words. Strong entity recognition is what lets an engine confidently recommend you; a thin entity gets skipped even when your pages are good.
Sentiment
How an engine characterizes you when it names you: the safe leader, a risky unknown, the budget option. Two vendors can share a visibility rate while one is described in language that wins the deal and the other in language that loses it.
Recommendation position
Where you fall when an engine lists options. Being named third in a list of five is visibility; being the one it says to evaluate first is a win. Order is a signal buyers act on.
Hallucination
When an engine states something false about your brand: a feature you don't have, a wrong price, a competitor's fact attributed to you. Finding and correcting hallucinations is the defensive side of GEO, covered in Defensive GEO.
Measurement and tracking
The terms for turning single readings into a trend you can act on. The full methodology, cadence, and attribution live on How We Measure GEO Results; these are the definitions.
Buyer-intent query
A question a real buyer would ask an engine while choosing a vendor, from problem identification ("how do I fix X") through vendor shortlisting ("best tool for X in B2B SaaS"). GEO is measured on these, not on branded or navigational searches.
Baseline (fixed query set)
A stable, unchanging set of queries scored the same way every cycle, so movement is comparable over time rather than an artifact of asking different questions. Without a fixed baseline, month-to-month numbers are noise.
Leading and lagging indicators
The two speeds of GEO signal. AI crawler traffic is leading: engines pull new content within days. Citation visibility is lagging: it moves once the engines have indexed and started citing. Reading them together keeps a working program from looking like a failing one.
AI crawler
The bots that fetch your pages for answer engines: GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and others. Their hits in your server logs are the earliest measurable sign that new content is being pulled.
AI referral traffic
Sessions that arrive on your site from an answer engine's citation link. Lower in volume than search, but often higher-converting, because the visitor arrived pre-qualified by the answer that sent them.
AI impressions
The impressions Google Search Console now reports for your pages inside AI features on Search. A free, real leading indicator for the Google surface, though it covers only Google and not yet click data for AI features.
Platform divergence
The fact that the four engines return different answers, cite different sources, and rank vendors differently for the same query. It's why a single blended "AI visibility score" hides the truth and every metric is scored per engine. Platform divergence covers why the score has to stay split.
On-page and technical
The terms for the content and markup that decide whether an engine can fetch, parse, and lift your pages. The working checklist is the Technical GEO Checklist.
Structured data (schema markup)
Machine-readable tags (Schema.org JSON-LD) that tell engines what a page is: an article, an FAQ, an organization, a defined term. Structured data helps an engine parse and trust your content, which supports citation.
Crawlability
Whether AI crawlers can actually reach and read your pages: not blocked in robots.txt, not hidden behind JavaScript the crawler won't run, served fast and clean. Content an engine can't fetch can't be cited. Why AI Can't See Your Site is the deeper diagnosis.
Atomic content
A self-contained passage that answers one question completely, so an engine can lift it into an answer without needing the rest of the page. Answer engines cite passages, not whole documents, so content structured as atomic answers gets cited more.
llms.txt
A proposed root-level file, like robots.txt, that points AI crawlers to the pages a site most wants surfaced. Adoption is early and uneven, so it's a low-cost signal, not a guarantee.
Frequently asked questions
What's the difference between visibility and citation?
Visibility means an AI engine mentioned you; citation means it used one of your pages as the source. You can be visible without being cited, and citation is the metric on-page content work moves most directly.
What's the difference between share of voice and win rate?
Share of voice is your portion of all vendor mentions across the full query set, so it measures breadth of presence against your competitors. Win rate looks only at the queries where you appear and asks how often you're the primary recommendation, so it measures quality of position. You can lead on one and trail on the other.
Is a high visibility rate good on its own?
Not by itself. Visibility is only the floor. A vendor can appear in many answers and still rarely be the recommendation. Win rate is what tells you whether you're the one being chosen.
What are the core GEO metrics to track?
Four outcome metrics carry the most signal: visibility rate (are you in the answer), citation rate (is your page the source), share of voice (how much of the conversation is yours), and win rate (when you appear, are you the recommendation). Each is read per engine, because the four engines cite differently. Crawler traffic and AI referral traffic are useful leading indicators alongside them.
How does Resonate Labs measure these metrics?
On a fixed set of 150 buyer-intent queries, scored every 30 days across Gemini, Claude, ChatGPT, and Perplexity. The full method is on How We Measure GEO Results.
Go deeper
See how these get measured.
The methodology behind these metrics, the query set, the monthly cadence, leading and lagging signals, is on the measurement pillar.
Free, delivered in 3 business days. No pitch, no commitment.