geo

AI Visibility Metrics: A Glossary for Measuring GEO

A field guide to the metrics that actually explain your brand's standing in ChatGPT, Claude, Gemini, and Perplexity — and how each one gets measured.

AD

Alex Dabson

Founder, DabaRank

8 min read

Ask five people what "AI visibility" means and you'll get five different dashboards. One means "does ChatGPT mention us at all." Another means "are we cited more than our competitor." A third is tracking sentiment in a spreadsheet nobody else has seen. None are wrong — they're just measuring different things and calling it the same name.

That looseness is the real problem. When we onboard a brand and start running prompts across the major AI platforms, the first conversation is rarely about the numbers — it's about agreeing what the numbers mean. A "mention" is not a "citation," and share of voice in AI answers isn't calculated like share of voice in a press clipping report. Get the definitions wrong and you'll report progress that isn't real, or miss a decline that is.

TL;DR

GEO measurement rests on nine core metrics: presence, share of voice, rank, citation frequency, citation share, sentiment, prompt coverage, competitor co-mention, and source attribution. Each requires running real prompts against real models on a schedule — there's no console to pull this from.

Quick reference

| Metric | What it measures | What "good" looks like | |---|---|---| | Presence / Mention Rate | % of prompts where your brand appears | Rising on unbranded, category prompts | | Share of Voice | Your mentions vs. all brand mentions in the category | Growing share vs. named competitors | | Rank / Position | Where you land within the answer | Named in the first recommendation or two | | Citation Frequency | How often the model cites a source tied to you | Consistent across multiple platforms | | Citation Share | Owned vs. third-party citations about you | A healthy mix, not owned-only | | Sentiment | Tone of how you're described | Net positive, no recurring errors | | Prompt Coverage | Query types (by intent) where you appear | Broad, spanning the buying journey | | Competitor Co-mention | How often, and in what role, competitors appear with you | Named instead of, not just alongside | | Source Attribution | Which domains the model cited | Top sources known by name |

Presence / Mention Rate

Presence is the simplest question: across your tracked prompts, does your brand show up at all? Mention rate is that same question as a percentage — mentions divided by prompts run.

We measure it by running a fixed prompt set against each platform on a schedule, logging a yes/no per prompt, rolled up by category. Good looks like mention rate climbing on unbranded, category-level prompts — "best project management tools for small teams" — not just prompts that already contain your name. The pitfall: mistaking branded-prompt presence for real progress, when you were always going to show up there.

Share of Voice (in AI answers)

Share of voice is the percentage of relevant AI responses that mention or recommend your brand versus named competitors, across your prompt set. Industry sources split this into finer flavors — share of answer, share of citation, share of mention — but the core idea holds: presence is relative, not a solo number (LLM Pulse).

It's measured by tallying mentions per brand across the same prompt set and dividing. Good looks like share growing even when your raw mention count is flat, since a competitor's mentions can fall while yours hold. The pitfall is tracking against a stale competitor set — category leadership shifts fast.

Rank / Position within an answer

Rank is where your brand lands within a single answer — named first, buried in a closing list, or absent from the recommendation entirely. There's no fixed "position 1 through 10" like a SERP; the model composes prose, so position has to be read structurally.

We measure it by parsing where your mention falls relative to the answer's structure and competitor mentions in the same response. Good looks like being named in the first recommendation or two, not the last item a user may never finish reading. The pitfall: treating every mention as equal, when named-first-and-praised beats buried in "some other options include."

Citation Frequency

Citation frequency is how often a model cites a specific source — a page, a domain — when answering a prompt relevant to your brand. It's distinct from a mention: a mention names your company in the text; a citation points to a page as evidence (Semrush).

We measure it by capturing the source list a platform surfaces alongside its answer and counting how often your domain, or domains favorable to you, appear. Good looks like consistent citation across platforms, not one lucky hit. The pitfall is chasing volume on pages nobody would cite for the query that matters — track per topic, not domain-wide.

Citation Share (owned vs. third-party)

Citation share splits citation frequency by ownership: your domain versus third-party sites that discuss you — reviews, comparisons, forums, press.

91%

of AI citations point to third-party sources rather than brand-owned domains

Source: Growth Unhinged / Onely

We measure it by tagging every citation as owned or third-party and tracking the ratio over time. Good looks like a deliberate mix — a program leaning entirely on owned content is missing where most citations originate. The pitfall follows directly from the stat above: optimizing only your own site ignores the larger half of what models trust.

Sentiment

Sentiment is the tone of how a model describes your brand when it mentions you — positive, neutral, or negative — aggregated across prompts and time, not just whether you were mentioned (LLM Pulse).

We measure it by classifying the language around each captured mention and rolling it up per platform and prompt category, so a drop traces back to a specific prompt. Good looks like net-positive sentiment with no recurring factual errors — outdated pricing, a discontinued feature — since models confidently repeat a mistake once it's baked into enough source material. The pitfall: averaging sentiment into one score and losing which claim actually needs correcting.

Answer Coverage / Prompt Coverage

Prompt coverage is the breadth of relevant query types — informational, comparison, "best of," problem-aware, buyer-stage — where your brand shows up, distinct from mention rate on any single prompt.

We measure it by building a prompt set that deliberately spans those intent categories, then tracking presence per category rather than one blended average. Good looks like coverage across the full buying journey, not just the "vs [competitor]" prompts that are easiest to win. The pitfall is a prompt set built entirely around queries you already rank well on — it flatters you and hides where you're actually invisible.

Competitor Co-mention

Competitor co-mention tracks how often your brand appears in the same answer as named competitors, and — critically — in what role: recommended alongside them, recommended instead of them, or mentioned only because the prompt named them first.

We measure it by logging every other brand in a response where you're also mentioned, then classifying whether you were positioned as peer, leader, or also-ran. Good looks like appearing instead of a weaker competitor, not just alongside a stronger one. The pitfall: treating any co-mention as a win — sharing a sentence with the category leader isn't share of voice if you're the caveat, not the pick.

Source / Domain Attribution

Source attribution identifies exactly which domains a model relied on to construct its answer about you — the specific, actionable version of citation frequency.

We measure it by aggregating cited domains across captured answers and ranking them, so you see the actual top sources shaping what models say about your category, not a vague "third-party" bucket. Good looks like knowing those sources by name and having a plan for each. The pitfall: stopping at "we got cited" without finding out where — you can't fix a bad source you haven't identified.

Turning definitions into a program

Definitions only matter if they become a habit: the same prompt set, on the same schedule, against the same platforms, read the same way every time. That's what turns "we think we're doing better in AI answers" into a number you can defend in a board meeting — and it's exactly what DabaRank automates.

We run your prompt set against the major AI platforms and growing — including dedicated ChatGPT, Perplexity, and Gemini rank trackers — and roll every metric above into one view. Agencies report all of it to clients under their own brand instead of building the tracking themselves. Still weighing GEO measurement against classic SEO and AEO tracking? See SEO vs. AEO vs. GEO. Then see where your brand stands — DabaRank pricing starts at $99/mo with a 14-day free trial.

AD

Written by

Alex Dabson

Founder, DabaRank

Alex has spent his career across marketing agencies, local-services businesses, and multi-location, multi-brand companies, with a background building SaaS products — the exact teams now working to measure AI visibility across many brands at once. He founded DabaRank to track how brands rank and get cited across ChatGPT, Claude, Gemini, Perplexity, and other AI platforms.