Most AI visibility dashboards bury you in numbers and call it insight. Four metrics carry real signal. The rest is noise dressed up as a chart. Here is what each one measures, what it cannot tell you, and how long you need to watch it before the trend means anything.
Four numbers, not forty
The ai visibility metrics worth tracking come down to four: Share of Model, Ad Exposure Score, Citation Health Index, and your cited pages. Everything else on a typical dashboard is a derivative of these or a vanity figure that moves without changing your outcome. If a number does not tell you where you stand in an AI answer or what to do next, it is decoration.
AI search behaves differently from the blue-link era. A user asks ChatGPT or Perplexity a question and gets one synthesized answer. Either your brand is named in that answer or it is not. There is no page two. That binary outcome, repeated across hundreds of prompts and several engines, is what these four metrics turn into something you can read and act on.
The trap is treating AI visibility like a leaderboard you check once. Citation data is volatile. The same prompt can name three different brands across three runs as engines re-weight their sources and refresh their indexes. A single snapshot is a coin flip. A 60 to 90 day trend is a signal. Keep that distinction in mind for every metric below.
“If a number does not tell you where you stand in an answer or what to do next, it is decoration.”
The filter for every metric
How often the answer names you
Share of Model measures how often your brand appears in AI answers across a defined set of prompts, expressed against the brands that show up instead. Run 200 high-intent prompts, count how many name you, and you have your share. It is the closest equivalent to share of voice, rebuilt for a world where the answer is generated rather than ranked.
Read it relative, not absolute. A 20 percent Share of Model means little until you see that two competitors sit at 45 and 30. The gap is the story. Share of Model also depends entirely on which prompts you measure, so the prompt set has to reflect real audience questions, not keywords you wish you owned. Measure the wrong prompts and the number is confident and useless.
Watch it per engine. ChatGPT, Gemini, Perplexity, Claude, and Copilot pull from different sources and weight them differently, so your share can be strong in one and absent in another. A blended average hides that. Break it out, find the engine where you are losing, and you know where to spend effort first.
“Share of Model is meaningless in isolation. The gap to the brands ranked above you is the whole point.”
Reading the number right
Ad Exposure and Citation Health
Ad Exposure Score tracks how visible your category is to paid placement inside AI answers, as engines like ChatGPT roll out ad formats. It estimates where ads can intercept the prompts your buyers ask and where a competitor could buy the moment you currently win organically. It does not detect any specific competitor’s ad spend, and any tool that claims to is guessing. Read it as opportunity and exposure, not as surveillance.
Citation Health Index looks at the quality and stability of the sources AI engines pull when they answer about you. Healthy citations come from pages you control or trust, stay consistent across runs, and reflect current information. An unhealthy profile leans on outdated pages, a forum thread, or a competitor’s comparison post. The index turns that mess into one number you can move by fixing or publishing the right sources.
Both scores are directional, not precise to the decimal. Their value is the trend line. A Citation Health Index climbing over two months after you shipped fresh, schema-clean pages tells you the work landed. A single reading tells you almost nothing. Treat any week-to-week swing as noise until the longer pattern confirms it.
“Ad Exposure Score reads opportunity and risk. It never claims to detect what a competitor spends.”
The honesty line
The pages doing the work
Your cited pages are the specific URLs AI engines reference when they mention you. This is the most actionable metric of the four because it points at exact assets. If three pages earn most of your citations, you know what to protect and reinforce. If your strongest commercial page never gets cited, you have found the gap to close.
Cited pages also expose the uncomfortable cases: the engine that cites a three-year-old blog post instead of your current product page, or names a competitor’s roundup as the source for a question about you. Each cited URL is a lead. Update the stale page, add structure an engine can parse, or publish the page that should exist and does not yet.
Tie cited pages back to the other three metrics and the picture closes. Share of Model tells you how often you appear, Citation Health tells you whether the sources behind that are sound, Ad Exposure tells you where paid placement can shift the result, and cited pages tell you exactly which assets to change. Verify the trend across 60 to 90 days, then act on the page level where the data is concrete.
“Every cited URL is a lead. The stale ones are a to-do list, and the missing ones are a brief.”
Turning data into action
Numbers only matter once you can see the answers behind them. A scan shows you your Share of Model, your cited pages, and the engines where you are missing, with the real AI responses attached.