You opened an AI visibility dashboard, and eleven numbers stared back at you. None of them told you what to do on Monday.
That's normal. The category is young, every vendor names things differently, and nobody handed you a rubric.
Here's the good news: there are only five ai visibility metrics that do real work. Everything else on that dashboard is a remix of those five. Once you can name them, you can ignore most of the screen.
This is a catalog, not a checklist. We'll define each metric, say plainly what it tells you and where it lies to you, and then help you pick the two or three that match your goal. You don't need all five. You need the right two.
Let's take them one at a time.
What Counts as an AI Visibility Metric?
An AI visibility metric measures how your brand shows up inside an AI-generated answer. Not where you rank on a page of links. Inside the answer itself.
That distinction is the whole ballgame, and it's why your old reporting doesn't transfer.
Think of the measurement space in three layers:
- Presence. Are we there at all? (mention rate, citation rate, share of voice)
- Representation. How are we portrayed when we're there? (prominence, sentiment)
- Outcomes. Did any of it pay? (referred sessions, pipeline, revenue)
The five metrics in this piece live in the first two layers. That's on purpose. Outcome measurement is a different job with different plumbing, and you can't do it credibly until the first two layers are stable.
Why does this matter now rather than next year? Because the surface got big, fast. Google AI Overviews appeared on roughly 6.5 percent of keywords in January 2025, climbed to nearly 25 percent by that July, then settled into a 15 to 16 percent band, according to Semrush's tracking. And the click behavior underneath changed with it. Pew Research found that when an AI summary was present, users clicked a traditional result about 8 percent of the time, against 15 percent when no summary appeared. Ahrefs, looking at a 300,000-keyword sample, measured a 58 percent drop in organic click-through for position one on AI Overview queries.
Read those together and the conclusion is uncomfortable but clear. Being in the answer matters more than it used to, and the answer sends you less traffic than you'd hope. That gap is exactly what these metrics exist to measure.
Your next step: before you touch a tool, write down the one question you want AI answers to settle for you. Awareness? Authority? Competitive position? Reputation? That answer picks your metrics for you, and the rest of this piece is easier with it in hand.
The Five AEO KPIs, Defined
These are the aeo kpis worth knowing by name. Narrow the ai search kpis to track down to these five and you can read almost any dashboard in the category. Each one gets a definition, the question it answers, and the way it misleads you if you trust it alone.
Mention Rate
What it is: the percentage of relevant AI answers that name your brand at all, link or no link.
The math is simple. Divide the number of answers that name you by the total relevant prompts you ran, then multiply by 100.
What it tells you: awareness. When buyers ask AI the questions you care about, does your name come up?
Why it comes first: if you're not mentioned, you can't be cited. It's the floor. A high mention rate with a low citation rate is the known-but-not-linked pattern, which is a real and fixable position.
Where it misleads: a mention isn't a click. In one panel study of AI search results, roughly 58 percent of observed brand mentions carried no clickable link at all, so mention rate consistently overstates how actionable your presence is. It also can't tell a "brands to consider" list from a "brands to avoid" list. Both count as one mention. And vanity prompts like "what is CRM" will inflate the number without measuring anything you can act on.
Citation Rate
What it is: the percentage of relevant AI answers that link to your domain as a source. Inline, in a sources panel, or in a footnote. All count.
Same math. Answers that link to you, divided by relevant prompts run, times 100.
What it tells you: trust. When AI discusses your category, is your content considered authoritative enough to reference?
Why it's the one most content teams should watch: it maps straight to a decision you can actually make. Which page do I write or improve so AI cites me for this prompt? No other metric in the set hands you a task that cleanly.
Where it misleads: citation behavior is fragmented across engines, and badly. Perplexity typically carries several sources per answer by default. ChatGPT may carry none. Pew found Google AI Overviews included three or more sources in 88 percent of cases. Worse, engines disagree about who deserves the link: on overlapping prompts, only about 11 percent of cited domains show up in both ChatGPT and Perplexity answers. A single blended "citation rate" across engines is an average of things that barely correlate. Citation rate is also blind to position. Being source five looks identical to being source one.
Share of Voice
What it is: your slice of all the citations (or mentions, depending on the vendor) earned across a defined prompt set, measured against a defined competitor set. The standard formulation puts citations in the numerator.
Many vendors weight it by position, giving something like 40 percent of the credit to the first brand named, 30 to the second, 20 to the third, and 10 to everyone after.
What it tells you: competitive position. Of the brands AI treats as players in your space, how much of that conversation is yours?
Why leaders like it: it collapses to one number per prompt set, which survives the trip to a board deck. It also forces you to name a comparison set, which makes it much harder to game with soft prompts.
Where it misleads: this is the least standardized metric of the five, and it isn't close. Vendors disagree on whether the numerator counts mentions or citations, whether to weight by position and how, whether the competitor list stays fixed or floats between runs, and how to handle a brand named three times in one answer. Two tools measuring your brand in the same week can report share of voice figures 10 to 20 points apart. Neither is lying. They're answering different questions.
If you're going to report this number upward, you owe it three things: a stable prompt set (20 to 50 buyer-intent prompts is a reasonable start, more at enterprise scale), a fixed competitor list, and a consistent cadence. Change any of them and your trend line becomes fiction.
Prominence
What it is: where you land in an AI's list, not just whether you're on it. Usually expressed as an average position, a top-three rate, or a weighted score that pays more for first place.
What it tells you: attention. When AI lists options for your buyer's question, are you near the top, where attention actually lives?
Why it earns its slot: position bias in language models is real and heavily replicated. In one controlled study out of MIT's computational law group, ChatGPT picked the first candidate in an unordered list about 87 percent of the time unless it was told to do otherwise. Benchmark work in AEO puts first position at roughly 32 percent higher purchase intent than later slots. The advantage compounds: models over-favor the top of the list, and so do humans reading the output.
Where it misleads: prominence is the twitchiest number you'll track. Testing suggests the odds that two runs of the same prompt return the same list in the same order sit somewhere under a quarter of one percent. One run tells you close to nothing. Run each prompt ten or more times and use the modal position, the one that shows up most often. And remember that being listed first while described inaccurately is a weak win, because the description holds more attention than the ranking does.
Sentiment
What it is: the framing of each mention, classified as positive, neutral, or negative, sometimes with a score attached. Most platforms run a classifier or an LLM over each mention to label it.
What it tells you: narrative. When you show up, is the framing pulling a buyer toward you or away?
Why it's the only quality metric here: the other four count things. This one judges them. You can post a rising mention rate and a rising citation rate while quietly losing, because the framing drifted from "leading option for X" to "the cheaper choice if you can't afford Y."
Where it misleads: sentiment produces more false comfort than any other metric in the set. The sharpest critique is that scoring sentiment on solution-seeking prompts conflates recommendation with favorability. The AI recommends you because you fit the question, and the tool reads that as praise. There's a circularity problem too: the model scoring the mention is often the same model that wrote it, so scores drift toward the model's own priors. The cleanest read comes from multi-brand comparison prompts, where you can see how the framing of your brand differs from the framing of a rival in the same breath.
Citation Rate vs Share of Voice: Which One Answers Your Question?
Short answer: citation rate tells you whether you're earning trust. Share of voice tells you whether you're earning it faster than the competition. One is absolute, the other is relative, and confusing them is the most common reporting mistake in AEO.
The citation rate vs share of voice question comes up constantly, so let's make it concrete.
Your citation rate can climb while your share of voice falls. That happens when your category is getting more visible overall and rivals are climbing faster. You'd feel good reading one number and alarmed reading the other, and the alarm would be the correct response.
The reverse happens too. Share of voice can rise because a competitor stopped publishing, not because you did anything. Absolute performance flat, relative performance up.
So which do you use?
- Use citation rate when you own a content roadmap and need to know what to produce next. It points at pages.
- Use share of voice when the question is competitive and the audience is your leadership. It points at position.
- Use both when you're trying to overtake a specific named rival. Citation rate is your leading indicator, share of voice is the scoreboard.
Your next step: pick which of those two is your primary right now, and demote the other to context. Not forever. Just for this quarter.
Which AI Visibility Metrics Matter for Your Goal
Pick two or three. Stop there.
That's the discipline, and it's the part most teams skip. Every metric needs its own prompt set, its own configuration, and its own review rhythm. Track all five from day one and you'll under-collect on the one that actually matters, while maintaining a dashboard nobody opens.
Which ai visibility metrics matter for you depends entirely on the question you wrote down earlier. Here's the mapping:
- Goal: category awareness. Mention rate as primary, share of voice for context. Are we a known option at all? Most programs should live here for the first six months.
- Goal: authority and citation capture. Citation rate as primary, share of voice for context, prominence as a diagnostic. This is the stack for content leads whose roadmap is producing the cited pages.
- Goal: reputation and narrative control. Sentiment as primary, mention rate for breadth, citation rate to catch shifts in who's being credited. Right when the fear isn't being unseen, it's being mis-described.
- Goal: outpacing a named competitor. Share of voice as primary, citation rate as leading indicator, prominence as tiebreaker.
- Goal: defending against traffic loss. Citation rate as primary, paired with sentiment for narrative and share of voice for defensibility.
There's a second filter worth applying, and it's the buyer journey:
- Awareness prompts. Mention rate dominates.
- Consideration prompts. Citation rate, share of voice, and prominence dominate. These are the "compare these five tools" questions.
- Decision prompts. Sentiment and accuracy dominate. This is where AI closes its recommendation.
Notice that no goal in that list needs all five. Two or three, every time.
Why One Snapshot Will Lie to You
Whatever you track, run it more than once. This is the least glamorous advice here and the most load-bearing.
AI answers are unstable in ways that will embarrass you in a meeting. Only about 30 percent of brands stay visible from one answer to the very next one, and roughly 20 percent hold visibility across five consecutive runs. Visibility can slide by about a third inside five weeks. The same brand can vary by nearly 3x across geographies, so a global average quietly hides the market you actually sell into.
Two more forces worth knowing about, because they'll show up in your data as noise you'd otherwise misread:
Freshness bias. Around 95 percent of ChatGPT citations point to content published or refreshed in roughly the last ten months. Your citation rate is partly a measure of how recently you touched the page.
Rank is a weak predictor. Roughly 80 percent of LLM citations point to pages that don't appear in Google's top 100 organic results. Your best-ranking page is not automatically your most-cited one. Benchmark data does show position still helps at the AI layer, with top-ranked pages cited meaningfully more often than page-one stragglers, but the effect is far from a guarantee.
Your next step: fix your prompt set, run each prompt at least ten times, and check weekly rather than monthly. A monthly snapshot of an unstable system isn't a trend. It's a coin flip you're reading as a strategy.
How to Read an "AI Visibility Score"
Most dashboards will hand you one composite number. Treat it as a trend line for your own brand, never as a benchmark against a competitor measured by a different tool.
Composite scores are just the five metrics in a blender, with weights the vendor chose. One published methodology weights it roughly 35 percent mention rate, 20 percent position, 15 percent citation rate, 10 percent sentiment, 10 percent answer share, and 10 percent engine breadth. That's a defensible set of choices. It's also completely arbitrary relative to your goals. A brand that cares about citations is being scored mostly on mentions.
Every visibility score on the market is proprietary. Run the same brand in the same week through three tools and you can get three numbers that disagree by double digits, because they're weighting different things over different prompt sets.
So ask for the methodology spec before you believe any of it. What's in the numerator? Which engines? How many runs per prompt? If a vendor can't answer that in a sentence, the number isn't measurement. It's decoration.
This is also where the tool you choose starts to matter, because it decides what you can act on. DeepSmith's AEO module reports mention rate, citation rate, and share of voice with trends over time, alongside a competitor leaderboard and the specific pages AI cites, and engine coverage scales with the plan (ChatGPT on Pro, Perplexity added on Grow, Gemini on Scale, all five engines at Enterprise). The reason we built it next to content production is that a citation gap is a content problem, and the trip from "we're not cited for this prompt" to "here's the page that fixes it" is where most programs stall.
Whatever you use, the standard is the same: know the methodology, or don't quote the number.
Start With Two
You don't need a bigger dashboard. You need a smaller one.
Pick the metric that matches your goal. Add one that gives you competitive context. If you have the appetite, add a third that explains why the other two look the way they do. That's it. Mention rate, citation rate, share of voice, prominence, and sentiment are the catalog you choose from, not a list you must complete.
Then hold your prompt set still long enough to learn something from it. Most teams change what they measure right when the data was about to get useful.
You're closer than the dashboard makes you feel. Two metrics, one stable prompt set, one weekly check. Start there this week.
If you'd like to see what your mention rate, citation rate, and share of voice actually look like today, start a free DeepSmith trial and watch real prompts run against your brand before you decide anything.
