Measurement
The AI visibility metrics that actually matter
By Arnav Mukherjee, founder of TofuBofu · July 6, 2026
Once you start measuring AI visibility, the next danger is measuring the wrong thing. It is very easy to build a dashboard that shows a big, satisfying number, "you were mentioned 240 times this month," and to feel like you are winning. Then you look at your pipeline and nothing has moved. The number went up and the business did not. That gap is almost always because the metric was a vanity metric.
AEO has a lot of these. The category is full of tools that show you plenty of data and leave you to guess what any of it means. So before you track anything, it is worth being clear about which metrics tell you whether AI is actually sending buyers your way, and which just look impressive. Here are the four that matter, and the ones to ignore.
The vanity metric to stop staring at
Raw citation count. The total number of times you were mentioned across all queries and engines. It is the AEO equivalent of impressions or follower count: big, easy to grow, and disconnected from outcomes. A count of 240 tells you nothing about whether you were the recommendation or a footnote, whether you were described accurately or wrongly, or how you stacked up against the three competitors named alongside you. It only becomes useful when you turn it into a ratio, which is where the real metrics begin.
The four metrics that matter
Vanity vs value
1. Presence rate. Of your buyer questions, what share name you at all? If you built a set of 30 questions and you appear in 6, your presence rate is 20 percent. This is the honest baseline: it answers "does AI know I exist for the things buyers ask," and it is the first number that should move when your AEO work starts landing.
2. Share of voice. Presence rate in isolation can flatter you. Share of voice fixes that by measuring you against the competitors answering the same questions. AI recommendation is comparative, the buyer gets a shortlist, so what matters is your slice of that shortlist. This is the metric with real benchmarks: a leader holds 40 to 70 percent, a challenger 20 to 35, a new entrant 2 to 10. Knowing your band, and climbing it, is the game.
3. Sentiment. Being named is not the same as being named well. AI can describe you with stale facts, the wrong specialty, or flat, unconvincing language. For a newer brand the usual issue is not negativity but vagueness, because the engine barely knows you. Tracking how you are described tells you whether each mention is helping or quietly costing you a deal.
4. Recommendation versus citation. These are different states with different value. Being cited means your content sits in the sources, sometimes a footnote that never shapes what the buyer reads. Being recommended means the engine actively puts you forward. Some engines list many sources with low individual influence, so you can be "cited" without affecting the answer at all. Measure recommendation separately, because it is the state that actually drives a decision.
The metric that ends the argument: revenue
The four above are the ones you can move month to month. But the metric that settles whether any of this was worth it is the chain from citations to clicks to leads to revenue. Early on it is hard to measure, so presence rate and share of voice are your leading indicators. As AI referral traffic grows, connecting it through your analytics to real pipeline is what proves the return. It is the hardest loop to close and the most valuable, and it is why measurement is not a vanity exercise: done right, it ties an AI mention all the way to a deal.
What to do
1. Demote raw count to an input
Stop treating total mentions as a headline. Keep it only as the raw material you turn into presence rate and share of voice.
2. Track presence rate and share of voice monthly
The share of buyer questions you appear in, and your slice against competitors. These are the two numbers that should move when your work lands.
3. Read the sentiment, not just the mention
Check how AI describes you, and flag anything stale, vague, or wrong. A bad description is a fixable problem you can only fix if you see it.
4. Separate recommended from cited
Note when you are actively recommended versus merely listed as a source. Recommendation is the state that moves buyers.
5. Build toward the revenue loop
Wire AI referral traffic into your analytics so that, over time, you can connect citations to clicks to leads. That is the metric that proves it all.
See your real metrics, not a vanity number
Run a free scan for your presence rate, share of voice, and how AI describes you across six AI engines.
Get your free auditFrequently asked questions
What AI visibility metrics actually matter?
Four. Presence rate: the share of your buyer questions where you are named at all. Share of voice: your mentions against the competitors answering the same questions. Sentiment: whether the way AI describes you is accurate and positive. And recommendation versus citation: whether the engine actively suggests you or merely lists you in a footnote. Together these tell you if AI is sending buyers your way. Raw citation count on its own does not.
Is citation count a good AI visibility metric?
On its own, no. A big number of raw mentions feels good but hides everything that matters: whether you were recommended or buried, described accurately or wrongly, and how you compare to competitors on the same questions. A brand can rack up citations in footnotes while never being the recommendation a buyer acts on. Use citation count only as an input to share of voice, not as a headline metric.
What is share of voice in AI search?
Share of voice is your presence relative to competitors across a defined set of buyer questions. If you are named in 6 of 30 answers and the whole competitive set is named 30 times, your share of voice is your slice of that total. It matters more than raw count because AI recommendation is comparative: buyers get a shortlist, and what counts is your share of that shortlist, not your absolute number of appearances.
What is a good AI share of voice?
It depends on your position in the category. Published ranges put a category leader at roughly 40 to 70 percent share of answer, a top-three challenger at 20 to 35 percent, a top-ten player at 10 to 20 percent, and a new entrant at 2 to 10 percent. The useful move is to identify which band you are in and aim to climb one band, rather than chasing an absolute number that ignores how crowded your category is.
Why does sentiment matter for AI visibility?
Because being named is not the same as being described well. AI can mention you with outdated information, the wrong specialty, or lukewarm framing, all of which cost you deals even though you technically appeared. For newer brands the common problem is not bad sentiment but thin, vague description, because the engine barely knows you. Tracking sentiment tells you whether the mention is helping or quietly hurting.
What is the difference between being cited and being recommended?
Being cited means your content appears as a source, often a footnote, without necessarily shaping the answer. Being recommended means the engine actively puts you forward as an option the buyer should consider. Some engines cite many sources with low individual influence, so you can sit in the references without affecting what the buyer reads. Recommendation is the state that drives decisions, so it is worth measuring separately from raw citation.
Should I track AI visibility down to leads and revenue?
Eventually, yes, and it is the metric that ends the debate about whether AEO is worth it. The chain runs from citations to referral clicks to leads to revenue. Early on, presence rate and share of voice are the leading indicators you can move quickly. As AI referral traffic grows, connecting it through analytics to actual pipeline is what proves the return, and it is the hardest and most valuable metric to close the loop on.
Sources and further reading
- Digital Applied: AI share of voice framework: the share-of-voice benchmark bands and the cited-versus-used distinction.
- Ahrefs: AI Overview brand visibility factors: context on what actually drives the mentions you are measuring.