For twenty years, search measurement meant rankings, impressions, clicks, and conversions. Those numbers described a world where a buyer typed a query, scanned ten blue links, and clicked one. That world is shrinking. More of your prospects now ask ChatGPT, Claude, Gemini, Perplexity, or Google’s AI Overviews a full question and act on the single answer they get back. If you only measure the old metrics, you can be losing this new channel badly and never see it in a dashboard.
The fix is not more reports. It is the right reports. The AI search KPIs that matter answer four plain questions: when buyers ask AI about your category, does it name you? When it does, does it link to you? What does it say about you? And does any of it turn into business? This guide walks through each one, how to measure it without enterprise tooling, and the vanity metrics we tell clients to ignore. If you are still getting oriented to the discipline, our guide to answer engine optimization covers the fundamentals these KPIs sit on top of.
Why AI search needs its own KPIs
Traditional SEO measures a ranked list. AI search measures a recommendation. That difference changes what is worth counting. A keyword can rank number one and still never appear when someone asks an assistant “who is the best mortgage broker in Seattle for first-time buyers.” The model does not read your meta title and place you in a list; it synthesizes an answer from what it has learned and what it can retrieve, then hands the user a short, opinionated response.
Because the output is generated text, not a fixed page of links, two things break. First, most answers are zero-click, so click-based analytics miss the moment that mattered. Second, the same prompt can produce different answers on different days, so a single spot check tells you almost nothing. Measurement has to shift from “where do I rank for this keyword” to “how reliably am I the answer.” If you want the deeper mechanics of how the old and new disciplines diverge, our explainer on generative engine optimization is a good companion to this piece.
The four AI search KPIs that matter
Across the audits we run, almost every meaningful question about AI visibility collapses into four metrics. Track these and you have a real picture; track everything else and you have a cluttered one.
1. Presence rate
Presence rate is the percentage of your tracked prompts where the AI names your business at all, in any position. It is the foundation. If a buyer asks “who should I hire to do X near me” and your name never surfaces, nothing else matters. We treat presence as the first number a client should know and the first one we move, because going from zero to “mentioned” is the hardest and most valuable jump.
Measure it by holding a fixed prompt set and counting appearances. A presence rate of 10 percent means you show up in one of every ten buyer questions; 60 percent means you are a default option in your category.
2. Citation share
Citation share is the percentage of answers in which the AI links to your domain as a source. Engines like Perplexity and Google AI Overviews cite explicitly, so this is directly observable. Citation share is the closest AI-era equivalent to share of voice: it tells you whether the model trusts your content enough to send the user to you, not just mention you. Being named is good; being named and linked is better, because the link is where attributable traffic and authority compound.
3. Answer sentiment
Being mentioned is not automatically good. AI can name you and then describe you as “limited,” “pricier than competitors,” or “better for X than Y.” Sentiment tracks whether the framing is positive, neutral, or negative, and what specific claims drive it. This is where reputation data, reviews, and the language on your own site show up in the output. It is worth its own discipline; we go deeper in our piece on brand sentiment in AI, but at minimum log the adjectives and caveats the model attaches to your name.
4. AI referral leads and revenue
The last KPI is the one that pays for the rest: leads and revenue you can attribute to AI search. Some of this is visible as referral traffic from AI domains; much of it arrives as “I asked ChatGPT and it recommended you,” which only surfaces if you ask. Add a “How did you hear about us” field to your intake and train whoever answers the phone to ask. This is the metric that converts a visibility program into a business case.
How to measure AI search KPIs without enterprise tools
You do not need a platform to start. The most reliable method we use is also the simplest: a controlled prompt set, run repeatedly, logged consistently.
- Build a prompt set. Write 20 to 40 questions a real buyer would type, in their words, not yours. Mix category prompts (“best [service] in [city]”), comparison prompts (“[you] vs [competitor]”), and problem prompts (“how do I [solve the thing you fix]”).
- Run them across engines. Test the same prompts in ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews. Each engine pulls from different sources and will give you different results.
- Log four columns. For every prompt and engine, record: were you present (yes/no), in what position, were you cited (yes/no), and a one-line note on sentiment.
- Repeat on a cadence. Weekly or biweekly. One run is noise; a trendline is signal.
- Layer in analytics. In GA4, segment referral traffic from AI domains to capture the click-through slice, and watch your “How did you hear about us” field for the zero-click slice.
This manual approach is honest, cheap, and it is exactly how we baseline a new client before any optimization work begins. Dedicated visibility tools automate the logging once you outgrow a spreadsheet, but the metrics they report are the same four above.
A simple AI search KPI scorecard
Here is how the four metrics fit together into a scorecard you can keep in one tab. The targets are starting points, not promises; what counts is the direction over time.
| KPI | What it answers | How to measure | Healthy direction |
|---|---|---|---|
| Presence rate | Does AI name me? | % of prompt set where you appear | Climbing toward 50%+ |
| Citation share | Does AI link to me? | % of answers citing your domain | Rising vs. competitors |
| Answer sentiment | What does AI say? | Positive / neutral / negative tally | More positive, fewer caveats |
| AI referral leads | Does it make money? | Intake field + GA4 referral segment | Growing, attributable |
The vanity metrics to ignore
Plenty of numbers look like AI KPIs and tell you nothing. We steer clients away from these:
- Raw mention counts with no denominator. “AI mentioned us 14 times” is meaningless without knowing how many prompts you tested. Always report presence as a rate.
- A single screenshot. One favorable answer proves the model can say nice things once, not that it does so reliably. Trends beat snapshots.
- Total AI referral traffic in isolation. Clicks are a fraction of AI’s influence because so much is zero-click. Pair traffic with prompt testing or you will dramatically undercount the channel.
- Keyword rankings restated as AI metrics. Your Google position does not predict whether you are the recommended answer. They are related but not the same, and treating them as interchangeable hides the gap.
What good progress looks like
The reason to track these KPIs is that they move when you do the work, and the movement is legible. The clearest public example is Keith Akada, a Seattle mortgage broker who went from invisible in AI answers to the most-recommended broker in his market, generating roughly 30 leads and four closed deals in about six weeks. On a scorecard, that story reads as presence rate climbing from near zero, citation share appearing where there was none, sentiment turning consistently positive, and referral leads landing in the intake form. The KPIs did not cause the result, but they made it visible and repeatable, which is the entire point of measuring.
If you want to see how this connects to the specific engine buyers use most, our walkthrough on showing up in ChatGPT pairs naturally with the presence-rate work above.
Where to start
You do not have to instrument everything at once. Pick the 20 prompts your best customers would actually ask, run them across the major engines this week, and write down whether you appear. That single baseline will tell you more about your real position in AI search than any keyword report you have ever pulled. From there, the four KPIs give you a steady way to watch the line move in the right direction, and a clear story to tell when it does.