Classic SEO dashboards were built for a world of blue links and ranking positions. AI assistants do not return ten links. They return one answer, and your brand is either inside it or it is not. Here is the metric set that reflects that reality.
Daniel Arons · Jun 2026 · 7 min read
Most marketing teams already have a reporting rhythm for organic search. They watch keyword rankings, sessions, click-through rate, and the slow climb of domain authority. Those numbers describe a channel where the goal is to earn a position on a page and win the click. They are useful, and they are not going away. But they describe only part of how buyers now find and evaluate options.
When a prospect asks ChatGPT, Claude, Gemini, or Perplexity for the best tools in a category, the assistant synthesizes an answer. It names a short list, describes each option in its own words, and sometimes cites sources. There is no position one. There is no impressions column. If your reporting only measures rankings and clicks, you are blind to the moment where a recommendation is actually formed. This post lays out the KPIs that make that moment visible, and why the old scorecard falls short on its own.
Why classic SEO KPIs fall short
Ranking position assumes a ranked list. AI answers do not rank ten results for the user to choose from. They make a selection on the user's behalf and present it as guidance. A page that sits at position four for a query may never surface in the assistant's answer, while a page that ranks lower may be paraphrased into the recommendation. The relationship between rank and inclusion is loose, so rank alone cannot tell you whether you are being recommended.
Click-through rate has the same problem. In an answer-first interface, many buyers never click anything. They read the synthesized response, form an impression, and move on to a shortlist. The influence happened, but no click recorded it. Impressions and sessions undercount this entirely. So the first shift is conceptual: you are no longer only measuring traffic you earned, you are measuring how often and how well you are represented inside answers the buyer never leaves.
There is also a structural difference. Search rankings are relatively stable from one user to the next. AI answers vary by phrasing, by engine, by session, and by the buyer's stated context. Measuring presence in AI answers means sampling across many prompts and engines, not checking a single position. That is why a different metric set is needed, not a tweak to the existing one. For a fuller treatment of the underlying discipline, see how to measure AI visibility.
“Rank tells you where you sit on a page. It does not tell you whether the assistant put you in the answer.”
The core KPI: presence or inclusion rate
Presence rate is the foundation. It answers a simple question: across a representative set of buyer prompts, how often does your brand appear in the answer at all? You define a panel of prompts that mirror how real prospects ask, run them across the engines that matter to you, and record whether your brand is mentioned. The output is a percentage. If you appear in 30 of 100 relevant prompts, your presence rate is 30 percent.
Presence is binary at the prompt level and a rate in aggregate, which makes it easy to communicate to leadership and easy to track over time. It is also the metric most directly tied to whether you exist in the consideration set. A brand with a low presence rate is invisible in the channel no matter how strong its website is. Before optimizing anything else, teams should know this number and watch it move. One caution: do not report presence as a single global figure. The same brand can sit near the top in one category of prompts and vanish in another, so a blended average can hide exactly the gaps you most need to close.
Defining a prompt panel that holds up
A presence rate is only as credible as the prompts behind it. Build the panel from the language buyers actually use: category questions ("best tools for X"), comparison questions ("X versus Y"), problem-framed questions ("how do I solve Z"), and qualifier questions ("affordable X for small teams"). Keep the panel stable so movement reflects real change rather than a shuffled question list. When you add prompts, version the panel so trend lines stay honest.
Share of recommendation against competitors
Presence tells you whether you show up. Share of recommendation tells you how you stack up. For each prompt where any vendor is named, record which brands appear and how prominently. Aggregate that into a share figure: of all the recommendations surfaced across your panel, what proportion are yours versus each named competitor? This is the AI-era equivalent of share of voice, and it is far more decision-useful than your own presence rate in isolation.
Share of recommendation reframes the goal from "appear sometimes" to "win the consideration set." It surfaces the competitors who consistently get named ahead of you, the prompts where a rival dominates, and the categories where you are absent while others are entrenched. Reported next to presence rate, it gives leadership a competitive read rather than a vanity count. It also makes prioritization obvious: chase the high-intent prompts where a competitor owns the answer and you do not. The competitive frame matters because answers are zero-sum in a way pages are not. A search results page can hold ten listings; an assistant's shortlist usually names a handful, so a gain for a rival is often a loss for you on that exact prompt.
Accuracy and sentiment of how AI describes you
Being named is necessary but not sufficient. The assistant also describes you, and that description shapes the buyer's impression before they ever reach your site. Two qualitative KPIs capture this. Accuracy measures whether the assistant's description of your product, pricing model, positioning, and capabilities matches reality. Sentiment measures whether the framing is favorable, neutral, or skeptical. A confident, accurate, positive description does very different work than a hedged or outdated one.
These are harder to quantify than presence, but they are trackable. You can score descriptions on a simple scale, tag specific inaccuracies (a wrong integration, a stale price tier, a misattributed weakness), and watch whether corrections to your public content move the framing over time. Accuracy issues are often the highest-leverage thing to fix, because a single repeated error can quietly suppress conversion across every answer the buyer reads.
“If the assistant names you but describes you wrong, you are losing deals inside a sentence you never wrote.”
Citation share and coverage by engine and segment
Some engines, Perplexity most visibly, attach citations to their answers. Citation share measures how often your owned content is the source the assistant draws from. This matters for two reasons. It is a more durable signal than a passing mention, and it gives you a content lever: when your pages are the cited authority, you have direct influence over how the answer reads. Track which of your URLs get cited, for which prompts, and on which engines.
Coverage adds two dimensions that a single blended number hides. Per-engine coverage acknowledges that ChatGPT, Claude, Gemini, and Perplexity do not behave identically; you may be strong on one and absent on another, and the mix of engines your buyers use should weight your priorities. Per-segment coverage breaks results out by buyer context, since an answer for "enterprise" can differ sharply from one for "small team." Reporting these separately keeps a strong average from masking a segment where you are losing. For day-to-day mechanics, see how to track your brand's visibility in AI search.
Movement over time and tying KPIs to the funnel
Every metric above is most valuable as a trend, not a snapshot. A presence rate of 30 percent means little until you know whether it was 18 percent last quarter or 42 percent. Hold your prompt panel and engine mix stable, sample on a consistent cadence, and report movement. Trend lines let you connect specific actions, a new comparison page, a corrected pricing description, a fresh proof point, to changes in presence, share, sentiment, and citation share.
These are mid-funnel KPIs, and it helps to name them that way. They do not replace pipeline and revenue, and they should not be reported as if they did. They measure the consideration stage, the moment where a buyer is forming a shortlist and an assistant is shaping it. Treat them as leading indicators that sit upstream of demo requests and opportunities, and pair them with your existing downstream numbers. The ROI of AI visibility becomes legible when you can show movement in mid-funnel presence ahead of movement in pipeline.
This is the layer Aethon AI is built to map. Rather than treating AI visibility as a single score, contextual AI presence mapping samples across prompts, engines, and buyer segments to show where you are recommended, how you are described, and how that picture changes. It sits on top of your SEO work, not against it, and it gives the consideration stage a scorecard it has been missing. If you want to see the metric set applied to your own category, take a look at how Aethon works or request a demo and we will map your current presence against your competitors.
Frequently asked questions
What is the single most important KPI for AI search?
Presence or inclusion rate is the foundation: across a representative set of buyer prompts, how often does your brand appear in the answer at all? It tells you whether you exist in the consideration set before you optimize anything else.
How is share of recommendation different from presence rate?
Presence rate measures how often you appear. Share of recommendation measures how you stack up against named competitors across the same prompts. It is the AI-era equivalent of share of voice and is more decision-useful than your own presence rate in isolation.
Why do classic SEO KPIs fall short for GEO?
Ranking position assumes a ranked list of links, but AI assistants synthesize one answer and make the selection for the user. Click-through rate misses buyers who never click. Neither metric captures whether or how you are represented inside the answer itself.
Can you measure how accurately AI describes your brand?
Yes. You can score descriptions for accuracy and sentiment, tag specific errors like a wrong integration or a stale price tier, and track whether updates to your public content shift the framing over time. Accuracy fixes are often the highest-leverage work.
Are AI visibility KPIs top-funnel or mid-funnel?
They are mid-funnel. They measure the consideration stage where a buyer is forming a shortlist and an assistant is shaping it. Treat them as leading indicators upstream of demo requests and pipeline, reported alongside your existing downstream numbers rather than in place of them.
Written by
Daniel Arons
Co-founder & CEO, Aethon AI
Daniel co-founded Aethon AI in November 2025 to close the gap between how marketers measure AI visibility and what AI is actually doing with their brands. Before Aethon, he spent eight years building digital marketing programs in New York across SaaS, financial services, and consumer brands. He holds an MPA from Baruch College and a BA in Public Relations from SUNY Oswego.