For twenty years, measuring search performance was a solved problem: rankings, organic clicks, and conversions told you almost everything you needed to know. That scorecard is now half-blind. When an AI answer — Google's AI Overview, ChatGPT, Perplexity, Gemini — resolves a query inside its own interface, there is no ranking to hold and no click to count, yet the reader has still formed an impression of your brand and often made a decision. Pew's July 2025 study put a number on the gap: users clicked a traditional result in 8% of searches that showed an AI summary, versus 15% without one, and just 1% of the time on a link inside the summary itself. The instinct is to reach for a brand-new AI-only dashboard and treat the old SEO KPIs as obsolete. That is the wrong move in both directions — some legacy KPIs are more important than ever, some are now actively misleading, and the new AI-search KPIs only make sense read alongside them. This guide builds the unified scorecard a marketer actually needs: the SEO KPIs that still matter (and what they now mean), the ones that quietly lie to you, the AI-search KPIs to add — share of voice, citation rate, prompt coverage, recommendation rank, AI referral traffic, and answer sentiment — how they are defined and gathered, and how to read the whole set together instead of chasing any single number in a world where no single number is trustworthy on its own.
For twenty years, measuring search performance was close to a solved problem. Rankings told you where you stood, organic clicks told you what that position was worth, and conversions closed the loop. In 2026 that scorecard is half-blind. When an AI answer resolves a query inside its own interface — Google's AI Overview, ChatGPT, Perplexity, Gemini's answers — there is no ranking to hold and often no click to count, yet the reader has still seen your brand named (or not), formed an impression, and sometimes made a decision. Pew's July 2025 study put the gap in numbers: users clicked a traditional result in 8% of searches that showed an AI summary versus 15% without one, and in just 1% of searches on a link inside the summary itself. The measurement that used to be reliable now misses a growing share of what actually happened.
The tempting response is to declare the old KPIs dead and stand up a brand-new AI-only dashboard. That is wrong in both directions. Some legacy SEO KPIs matter more than ever — they are among the only ways to see AI-driven awareness that never produced a click. Others now actively mislead, because they measure a click that is disappearing for reasons unrelated to your performance. And the new AI-search KPIs — share of voice, citation rate, prompt coverage, recommendation rank, referral traffic, answer sentiment — only make sense read alongside the legacy set, not instead of it. This guide builds the one scorecard that spans both worlds: what to keep, what to stop trusting, what to add, and how to read the whole thing together.
Start with the mechanism, because it explains every KPI decision that follows. The traditional search funnel assumed a click: the searcher saw a list of results, clicked one, landed on a page, and that page did the persuading. Every legacy KPI is built on that assumption. Rankings measure your position in the click list. Organic click volume measures how many people took the click. Time-on-page and bounce measure what happened after. When an AI answer sits above the results and satisfies the query in place, the click becomes optional — and for a large share of informational queries, it simply does not happen. The reader still learns whether you exist, what you do, and whether you are recommended, but none of it registers in a click-based KPI.
This is why reading organic clicks in isolation now produces a false signal. A page can be steadily losing clicks while gaining influence — being cited more often in AI answers, mentioned to more buyers, driving more branded searches later — and the click chart will read as pure decline. The KPI is not wrong about clicks; it is wrong as a proxy for performance, because the thing it was standing in for (did people encounter and consider us?) has partly moved to a surface it cannot see. The fix is not to throw the metric out. It is to demote it from headline KPI to one input among several, and add the metrics that see the part it now misses. The broader case for why click-based measurement broke and what replaces it is worked through in AI content didn't stop working — your metrics did.
Four legacy KPIs survive the transition, and two of them gain importance. Branded search volume and direct traffic are the first pair, and they are the quiet heroes of the AI era. When an AI answer names your brand without linking — the most common outcome — the reader who is interested does the one thing the answer box cannot stop them from doing: they search your name, or type your URL directly. A rising branded-search trend that is not explained by a campaign or PR spike is one of the cleanest available proxies for AI-driven awareness that produced no click. Watch the ratio of branded to non-branded search over time; a shift toward branded is often the fingerprint of answer-engine mentions doing their work upstream of your analytics.
The second pair is impressions and rankings, both re-read for the new surface. Impressions still matter because they show you are being surfaced even as clicks fall — and Search Console now reports AI Overviews and AI Mode impressions specifically, so you can see when Google is putting you into an AI answer rather than only a blue link (how to read that data is covered in AI search impressions in Google and AI search performance reporting in Search Console). Rankings for informational pages matter for a subtler reason: AI Overviews overwhelmingly summarize content that already ranks near the top, so a strong ranking on a citable page is now partly a bid to be the source the answer is built from. Conversions and revenue, finally, never lose meaning — they are the one place the funnel still terminates in a measurable outcome, and the ultimate check on whether any visibility metric is translating into business.
Some metrics have quietly become traps — not because they are inaccurate, but because their meaning has drifted while the dashboard label stayed the same. Raw organic-click volume, read alone, is the biggest one: it now conflates real performance change with the structural decline of the click itself, so a falling number tells you almost nothing about whether your content is winning or losing. Average click-through rate on informational queries is a close second — AI Overviews compress CTR across the board, so a lower rate may mean nothing about your page and everything about whether an answer box appeared above it. Position-one obsession is the third: being the top blue link matters far less when an AI summary sits above position one and answers the question before anyone scrolls to you.
The rule for this category is not to delete these metrics but to stop reading them as standalone verdicts. Segment them — a click-through rate on transactional, high-intent queries is still meaningful because those queries trigger AI answers less often and still route to a click; a click-through rate averaged across informational queries is now noise. And always pair a declining click metric with the awareness proxies (branded search, impressions, AI share of voice) before concluding anything. A drop in clicks alongside rising branded search and citation rate is not a problem; it is the AI era working as designed. A drop in clicks alongside falling everything else is a real problem. The click number alone cannot tell you which situation you are in — that is precisely why it stopped being a headline KPI.
Six metrics fill the gap the click left, and they share one measurement backbone: you cannot passively collect them from an analytics tag, so you build a fixed set of buyer-intent prompts — typically 20 to 50 — and run them through the engines that matter (ChatGPT, Claude, Gemini, Perplexity, Google AI Mode) on a repeating monthly schedule, recording what each answer says. Everything below comes from that recurring panel. The formulas and benchmark ranges behind each are laid out in detail in how AI search visibility metrics are actually calculated; the goal here is to know what each one is for.
Share of voice is the flagship: for a given prompt set, your brand's total mentions divided by the total mentions across all brands, as a percentage. It answers "of all the times a competitor could have been named, how often was it us?" — the AI-era equivalent of share of search. Citation rate measures how often answers cite your content specifically as a source, which is distinct from being mentioned: a model can recommend you without linking you, or cite your page without naming you as the recommendation, and the two behave differently, a distinction drawn out in AI answer visibility and citations. Prompt coverage is the breadth metric — of your full buyer-intent query set, what share names you at all — and it surfaces the queries where you are simply absent, which are usually the highest-value gaps to close.
The remaining three round out the picture. Recommendation rank (or answer position) tracks your average placement when you do appear — first-named carries far more weight than a passing mention three brands down. AI referral traffic is the one metric you can read from analytics: visits arriving from ChatGPT, Perplexity, Gemini, and Google's AI surfaces, which is small today but is the only direct measure of clicks the answer engines do send, and its trend is worth watching as attribution improves. Answer sentiment and accuracy is the qualitative check that the others miss entirely — being mentioned is worthless if the model describes you wrongly or unfavorably, and a mention that misstates your pricing or category can be a net negative, which is why brand-accuracy monitoring belongs on the scorecard (the failure mode is detailed in AI search brand risk). Because AI outputs vary run to run, sample each prompt several times and read the monthly trend, never a single generation, as the truth.
The single most important discipline is refusing to crown a replacement headline number. The clean traffic KPI is not being swapped for a clean AI KPI; it is being replaced by triangulation — several imperfect signals read together, where the pattern across them is trustworthy even though no one of them is. Build a dashboard, not a metric. Put the awareness proxies (branded search, direct traffic, impressions), the AI-search panel (share of voice, citation rate, prompt coverage, recommendation rank), the direct AI referral traffic, and the terminal outcomes (conversions, revenue) side by side, and learn to read the shapes. Rising share of voice and branded search with flat clicks is health. Falling everything is decline. Rising clicks with falling share of voice is a warning that a competitor is winning the answer box even as you hold the link. No single row settles it; the constellation does.
Two practical guardrails keep the scorecard honest. First, hold the inputs fixed: the same prompt set, the same engines, the same monthly cadence, so month-over-month movement reflects reality and not a changed measurement. Change the panel and you have reset your own baseline. Second, match the KPI to the audience for the report. A client or executive wants the outcome view — branded search, share of voice, conversions — not the raw prompt logs; the practitioner needs the full panel to know what to fix. The reporting layer for each is worked through in AI visibility metrics for client reporting and, for the strategy behind the numbers, AI visibility measurement and AI visibility and GEO.
Be precise about the boundary. Kompozy is not a measurement tool — it does not run your prompt panel, compute your share of voice, or sit in your analytics stack, and any of the AI-visibility trackers named in the guides above do that job better. What Kompozy addresses is the other half of the sentence every marketer says after reading a scorecard: "so now we need to move these numbers." A dashboard tells you the score. It does not put points on the board. The AI-search KPIs are supply-driven — share of voice, citation rate, and prompt coverage rise when the same claim, stated consistently, shows up across more of the surfaces an answer engine pulls from: your blog, your social posts, your video, corroborated enough that a model grows confident it is true and names you as the source.
That corroboration-at-volume is exactly what Kompozy produces. It is a full AI content generation and multi-platform publishing engine — 18 output formats across the eight social platforms plus blog and email — driven by one Persona Brief that fixes your voice, claims, and positioning so every piece says the same thing wherever it lands. From a single source it generates the structured, citable blog article built to be the paragraph an AI answer quotes, plus the same message restated as text posts, Quote Graphics, Carousel Posts, and a talking-head Persona Short for the social and video surfaces those answers increasingly cite — HyperFrames keeping every asset brand-exact. That is the cross-surface corroboration a share-of-voice number rewards, produced in one pass instead of five separate workflows.
It also feeds the legacy KPIs the scorecard keeps. Kompozy generates an email newsletter from the same brief — an owned channel that drives the direct traffic and branded search a rising KPI depends on — and Autopilot schedules and publishes the whole spread across the supported platforms plus blog and email from one queue, behind a per-post review gate so a person signs off before anything ships. The steady cadence matters because AI models weight fresh, corroborated, consistently-published content, and a stop-start publishing pattern shows up as a stalled share-of-voice line. The honest framing: Kompozy will not measure your KPIs and cannot promise a citation — no tool can. It removes the production ceiling that otherwise leaves you with a scorecard full of numbers you have no realistic way to feed.
AI search did not make SEO KPIs obsolete; it split them into three groups and added a new set. Keep the KPIs that measure demand and citability — branded search, direct traffic, impressions (including Search Console's AI surfaces), rankings on citable pages, and conversions — because several of them are now your only window into AI-driven awareness that never produced a click. Stop trusting the click-based metrics that now conflate structural decline with performance: raw organic clicks, blended informational CTR, and position-one obsession. Add the AI-search panel — share of voice, citation rate, prompt coverage, recommendation rank, referral traffic, and answer sentiment — gathered from a fixed prompt set run monthly through the major engines. Then refuse to name a single headline number: read the whole constellation together, because in a zero-click world the pattern across imperfect signals is trustworthy and no one signal is. That unified scorecard is the KPI framework the AI-search era actually requires.
Six carry most of the load. Share of voice — your percentage of total brand mentions across a fixed prompt set. Citation rate — how often AI answers cite your content. Prompt coverage — how many of your buyer-intent queries name you at all. Recommendation rank — your average position when you do appear. AI referral traffic — visits arriving from ChatGPT, Perplexity, Gemini, and Google's AI surfaces. And answer sentiment/accuracy — whether the model describes you correctly and favorably. Track them monthly against the same prompt set so the trend, not the snapshot, is what you read.
The ones that measure demand and citability rather than clicks. Branded search volume and direct traffic rise when AI answers mention you without a click, so they become a proxy for AI-driven awareness. Impressions — including Search Console's AI Overviews and AI Mode impressions — show you are being surfaced even when clicks fall. Rankings for informational pages still matter because top-ranked, well-structured pages are what AI Overviews summarize. Conversions and revenue never stop mattering. What loses meaning is raw organic-click volume read in isolation.
Run a fixed set of buyer-intent prompts — usually 20 to 50 — through the AI engines you care about (ChatGPT, Claude, Gemini, Perplexity, Google AI Mode) on a repeating monthly schedule. For each answer, record every brand named. Share of voice is your brand's mentions divided by the total mentions across all brands in that prompt set, expressed as a percentage. Because outputs vary run to run, sample each prompt multiple times and track the trend across months rather than reading a single snapshot as fact.
Because neither set is complete alone, and swapping one for the other trades one blind spot for another. AI-search KPIs like citation rate and share of voice measure whether models mention you, but they do not measure demand, conversion, or the informational rankings that feed the answers in the first place. Legacy SEO KPIs still capture those. The reliable approach is triangulation: read branded search, impressions, AI share of voice, citation rate, and referral traffic together, and trust the pattern across them even though no single metric is trustworthy on its own.
A scorecard tells you the score; it does not put points on the board. The AI-search KPIs are supply-driven — share of voice, citation rate, and prompt coverage rise when the same claim is corroborated across more surfaces an answer engine pulls from. Kompozy is the production engine on the input side: from one Persona Brief it generates a citable blog article plus the same message as social posts, images, and short-form video, then publishes across the eight social platforms plus blog and email on a steady cadence behind a review gate — the supply that feeds the numbers.
The core AI-search KPIs are share of voice (your share of brand mentions across a fixed prompt set), citation rate, prompt coverage, recommendation rank, and AI referral traffic. Pair them with the SEO KPIs that still matter — branded search, impressions, and rankings for citable pages — and read the whole set together, because in a zero-click world no single number is reliable on its own.
Get started → · ← All guides · Compare Kompozy vs other tools