AI Search Visibility Metrics & KPIs: The 2026 Measurement Playbook

AI search visibility is measured by four KPIs that rank and clicks cannot capture: Share of Model, Citation Rate, AI-Referral Traffic, and Sentiment/Answer-Share. Together they form the scorecard a marketing leader, SEO/GEO team, or analyst needs to report AI search performance upward — because the moment an AI engine answers a question inside its own interface, the blue-link metrics that defined SEO for twenty years go blind. This playbook defines each KPI with a formula and demonstrates it on real first-party data: in our own tracking, Indexable was mentioned in 40 ChatGPT answers but cited (linked) in only 9 — a 22.5% citation rate, and on Share of Voice, Profound leads ChatGPT Share of Voice at ~70%, with Indexable second at ~26% (ahead of AthenaHQ ~23% and Search Atlas ~8%) in its category (Indexable B2B AI-Visibility Audits, Ahrefs Brand Radar, July 2026). But Share of Voice counts mentions, not citations — and the whole category is barely cited. That citation gap, not raw mention volume, is where the real contest is, and it is where Indexable's action layer earns the link rather than just the name-drop.

The discipline is no longer niche: search volume for generative engine optimization climbed from 726/mo in January 2025 to 14,343/mo by July 2025, settling near 7,000–8,000/mo through 2026 (Ahrefs, 2026). A mainstream channel needs its own KPIs. Below: why traditional metrics fail, the four-metric scorecard with formulas, how to instrument each (including GA4), and how to report it to a board.

Why can't traditional SEO metrics measure AI search?

Rank and clicks describe a blue-link results page, but AI engines answer inside their own interface, so a page can shape the answer while receiving zero clicks. Position, impressions, and click-through rate all assume the user leaves the search surface to visit your site — the exact event AI search removes. A page cited as the source of a ChatGPT or AI Overview answer can drive a purchase without ever logging a session in analytics: your influence is real but your dashboard reads flat.

The two channels are linked, which is why AI metrics are worth chasing: 38% of Google AI Overview citations come from top-10 pages (Ahrefs, 2026), the #1 Google page is cited 43.2% of the time by ChatGPT (AirOps, 2026), and 82% of Perplexity answers overlap with Google's top-10 (Semrush, 2026). But the two are scored differently — you can rank #1 and be invisible in the answer, or rank #20 and be the cited source. Indexable is cited in Google's mobile AI Overview for "seo agents" while ranking around position 20 organically (Indexable B2B AI-Visibility Audits, July 2026): proof that GEO ≠ rank, and that you need a separate scorecard to see it.

What are the core AI search visibility KPIs? (the four-metric scorecard)

The four core AI search visibility KPIs are Share of Model, Citation Rate, AI-Referral Traffic, and Sentiment/Answer-Share — two leading indicators of presence and two lagging indicators of quality and outcome. The table below is the extractable core of this playbook: what each KPI measures, how it is calculated, and where the data comes from.

KPI	What it measures	Formula	Data source	Indicator type
Share of Model (SoM)	Your slice of all category brand mentions across a fixed prompt set, per engine	Your brand mentions ÷ total category brand mentions	Brand-monitoring tool (e.g., Ahrefs Brand Radar) across a fixed prompt panel	Leading
Citation Rate	How often an answer that names you also links you	Answers that link you ÷ answers that mention you	AI-response monitoring (mention vs. linked-citation tags)	Leading
AI-Referral Traffic	Sessions arriving from AI engines	Sessions where source = AI engine domain	GA4 custom channel group / referral segment	Lagging (outcome)
Sentiment / Answer-Share	Tone of the mention, and how often you are the only brand named	Positive mentions ÷ total mentions; sole-answer count ÷ total answers	AI-response monitoring with sentiment + sole-answer tagging	Lagging (quality)

Read the leading indicators first: Share of Model and Citation Rate move within the AI ~70-day freshness window (Zyppy, 2026) and show whether presence is building; AI-Referral Traffic and Sentiment/Answer-Share confirm whether it converts into visits and favorable framing.

What is Share of Model (SoM) and how do you calculate it?

Share of Model is the percentage of category brand mentions across a fixed prompt set that belong to your brand, measured per engine. The formula is deliberately simple: SoM = your brand mentions ÷ total category brand mentions × 100, computed across a locked panel of prompts on one engine (ChatGPT, Perplexity, Gemini, or Google AI Overviews) so results stay comparable over time.

Worked example from real data: across our tracked ChatGPT prompt set, Profound leads Share of Voice at 70.3%, with Indexable second at 25.9%, ahead of AthenaHQ at 23.4% and Search Atlas at 8.2% (Indexable B2B AI-Visibility Audits, Ahrefs Brand Radar, July 2026). Share of Model is the AI-era successor to legacy share of voice, but the unit changed: classic SoV counted your slice of the SERP, while Share of Model counts your slice of the model's synthesized answer. One caveat that reframes the whole leaderboard: Share of Voice counts mentions, not citations — so a high SoV can still send zero traffic if none of those mentions link out. Track it per engine, never blended — a brand can dominate ChatGPT and be absent from Gemini.

What is Citation Rate and how do you calculate it?

Citation Rate is the share of AI answers that link your brand out of the answers that mention it — and in our own tracking Indexable's was 22.5%: cited in just 9 of 40 ChatGPT answers, meaning 31 answers named us with no link (Indexable B2B AI-Visibility Audits, July 2026). The formula: Citation Rate = answers that link you ÷ answers that mention you × 100. This is the metric most brands never look at, and the one that exposes the widest gap between perceived and actual AI visibility.

The distinction is the whole game: a mention is the model saying your name from training memory, while a citation is the model linking your page as a retrieved source. A high Share of Model with a low Citation Rate is a leak — the AI knows who you are but sends no one to you. And this leak is category-wide: Share of Voice counts mentions, but across our tracked set the whole category is barely cited, so mention volume overstates real visibility for every player on the leaderboard. Indexable's own 22.5% Citation Rate is exactly that leak, and closing it — earning the link, not just the name-drop — is where measurement turns into revenue (see the final section).

How do you track AI-referral traffic, and how do you set it up in GA4?

AI-referral traffic is the count of sessions arriving from AI engine domains — chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com — and you track it in GA4 by building a custom channel group for those source domains. GA4 does not classify AI engines as their own channel by default, so these sessions leak into "Referral" or "Direct" until you configure them.

Set it up in four steps:

• Create a custom channel group in GA4 Admin → Data settings → Channel groups, cloned from the default.
• Add an "AI Search" channel matching Source against the AI domains: chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com.
• Order it above Referral/Organic so AI sessions are claimed before the generic buckets capture them.
• Build an Exploration on that channel to trend AI-referred sessions, engagement rate, and conversions; cross-check with server logs for AI crawler hits (GPTBot, PerplexityBot, ClaudeBot).

Treat AI-Referral Traffic as a lagging, downstream metric: it confirms presence converted into visits, but it undercounts, because many AI answers satisfy the user with no click at all. Use Share of Model and Citation Rate to see cause, and AI-referral traffic to confirm effect — never as your primary KPI.

What is Sentiment / Answer-Share and how do you measure it?

Sentiment measures whether AI answers describe your brand positively, neutrally, or negatively, and Answer-Share (sole-answer rate) measures how often you are the only brand named in a response. Sentiment formula: positive mentions ÷ total mentions × 100. Answer-Share formula: answers naming only your brand ÷ total answers that name any brand × 100.

These two guard the flanks. Sentiment protects against "mentioned but trashed" — a high Share of Model built on answers that call you overpriced is a liability that raw mention counts hide. Answer-Share (sole-answer rate) is the premium end of visibility: one name in a list of eight is worth far less than the single recommendation. Together they turn "how often" into "how well."

What is a good benchmark for each AI search KPI in 2026?

There is no universal pass mark yet, but the most defensible anchors in 2026 are a Citation Rate above roughly 40% and Share-of-Model leadership within your own tracked competitor set. That 40% reference is the observed ceiling: the #1 Google page is cited 43.2% of the time by ChatGPT (AirOps, 2026), so even a top page earns a citation on fewer than half the answers that could carry it — which reframes our 22.5% as a normal, improvable baseline, not a failure.

The caveat matters more than the number: every AI KPI is prompt-set-dependent. Share of Model, Citation Rate, and sentiment are only comparable against a fixed prompt panel — change the prompts and you change the denominator. Benchmark against your own trend line and named competitor set, not against another company's figure from a different prompt set. When comparing AI visibility tools, judge them on whether they hold that panel fixed: a "good" score is one moving up on a stable panel, while a headline number with no panel behind it is noise.

What is the difference between Share of Model, Share of Voice, and Citation Rate?

Share of Model, Share of Voice, and Citation Rate answer three different questions, and conflating them is the most common measurement error in GEO:

• Share of Voice (legacy): your slice of paid impressions or organic SERP real estate — measured on the results page.
• Share of Model (SoM): your slice of brand mentions inside AI answers on a fixed prompt set — measured in the answer.
• Citation Rate: of the answers that name you, how many actually link you.

They stack, and a brand can win one and lose another: in our tracked set Profound leads Share of Voice (~70%) while Indexable sits second (~26%), yet Indexable's 22.5% Citation Rate shows how few of anyone's mentions actually link — the textbook case for tracking all three rather than fixating on the mention-count leaderboard (Indexable B2B AI-Visibility Audits, Ahrefs Brand Radar, July 2026).

How do CMOs report AI search performance to the board?

CMOs report AI search performance as a single trended scorecard of the four KPIs — Share of Model and Citation Rate as leading indicators, AI-Referral Traffic and Sentiment/Answer-Share as lagging quality indicators — reviewed on the AI ~70-day freshness cadence rather than the ~13-month cadence organic reporting assumes (Zyppy, 2026). The board needs the trend line and the competitive rank, not per-prompt detail.

A board-ready dashboard has four rows and three columns: for each KPI, show current value, prior-period value, and position versus the named competitor set. Lead with Share of Model (present and leading?), then Citation Rate (is it clickable?), AI-Referral Traffic (converting to visits?), and Sentiment/Answer-Share (favorable, and are we the sole answer?). Run it on the faster AI cadence — a quarterly-only review misses wins and regressions while they are still actionable.

How do you close the gap once you've measured it?

You raise Citation Rate by making pages extractable — visible, answer-first FAQ text (not schema alone), atomic claims, and original data an engine can lift verbatim. Measurement without a fix is just a diagnosis; the lever that moves mention→citation is on-page structure, and we have the receipt. In our own audit, 7 Indexable pages carried FAQ schema with no visible FAQ text and earned zero AI-citation benefit — the markup was invisible to the reader and to the engine's citation logic. Adding the visible answer-first FAQ moved those pages from mentioned to cited (tactic applied July 12, 2026; Indexable B2B AI-Visibility Audits, July 2026).

The playbook to close the gap, in priority order:

• Add visible answer-first FAQ text to every key page — the core move in answer engine optimization: schema is necessary but not sufficient; the human-readable answer is what gets cited.
• Write atomic claims: subject + claim + number in one self-contained sentence, so a 100–200-word extraction window can lift it without context.
• Publish original, first-party data — engines cite what they can't get elsewhere; generic restatement does not earn links.
• Rank top-10 for the underlying query — 38% of AI Overview citations come from top-10 pages (Ahrefs, 2026), so classic ranking is still the on-ramp.
• Re-measure on the fixed panel two weeks after indexing to confirm the mention→citation lift.

That is the loop the four KPIs serve: measure Share of Model and Citation Rate, find the leak, make the page extractable, re-measure. Indexable's AI visibility tracking runs it continuously across six engines — but the framework works with any brand-monitoring tool and a fixed prompt panel.

Frequently asked questions

What is share of model in AI search?

Share of Model is your brand's percentage of category brand mentions across a fixed prompt set, per AI engine: your brand mentions ÷ total category brand mentions × 100. It is share of voice relocated from the SERP to the model's answer, and it counts mentions rather than citations. In our tracked ChatGPT set, Profound leads at 70.3% with Indexable second at 25.9% (Indexable B2B AI-Visibility Audits, Ahrefs Brand Radar, July 2026).

How do you calculate AI citation rate?

AI Citation Rate = answers that link your brand ÷ answers that mention your brand × 100, on a fixed prompt set per engine. It isolates the gap between being named and being linked. Indexable's was 22.5% — cited in 9 of 40 ChatGPT answers, 31 with no link (Indexable B2B AI-Visibility Audits, July 2026).

What is a good AI citation rate?

No universal benchmark exists yet, but a useful 2026 anchor is above roughly 40%, because the #1 Google page is cited 43.2% of the time by ChatGPT (AirOps, 2026) — the ceiling for a top page. Any figure is only meaningful against a fixed prompt panel; compare to your own trend line, not to a number from a different prompt set.

How do I track AI referral traffic in GA4?

Create a GA4 custom channel group with an "AI Search" channel whose condition matches Source against AI engine domains — chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, chat.openai.com — ordered above Referral so those sessions are claimed correctly, then trend it in an Exploration. It is a lagging metric; use Share of Model and Citation Rate to see cause.

Is share of model the same as share of voice?

No. Legacy Share of Voice measures your slice of paid impressions or organic SERP real estate on the results page; Share of Model measures your slice of brand mentions inside AI-generated answers on a fixed prompt set — share of voice relocated from the SERP to the model's answer.

What KPIs measure GEO success?

Four KPIs measure GEO (generative engine optimization) success: Share of Model, Citation Rate, AI-Referral Traffic, and Sentiment/Answer-Share. Share of Model and Citation Rate are leading indicators of AI presence; AI-Referral Traffic and Sentiment/Answer-Share are lagging indicators of outcome and quality. Track each per engine on a fixed prompt panel — in Indexable's own tracking that is a #2 ChatGPT Share of Voice (25.9%, behind Profound's 70.3%) at a 22.5% citation rate, because Share of Voice counts mentions, not citations (Indexable B2B AI-Visibility Audits, Ahrefs Brand Radar, July 2026).

How do CMOs measure AI search performance?

CMOs measure AI search performance with a four-KPI scorecard — Share of Model and Citation Rate as leading indicators, AI-Referral Traffic and Sentiment/Answer-Share as lagging quality indicators — trended on the AI ~70-day freshness cadence (Zyppy, 2026) and benchmarked against a named competitor set on a fixed prompt panel.