The AI search KPIs now circulating measure the answer and ignore the pages that build it. A September 23, 2026 BigGo Finance report on Japanese marketing firm Digital Identity lists three: brand mention rate, LLM-referred traffic and conversions, and branded search plus citations. All three are outputs. The input, meaning the publisher pages AI engines retrieve before they answer, is where the next media market sits.

  • AI search KPIs built only on mention rate report an outcome the brand cannot directly move.
  • SEO always paired an output metric, rank, with an input metric, backlinks; AI Search frameworks have copied only the output half.
  • The widely quoted 63.2% zero-click figure from CyberAgent measures how many users sometimes stop at AI Overviews, not the share of searches that do.
  • Source share, a brand's presence on the pages AI engines retrieve for category questions, is the missing input KPI and the unit publishers can sell.

What an input KPI for AI Search is

An AI Search input KPI measures the conditions that produce a brand mention, not the mention itself. The clearest one is source share: the percentage of pages an AI engine retrieves for a category question on which a brand is present, accurately and favorably. Mention rate tells you what the answer said. Source share tells you why it said it.

Rank and backlinks vs mention rate and source share

For twenty-five years, search KPIs came in pairs. Rank was the output and backlinks were the input, and PageRank made that input measurable. Brands moved rank by moving links, and an entire industry, legitimate and not, priced that input.

In the agentic web, mention rate is the new rank. The new backlink is presence on the third-party pages that retrieval systems pull into the answer. Most AI Search dashboards track the first and have no line for the second.

What BigGo Finance and Digital Identity reported, and what we verified

BigGo Finance summarizes Digital Identity's framework and tactics: topic clusters over single keywords, FAQ content, conclusion-first writing, one topic per heading. It explains RAG well, including how an engine splits one question into several sub-searches before synthesizing. The advice is sound and mirrors the standard Japanese LLMO playbook.

The headline number needs a correction. The source says 63.2% of users end their searches with AI Overviews alone. CyberAgent's GEO Lab survey, run October 7 to 8, 2025 among 9,278 respondents, actually measured the share who sometimes, frequently or often stop there, and found 54.2% at least sometimes click links inside AI Overviews. For behavioral data, a VALUES and note study published October 30, 2025 estimated that 63.5% of Japanese Google sessions ended without a site visit. The direction holds, the unit differs, and we use the behavioral figure.

Why brand mention rate cannot be managed on its own

The strongest case for output-only AI search KPIs

The case for measuring only outputs is serious. Buyers see the answer, not the retrieval log, so mention rate is the closest proxy for customer experience. SparkToro and Gumshoe's January 2026 study showed consideration sets are stable enough to measure, with leading headphone brands appearing in 55 to 77% of responses. And Muck Rack's May 2026 analysis of 25 million AI-cited links found paid and advertorial content made up just 0.3% of citations. On that reading, inputs are earned, not bought, and belong to PR.

Why an output you cannot move is a report, not a KPI

The same SparkToro study found less than a 1 in 100 chance that two responses return the same brand list. Mention rate is a probability, and the source's own RAG example shows where that probability is set: in the pages the engine retrieves. Muck Rack attributes 84% of citations to earned sources the brand does not own.

The 0.3% figure proves advertorials rarely get cited. It says nothing about a brand's presence on the editorial pages that do. That presence is the input worth measuring, and worth pricing.

What this means for brands and for publishers

For CMOs, media buyers and agencies: report source share next to mention rate

Keep Digital Identity's three KPIs and add a fourth. For each priority category question, identify which publisher pages the engines retrieve, then track whether your brand is present on them and how. Budget against that page list the way SEA teams budget against a keyword list. Agencies that can produce this map will sell a media plan; those that cannot will sell a dashboard.

For publishers: retrieval frequency is the audience you can sell

Publishers should measure the mirror image: how often their pages are retrieved for commercial questions, by which engines, in which categories. That is audience data, even when no human clicks. Time's COO Mark Howard told Digiday in July 2026 that AI-bot impressions on authoritative content are scarce and valuable. A publisher that proves retrieval frequency by topic can price it; one that reports only residual traffic will keep discounting.

Three signals the AI Search input market is forming

First, Time and Mobian launched agent-facing ads in July 2026 with the KPI loop built in: a sponsored FAQ on the source page, then prompts run through AI engines to track visibility, favorability and accuracy. Input bought, output measured.

Second, Perplexity, a Time licensing partner, called those units deceptive, so disclosure standards will decide which input formats survive. Third, measurement is professionalizing: 60 to 100 runs per prompt per platform is becoming the floor for any credible mention rate.

Conclusion

Hold on to this: AI search KPIs that stop at the answer describe the market without giving anyone a lever in it. The lever is the source layer, and it needs a price on both sides. Smalk AI builds that layer as Generative Engine Advertising, placing native ads for AI agents on the publisher pages AI engines retrieve, giving brands an input they can buy and publishers revenue for the retrieval they already supply. Watch for the first media plans that list source pages instead of keywords; that is when AI visibility moves from the analytics budget to the media budget.