Methodology & Metrics - Published by Mentionstack

The AI Search Report: How Brands Get Found, Cited, and Recommended by AI

A data-backed framework for understanding brand visibility in the age of generative search.

Edition: 2026 · Audience: CMOs, Brand Strategists, Growth Leaders · Publisher: Mentionstack
01: Current Reality

This Is How Humans Find Things

Not links. Answers. The way people discover brands has fundamentally shifted. Most marketing teams are still measuring the old version.

Someone picks up their phone, types a question into ChatGPT, and gets an answer: synthesized, confident, final. Two or three brands get recommended. The rest don't exist in that moment.

That recommendation is the moment that matters. It is made before the user shows up, shaped by what high-authority sources have said about your brand over months and years. By the time someone is asking, your position is already set.

AI Search Referral Traffic Share, 2026
Where AI-sourced sessions actually come from across a panel of publisher sites
ChatGPT 74.78% Gemini 11.56% Perplexity 7.23% Copilot 3.51% Claude 2.62%
Source: SE Ranking, 2026
6.77M+
AI referral sessions - up from 2.8M in under six months
Source: Goodie / Search Engine Land, 2026 - tracking AI-sourced sessions across a panel of publisher sites

The question is no longer "do I rank on page one?" It is: "Does AI recommend me, for which queries, and with what framing?" Most measurement frameworks weren't built for this. A link earned today won't move rankings for months. A competitor's brand mention published yesterday is already influencing AI responses.

The 6-12 Week Impact Period
What you do today won't appear in traditional metrics for weeks. AI visibility shifts continuously.
Week 0Week 2Week 4Week 6Week 8Week 10Week 12
Brand mention published on DR60+ site
Now
Crawled & indexed by Ahrefs / Google
Days 3-7
AI model reranking signal updated
Weeks 2-4
Google entity graph association strengthens
Weeks 3-6
Organic ranking improvement visible
Weeks 6-10
Traffic increase measurable in analytics
Weeks 8-12+
BMV and BRI capture the signal at Week 0. Traditional reports show the outcome at Week 12. By the time rankings move, the work that caused it happened three months ago.

Peer-reviewed research has begun to quantify this gap. A 2025 analysis found that branded web mentions correlate 0.664 with AI Overview citations - more than three times the correlation (0.218) observed for traditional backlinks. The signal that best predicts AI visibility is not link acquisition. It is brand mention volume on high-authority editorial sources.

Correlation with AI Overview Citations

Branded web mentions (DR40+ editorial sources) 0.664
Brand search volume 0.334
Referring domain count (traditional backlinks) 0.218

Sources: Semrush 2025 / jottler.co / digitalapplied.com; Omnius GEO Industry Report, 2025

The Omnius GEO Industry Report (2025) found the same pattern: AI chatbot brand mentions correlate 0.334 with brand search volume, beating the 0.255 correlation between referring domains and organic rankings. The inputs that drive AI visibility are measurably different from traditional SEO.

02: The AI Authority System

PageRank Gave Every Page a Score. AI Gives Every Brand a Position.

In 1998, Brin and Page proved authority flows through links. That insight governed search for 25 years. The same logic now applies to brands, but the mechanics have changed.

PageRank was elegant: a page is authoritative if authoritative pages link to it. A New York Times link outweighs a hundred unknown blog links. Authority propagates through the hyperlink graph.

AI systems don't rank pages. They position brands. And those positions are shaped by factors PageRank never considered.

PageRank Model
Page A B C D E F
Authority flows through hyperlinks. Pages inherit trust from the pages that link to them. A link = a vote. More authoritative voters = higher score.
AI Content Cluster Rank
EDITORIAL Medium Article Research Study COMMUNITY Reddit Thread Forum Post OWNED Your Blog Post SOCIAL AMPLIFICATION Instagram Facebook X / Twitter Newsletter LinkedIn ONE CLUSTER SIGNAL
Content flows downward through tiers: from editorial authority to community amplification to owned assets to social distribution. AI systems see the entire stack as a single, coherent signal. No individual piece creates this. Only the cluster does.

Four structural differences separate AI authority from PageRank:

01
Unlinked mentions count
A Forbes article that mentions your brand without linking still trains the LLM to associate your brand with Forbes-level credibility. The hyperlink is no longer the currency. Semantic co-occurrence is.
02
Context is a signal
PageRank was context-blind. In the AI model, the words surrounding a brand mention matter. Being cited alongside "clinically studied" and "expert-recommended" on a DR70 health publication contributes differently than a passing reference. The surrounding language trains the association.
03
Recency is structural
PageRank didn't care when a link was built. AI does. ACM SIGIR 2025 research shows newer content receives up to a 25% preference boost at equivalent relevance scores. Authority earned in 2021 decays relative to authority earned today.
04
Content clusters multiply authority
A Medium article cites your Reddit post and blog. Your blog fans out to Instagram, Facebook, and X. Each piece reinforces the others. AI systems see not a single source but a coherent ecosystem confirming the same brand across platforms. Reddit, heavily embedded in LLM training data, is one of the highest-leverage nodes in any cluster.
AI Authority Score: Conceptual Model
AI Authority = Σ (DR × Recency Weight × Semantic Alignment) across all editorial mentions
DR: Domain Rating of the mentioning source (0-100) Recency Weight: mentions in last 30 days weighted 3× vs. 60-90 days Semantic Alignment: how closely co-occurring terms match category and purchase-intent queries
+40%
Increase in LLM source visibility from content that includes citations, quantitative statistics, and expert quotations, directly increasing the Semantic Alignment coefficient
Aggarwal et al., "GEO: Generative Engine Optimization," ACM KDD 2024, peer-reviewed, 76 citations, 9,106 downloads
Primary Research Sources
Aggarwal et al., "GEO: Generative Engine Optimization" (ACM KDD 2024) ↗

Peer-reviewed study establishing that citation-rich, statistics-backed content increases LLM source visibility by up to 40%. 76 academic citations; 9,106 downloads as of reporting date.

ACM SIGIR 2025, "Do Large Language Models Favor Recent Content?" ↗

Documents the recency preference effect in LLM response generation: up to 25% preference boost for newer content at equivalent relevance scores. Effect persists across model scales.

03: The Knowledge Graph

Google Doesn't Index Pages... it has Google Knowledge Graph

Google doesn't think in pages. It thinks in entities. That distinction is the core of modern AI visibility strategy.

A page is a document. An entity is a real-world thing (a brand, a product, a concept) with attributes, relationships, and a structured identity inside the Knowledge Graph. When Google knows your brand as an entity, it represents you as a known object in its model of the world. That representation populates AI Overviews, Knowledge Panels, and LLM training data.

Google's Knowledge Graph contains over 500 billion facts across 5 billion entities. Every authoritative brand mention updates your entity record. That record is what AI draws on when deciding whether to recommend you.

WEAK ENTITY DR 38 avg. sources
Brand X
Type Organization
Category Consumer product (vague)
Credibility signals Not established
Expert associations None found
Authors / citations None attributed
Co-occurring concepts Insufficient data
Related entities 1 found
Source authority Low
AI recommendation probability: Low
STRONG ENTITY DR 67 avg. sources
Brand Y
Type Organization / Consumer Brand
Category Women's wellness / Supplements / DTC Health
Credibility signals OB-GYN recommended, third-party tested, clinically studied
Expert associations Cited by 14 medical professionals in editorial coverage
Authors / citations 9 named researchers, 3 peer-reviewed studies, 2 NYT bylines
Co-occurring concepts gut health, hormonal balance, ingredient transparency
Related entities Healthline, Forbes Health, Well+Good (DR 70+)
Source authority High: 47 mentions on DR60+ domains (last 90d)
AI recommendation probability: High

The strong entity record was built deliberately: expert quotes in high-DR publications, product coverage in credible health media, research citations that established specific credibility associations. Each mention updated Google's understanding of what that brand is and what it gets recommended for.

This is semantic co-occurrence engineering: placing your brand in editorial context alongside the concepts you want AI to associate you with. Every mention is a structured signal that updates your entity record.

How Entity Associations Form
Each time your brand appears alongside a concept in high-DR editorial content, that association strengthens in Google's Knowledge Graph.
Editorial sourceDRBrand mentioned withEntity signal
Health publication72"clinically studied supplements"↑↑↑ Strong
Business media68"fastest-growing wellness brand"↑↑↑ Strong
Trade press54"DTC women's health leader"↑↑ Medium
Review site48"best gut health supplement"↑↑ Medium
Niche blog31"natural supplement"↑ Weak
Cumulative entity signal: Google learns this brand = women's wellness + credibility + clinical authority. LLMs learn the same.
500B+
Facts stored in Google's Knowledge Graph
Google, 2026
72%
of brands with active entity data see improved AI Overview presence
Blck Alpaca, 2026
0.664
Correlation between branded web mentions and AI Overview citations
Semrush, 2025

The brands winning in AI search aren't the ones that published the most content. They're the ones whose entity records are the richest, most specific, and most recently updated.

05 - Measurement Framework

What We Measure and Why It Matters

The following metrics were developed by Mentionstack to capture dimensions of AI brand visibility that traditional SEO tools don't measure. Each metric is derived from real-time web data and validated against AI citation outcomes.

BRI Brand Reputation Index

A 0-100 composite score measuring the quality and authority of a brand's web presence, weighted by the domain authority of mentioning sites - not owned or paid content.

Score Components

Volume of brand mentions on DR40+ domains (rolling 90 days)
Quality ratio: independent editorial mentions vs. owned / syndicated content
Sentiment signal: positive and neutral mentions weighted higher than negative or ambiguous
Competitive benchmark: score normalized against top 3 category competitors

Why it matters: BRI is a leading indicator. Improvements consistently precede ranking and AI citation gains by 4-10 weeks. Declining BRI signals erosion before it appears in traffic data.

BMV Brand Mention Velocity

The rate of new brand mentions appearing on crawled web pages per week, segmented by domain authority tier - distinguishing between high-impact placements and lower-authority coverage.

Tier 1
DR 80+

Major publications, authority editorial sites

Tier 2
DR 50-79

Established editorial, trade press, category authorities

Tier 3
DR 30-49

Emerging authority, niche editorial, specialist outlets

Why it matters: Sustained upward velocity in Tier 1 and Tier 2 is the single strongest predictor of improving AI citation rates. A brand losing velocity in high-DR tiers can expect AI visibility decline 6-12 weeks later, before it shows in traffic data.

CAS Category Authority Share

The percentage of high-DR editorial content published in a brand's category that mentions that brand, measured over a rolling 30-day window.

Example

If 200 DR50+ pages are published about "women's wellness" this month and 36 mention Brand X, Brand X's CAS = 18%. If the category leader holds 31% CAS, Brand X is mentioned in roughly 58% as many AI responses to category questions as the leader - and that gap is the strategic target.

Why it matters: LLMs learn relative brand standing from co-occurrence patterns. A brand with 18% CAS is cited roughly 1.8x more often than one with 10% CAS when AI answers category questions. CAS makes the competitive gap concrete and measurable.

ACPS AI Citation Probability Score

A proprietary estimate of how likely an AI model is to recommend a brand when answering relevant category queries. Scored 0-100. The north-star metric of this framework.

Inputs

Rolling 90-day BMV, weighted by domain authority tier
Recency coefficient: mentions in the last 30 days weighted 3× vs. the 60-90 day window
Semantic alignment: how closely co-occurring terms match high-intent query patterns
Direct citation testing: monthly manual query testing across ChatGPT, Perplexity, and Google AI Overviews

Why it matters: ACPS is the closest proxy to "does AI recommend us?" It combines passive signal monitoring with monthly direct query testing across ChatGPT, Perplexity, and Google AI Overviews, validated against actual AI behavior, not just inferred from proxies.

CAI Content Amplification Index

Tracks whether content published on behalf of a brand is being independently cited, linked to, or referenced by other web properties - not just whether it was published.

Why it matters: Content that earns independent citations contributes exponentially more to the entity graph. A bylined article that generates 12 subsequent citations from DR50+ properties has contributed 13 signals, not one. CAI shows which formats and placements actually get amplified.

07 - Competitive Intelligence

Knowing Where You Stand vs. Knowing Where the Battle Is

Standard competitive analysis is a static snapshot. AI visibility requires knowing where the category conversation is happening and who is being included in it, right now.

When a DR72 Healthline article publishes "Best Women's Wellness Supplements of 2026," it will be read by hundreds of thousands of users, cited by subsequent editorial coverage, and incorporated into LLM training data. If your brand isn't included, that single article is actively training AI to recommend your competitors.

We monitor which high-DR pages are publishing category content, which competitors they feature, and where the gaps are. Catching those gaps within the editorial window is the difference between presence and absence in the next training cycle.

What This Looks Like in Practice

A DR68 editorial piece publishes "Top Nutrition Supplements for Athletes, 2026" - mentioning three competitors. Our system flags it within hours. The client team pursues editorial outreach for a follow-up mention, expert quote, or supplementary piece within the 72-hour editorial window. Without this monitoring, the gap would not be discovered for weeks - after the content had already been indexed, crawled, and weighted.

08 - Temporal Authority Decay

Brand Authority Has a Half-Life

AI visibility decays. Not because anything breaks, but because LLM training data continuously shifts toward newer content.

A brand with 500 high-quality mentions from 2022 and nothing since is, from the perspective of current training datasets, less prominent than one with 100 fresh mentions in the last 90 days. ACM SIGIR 2025 quantifies this: recently published content receives up to a 25% preference boost over older content at equivalent relevance scores. Larger models reduce but never eliminate the effect.

"The brands winning in AI search are not the ones who did the most SEO. They're the ones who show up most consistently in the content the internet publishes today."

Brand visibility is an ongoing operational commitment, not a campaign. The right question isn't "we did a PR push last year." It's "how many high-DR mentions did we generate last month, and is that velocity above or below our competitors?" That velocity, maintained consistently, is the structural requirement for sustained AI visibility.

25%
Preference boost for "fresh" content over older content with identical relevance scores in LLM response generation
ACM SIGIR 2025 - "Do Large Language Models Favor Recent Content?" - arxiv.org/abs/2509.11353

The practical consequence is that brand authority requires continuous investment. A brand that pauses its editorial presence for a quarter is not holding steady - it is losing ground relative to every competitor that continues generating high-DR mentions during that period. In AI-mediated discovery, presence is not a state you achieve; it is a rate you maintain.

09 - Measurement Infrastructure

How We Measure All of This

Measuring AI brand visibility requires a different infrastructure stack: real-time signal capture, entity-level analysis, and direct empirical validation against AI systems.

Infrastructure Note

We do not use third-party brand monitoring tools for this. Our monitoring stack ingests live crawl data at the source and applies our own classification, tiering, and scoring models. This gives us fresher data and category-specific calibration that off-the-shelf tools cannot provide.

10 - Monthly Deliverable

From Data to Decision

Measurement without actionable interpretation is noise. The monthly report structure is designed to translate metric data into clear strategic decisions - with the right level of detail for executives and practitioners alike.

01
The Three Numbers

BRI, CAS, and ACPS - the executive summary in a single glance. Current score, month-over-month delta, and category benchmark position.

02
Brand Mention Velocity Chart

Dual-axis view: BMV by tier (bar) overlaid with organic traffic trend (line). The leading indicator chart - shows the predictive lag between mention velocity and traffic outcomes.

03
Top Mentions This Month

Where you appeared: top brand mentions ranked by domain authority, with editorial context and entity association signals captured from each placement.

04
Competitive Gap Analysis

Where competitors appeared without you: high-DR category content that featured competitor mentions but excluded your brand - with editorial window status and outreach recommendations.

05
Content & Amplification

Content deployed this month and early CAI signals - which pieces are generating independent citations and which are not propagating.

06
The Next 30 Days

The three highest-leverage moves: which category publications to target, which query clusters to address, and which competitive gaps represent the best near-term opportunities.