← Back to blog
How AI Search Engines Decide What to Cite — And What It Means for Your Brand
AI Visibility

How AI Search Engines Decide What to Cite — And What It Means for Your Brand

AI search engines decide what to cite based on three factors: earned authority, entity clarity, and citation architecture. Here is the full breakdown of each factor -- and what your brand needs to do about it.

AI search engines decide what to cite based on three factors: earned authority (third-party coverage in publications AI systems already trust), entity clarity (the degree to which machines can unambiguously identify and categorize a brand), and citation architecture (the structural formatting that makes content independently extractable). SEO ranking position does not reliably predict AI citation - 88% of Google AI Mode citations are not in the organic top 10, according to Moz's 2026 analysis of nearly 40,000 search queries.

That number deserves a pause. The top 10 positions that marketing teams have spent years optimizing for are largely irrelevant to whether your brand appears in an AI-generated answer. A completely separate set of criteria governs AI citation, and most brands have never addressed them.

This article defines those criteria, explains the mechanism behind each, and shows what your brand needs to do to appear in AI search results rather than remain invisible to them. The three factors described here correspond directly to Layers 1, 2, and 3 of the Machine Relations (MR) framework, the only discipline that addresses all three as an integrated system rather than separate optimizations.

Key Takeaways

  • AI search engines operate on citation logic, not ranking logic, the signals that predict AI visibility are fundamentally different from traditional SEO signals.
  • Earned media is the most frequently cited source type across all major AI engines, according to Muck Rack's AI citation research, making third-party coverage in trusted publications the highest-leverage action for AI citation.
  • Muck Rack's "What is AI Reading?" study analyzing millions of AI-cited links found that over 95% come from non-paid sources, with 85% of those originating from earned media—not owned blog content or paid placements.
  • Ahrefs analysis of ChatGPT's 1,000 most-cited pages found that 65.3% come from domains with a Domain Rating above 80, according to Ahrefs' citation research—domain authority built almost exclusively through editorial coverage in major publications.
  • Adding statistics to content improves AI visibility by 30–40%, according to the Princeton/Georgia Tech GEO study (Aggarwal et al., SIGKDD 2024), the single highest-leverage structural change a publisher can make.
  • These three citation factors are Layers 1–3 of the Machine Relations framework, the only approach that treats earned authority, entity clarity, and citation architecture as a single integrated system.

The Broken Assumption: Why SEO Rankings Don't Predict AI Citations

For two decades, getting cited online meant getting ranked. If your page appeared in the top 10, it got traffic and attribution. If it didn't, it was invisible. The signal and the outcome were inseparable.

AI search engines have broken that assumption. The evidence is consistent across multiple independent studies, each measuring AI citation behavior against traditional search performance:

This is not a marginal shift. The systems operate on different selection criteria. A brand that has spent five years building SEO authority through backlink acquisition and keyword optimization may be completely invisible in AI answers, while a competitor with lower domain authority appears consistently, because they have earned media coverage in publications AI engines trust.

Understanding why this happens requires understanding what AI search engines are actually doing when they decide what to cite. The answer is more specific, and more actionable, than most guidance in this space suggests.

How AI Search Engines Make Citation Decisions

Most AI search engines that retrieve live web content use a Retrieval-Augmented Generation (RAG) pipeline. The system converts the user query into a vector embedding, searches its index for semantically relevant content, and then filters and re-ranks candidates before synthesizing a response and attributing sources. The citation is not random. The system makes a selection, and that selection is based on specific, measurable signals.

A September 2025 arXiv study (Kumar et al., arXiv:2509.10762) analyzed 1,702 citations across Brave Summary, Google AI Overviews, and Perplexity using a 16-pillar auditing framework. The study found that overall page quality is a strong predictor of citation, with an odds ratio of 4.2, meaning a high-quality page is four times more likely to be cited than a low-quality one. Pages with a GEO score at or above 0.70 and 12 or more pillar hits achieved a 78% cross-engine citation rate.

More important than the aggregate score is what the individual pillars revealed: the signals most strongly associated with citation are Metadata and Freshness, Semantic HTML, and Structured Data. These are not traditional SEO metrics. They are machine-readability signals, and they point directly to what brands need to build.

Research across multiple studies points consistently to three factors that govern AI citation decisions. They correspond to Layers 1, 2, and 3 of the Machine Relations framework.

Factor 1: Earned Authority. the Publication Trust Signal

Earned authority is the foundation of AI citation: trusted third-party coverage in publications that AI systems already recognize as credible. Without external corroboration, a brand's content is self-assertion, and AI engines systematically deprioritize self-assertion in favor of third-party validation.

The data on this factor is the most consistent finding in AI citation research. A September 2025 controlled study by Chen et al. (arXiv:2509.08919) analyzing citation patterns across multiple verticals found that AI search systems show a systematic and overwhelming bias toward earned media over brand-owned and social content—a stark contrast to Google's traditional organic mix.

  • Earned media is the most frequently cited source type across all major AI engines, according to Muck Rack's analysis of millions of AI-cited links, the highest-leverage input available for improving AI visibility.
  • Muck Rack's "What is AI Reading?" study analyzed millions of AI-cited links and found that over 95% came from non-paid sources, with 85% originating from earned media—not owned blog content, not paid placements, not social posts. Editorial placements in publications that AI engines already indexed and trusted.
  • 65.3% of ChatGPT's top-cited pages come from domains with a Domain Rating above 80, according to Ahrefs' analysis of 1,000 ChatGPT citations—the authority threshold that maps almost exclusively to major editorial publications.

The mechanism is straightforward. AI engines are trained on the web, and the web has a trust hierarchy that has been established over decades. Publications like Forbes, Harvard Business Review, TechCrunch, Wired, and major industry trade outlets have accumulated editorial credibility through consistent, sourced journalism. AI engines inherit that credibility: when they encounter a claim in a high-authority publication, they weight it more heavily than the same claim on a company's own blog. A brand mentioned positively in a Forbes article carries a fundamentally different trust signal than the same brand's website making the identical claim.

This is why PR's original mechanism, earned media in respected publications, has not become obsolete in the AI era. It has become structurally more important. The publications themselves haven't changed. What changed is who is reading them. Machines are now the primary reader of the content that determines brand citation, and they respond to the same editorial trust signals that shaped human perception for decades.

A December 2025 arXiv study (Zhang et al., arXiv:2512.09483) confirmed this dynamic: "recent comparative research emphasises that generative engines heavily weight earned media and often exclude brand-owned and social platforms." The same study found that "even high-quality pages may not be cited if they reside solely on vendor blogs", meaning citation architecture alone, without earned authority, is insufficient.

The practical implication: AI visibility is downstream of editorial relationships. Not ad spend. Not link building. Not keyword density. Placement in publications that AI engines treat as credible sources.

Factor 2: Entity Clarity. Machine Legibility

Entity clarity is the degree to which AI systems can unambiguously identify, categorize, and relate a brand to its category. A brand that AI systems cannot confidently resolve will not be cited confidently, even if that brand has relevant content and earned media coverage.

This is the factor most marketing teams have never addressed, because it was irrelevant in keyword-based search. Keywords didn't require machines to understand who you are, they required machines to match text patterns. Entity clarity requires something fundamentally different: consistent, structured signals across multiple platforms that allow AI systems to build a coherent internal model of who your brand is, what category it belongs to, and what it is known for.

When a user asks ChatGPT "which agency specializes in AI brand visibility," the AI system doesn't just retrieve relevant pages. It attempts to resolve which entities, specific named companies, are credibly associated with that category. Brands that have built clear entity signals appear as resolved candidates. Brands that haven't built those signals are absent from the resolution, regardless of how much content they've published.

The signals that build entity clarity:

  • Consistent naming across platforms, the brand name, founder name, and company description use the same language everywhere: website, LinkedIn, Crunchbase, Wikipedia, press mentions, and any other platform where the brand is mentioned
  • Cross-platform presence, the brand exists as a named entity in multiple independent sources, not just its own web properties; each independent mention strengthens the entity signal
  • Structured data markup. JSON-LD Organization and Person schema on web properties that explicitly defines the company, its founder, its category, and its relationships to other entities
  • Category association, consistent linkage between the brand and the specific category it wants to own in AI responses, established across multiple independent sources

The Kumar et al. arXiv study (2509.10762) found that Structured Data and Semantic HTML were among the pillars most strongly associated with citation across all three AI engines studied. The December 2025 arXiv study (Zhang et al., 2512.09483) found that domains favored by LLM-based search engines exhibited "more structured, hierarchical HTML, easier-to-read text, and more outlinks to reputable sources", the structural signals that enable entity resolution.

Entity clarity is not a technical SEO exercise. It is the process of making your brand legible to machine readers who are trying to determine whether you are the right source for a given query. A brand that is unambiguously associated with a specific category and function, across its own properties, its earned media placements, and third-party data sources, as a brand that AI systems can cite with confidence. A brand that sends inconsistent signals, or that exists only on its own domain, is a brand that AI systems resolve with uncertainty. Uncertain resolution produces fewer citations.

Factor 3: Citation Architecture. structural Extractability

Citation architecture is the structural formatting of content, data density, FAQ sections, tables, and answer-first structure, that makes specific claims independently extractable by AI retrieval systems. A page that reads well for humans but lacks these structural signals is a poor candidate for AI citation, regardless of how it ranks or how authoritative its source publication is.

The foundational research on this factor is the Princeton/Georgia Tech GEO paper (Aggarwal et al., SIGKDD 2024), which studied how content formatting affects AI visibility. Its core findings:

  • Adding statistics to content improves AI visibility by 30–40%, the single highest-leverage structural modification a publisher can make to increase citation rates
  • Tables are cited 2.5x more often than prose by AI systems, because tables present structured data that AI retrieval pipelines can extract without parsing narrative context or inferring meaning from surrounding text

The mechanism is specific to how RAG pipelines work. An AI search engine that retrieves a page doesn't read it the way a human does, it identifies passages that directly answer the query and extracts them as citation candidates. A page that states its key claim in the first sentence, supports it with a named statistic, and presents related data in a table provides clean extraction targets. A page that buries its main point in the seventh paragraph, uses passive voice throughout, and presents data in flowing prose requires the system to do interpretive work that reduces extraction confidence, and therefore reduces the likelihood of citation.

The specific elements of citation architecture that the research identifies as citation signals:

  • Answer-first structure, the main claim appears in the first 40–60 words, stated declaratively, self-contained, usable by an AI system without any surrounding context
  • Data density, a minimum of 12 unique external statistics from named primary sources, inline-cited with direct URLs to the original document; each statistic gives the AI a specific, attributable claim to extract
  • FAQ sections, question-answer pairs that AI systems treat as direct extraction targets; the format mirrors how AI systems receive queries, making extraction more reliable than parsing body prose
  • Tables, structured comparison data that presents relationships between entities, metrics, and claims in a format AI systems can extract and represent without paraphrase
  • Metadata and freshness signals, the GEO-16 study found Metadata and Freshness to be the pillar most strongly associated with citation, including datePublished and dateModified markup, which signals to AI systems that the content is current and worth surfacing

Citation architecture is the layer where most "GEO optimization" efforts concentrate, and it's the most accessible layer for brands to address without external relationships. But the Kumar et al. study is clear that architecture alone does not guarantee citation: content that lives exclusively on brand-owned properties, regardless of how well-structured it is, faces a citation ceiling that earned authority breaks through.

Architecture without authority is optimization without foundation. The three factors work as a system, not as alternatives.

The Scale of the Shift: Why AI Citation Matters Right Now

Two data points define the urgency of getting this right:

AI search now reaches 1.5 billion users, according to Google's 2025 I/O figures. ChatGPT alone reaches over 800 million users per week, nearly double its usage from the year prior. These are not niche early-adopter numbers. They represent the mainstream buyer research journey moving into AI-generated answers.

Nearly 60% of Google searches now end without a click, according to SparkToro's 2024 zero-click search study, the answer is delivered in the AI-generated summary, and users never visit a source page. Pew Research Center confirmed that users encountering AI summaries click on links at just 8% of visits—half the 15% rate on search pages without AI summaries. The implication: brand discovery is shifting from "who ranks" to "who gets cited." A brand absent from AI answers is invisible to the majority of users who never leave the search interface.

70% of B2B buyers complete their research before first contact with a vendor, according to Forrester's State of Business Buying 2024 report. That research now happens substantially inside AI engines. A Forrester Buyers' Journey Survey found that as many as 89% of B2B buyers have begun using generative AI in their purchasing process, at every stage, from category discovery through vendor evaluation. The buyer who asks Perplexity "which agency specializes in AI brand visibility" is doing their due diligence before they ever visit a website. If your brand is not cited in that answer, you are not in the consideration set.

The stakes extend beyond individual queries. Brands that build AI citation infrastructure now, earned authority, entity clarity, citation architecture, accumulate a compounding advantage. Each placement in a trusted publication strengthens entity resolution. Each well-structured piece of content adds another extraction target. Each FAQ section planted in a credible source is a direct input to AI answers. Brands that wait to address these factors are not maintaining the status quo. They are falling behind competitors who are actively building the infrastructure that AI citation requires.

How Different Disciplines Approach AI Citation

The AI citation challenge is being addressed from multiple disciplines simultaneously. Each discipline has real value, and each has a gap. Understanding where each approach sits clarifies why a full-system response is necessary.

Discipline Optimizes for Success condition Scope
SEO Ranking algorithms Top 10 position on SERP Technical + content
GEO Generative AI engines Cited in AI-generated answers Content formatting + distribution
AEO Answer boxes / featured snippets Selected as the direct answer Structured content
Digital PR Human journalists/editors Media placement Outreach + storytelling
Machine Relations AI-mediated discovery systems Resolved and cited across AI engines Full system: authority → entity → citation → distribution → measurement

The gap in every discipline except Machine Relations: each optimizes for one layer of a three-layer problem. SEO addresses some citation architecture signals (technical structure, schema) but ignores earned authority entirely. GEO addresses citation architecture but typically treats earned media as an optional amplification layer rather than the foundational requirement it is. AEO and Answer Engine Optimization agencies are the closest to the full picture, but most operate at the content formatting level without the earned media infrastructure that determines whether that content gets cited in the first place. Digital PR builds earned authority but was not designed for machine extraction, placements in the right publications get AI citations, but the content itself may not be structured for reliable extraction.

Machine Relations is the discipline that addresses all three factors as a single integrated system rather than separate optimizations. Earned authority provides the trust foundation. Entity clarity provides the resolution signal. Citation architecture provides the extraction mechanism. All three must be present for AI citation to occur reliably.

What Brands Get Wrong About AI Citations

Three misconceptions account for most of the strategic errors brands make when trying to improve their AI citation rates:

Misconception 1: More content produces more citations. Volume of published content has no meaningful correlation with AI citation rate. What matters is whether any individual piece of content is independently extractable, self-contained, data-dense, answer-first, sourced from a credible entity. Ten tightly structured pieces with earned authority behind them outperform a thousand generic blog posts in every AI citation study conducted. The GEO-16 research found that page quality, not page count, was the strong predictor (odds ratio 4.2).

Misconception 2: High domain authority guarantees AI visibility. The 6.82% overlap between ChatGPT citations and Google's top 10 results is the clearest evidence against this assumption. Domain authority predicts ranking on Google. It does not predict AI citation. A startup with three Forbes placements and well-structured, data-dense content can outperform an enterprise with a DA-90 domain in AI answers. The citation systems are measuring different signals.

Misconception 3: AI citations are purely a content problem. This is the most costly misconception in practice. Brands invest in content formatting, FAQ sections, and structured data, all useful, while ignoring the earned authority layer that sits beneath them. A page with perfect citation architecture but no third-party corroboration is less likely to be cited than a page with average formatting that lives within a trusted, earned-media-backed domain. The December 2025 arXiv study was explicit: "even high-quality pages may not be cited if they reside solely on vendor blogs." Architecture without authority is optimization without foundation.

How to Get Cited by AI Search Engines: The Practical Sequence

Based on the three citation factors and the research supporting each, here is the sequence of actions that produces measurable improvement in AI citation rates:

  1. Build earned authority first. Secure placements in publications that AI engines already index and treat as credible. Forbes, Harvard Business Review, TechCrunch, Wired, and tier-1 industry publications in your specific vertical. Each placement is a trust transfer from the publication's established credibility to your brand. This is the layer no amount of on-page optimization can replicate. You can read your brand's current AI answer footprint at any point using AI monitoring tools, but fixing it starts with earned media, not with monitoring.
  2. Establish entity clarity across the graph. Audit your brand's presence across the entity graph: Wikipedia, Crunchbase, LinkedIn, your own website schema. Verify consistent naming, category association, and structured JSON-LD markup. Inconsistent descriptions across platforms create entity resolution uncertainty. Uncertainty produces fewer citations, even from brands with strong earned authority.
  3. Structure every piece of content for machine extraction. Answer-first openings that define the core concept in the first 40–60 words. Named statistics from primary sources (minimum 12 for a long-form piece). Tables for comparative data. FAQ sections with declarative question-answer pairs. These are not stylistic preferences. They are the structural signals that make content independently extractable by AI retrieval pipelines.
  4. Publish original research. The dominance of high-DR domains in ChatGPT citations (Ahrefs found 65.3% of cited pages come from DR 80+ domains) is not coincidence. AI engines cite sources that produce data other sources cannot replicate. A single original study, proprietary dataset, or systematic analysis creates a citation asset that earns attribution across every AI engine that encounters the relevant query.
  5. Maintain recency signals. The GEO-16 framework found Metadata and Freshness to be the pillar most strongly associated with citation. Pages with current datePublished and dateModified markup, regularly updated content, and recent references consistently outperform stale content, even when the underlying claims remain accurate. Update timestamps and refresh data annually at minimum.

The sequence matters. Brands that attempt citation architecture before establishing earned authority are optimizing the packaging of content that lacks distribution rights. Brands that earn media placements but fail to structure the content for extraction leave attribution on the table. The three factors work in sequence: authority enables extraction, entity clarity enables attribution, and citation architecture enables reliable, repeatable citation at scale.

Why Earned Media Is the Mechanism That Makes It All Work

Machine Relations (MR) is the discipline of earning AI citations and recommendations for a brand by making that brand legible, retrievable, and credible inside AI-driven discovery. Coined by Jaxon Parrott, founder of AuthorityTech, in 2024, MR extends Public Relations into a world where machines mediate how brands are surfaced, compared, and recommended.

The mechanism behind Machine Relations is the mechanism that made traditional PR valuable: earned media placements in publications that trusted audiences, and now trusted machines, rely on for authoritative information. A Forbes placement that generates an AI citation is not a new phenomenon requiring a new technology stack. It is the original PR mechanism, now operating with machine readers as the primary audience alongside human ones.

PR got one thing exactly right: earned media. A placement in a respected publication, secured through a real editorial relationship, as the most powerful trust signal available for AI citation. It was true when buyers were human readers. It is true now that AI systems perform the first cut of research on a buyer's behalf. The publications that shaped human brand perception for decades are the same publications AI engines treat as authoritative sources. The editorial relationships that produced those placements are the same relationships that now produce AI citations.

What changed is not the mechanism. What changed is the reader. Machine Relations is the name for understanding this shift, and for building the infrastructure that makes earned media produce both human and machine attribution simultaneously.

Brands that have been consistently building earned media infrastructure are discovering that their existing placements translate directly into AI citations. Brands that optimized exclusively for SEO rankings are discovering those rankings produce no equivalent benefit in AI answers. The investment thesis for which brands dominate is being rewritten by this shift.

Start your visibility audit →

Frequently Asked Questions

How do AI search engines decide what to cite?

AI search engines decide what to cite based on three primary factors: earned authority (third-party coverage in publications they recognize as credible), entity clarity (how unambiguously they can identify and categorize a brand or source), and citation architecture (structural formatting that makes specific claims independently extractable). SEO ranking position is not a reliable predictor - 88% of Google AI Mode citations are not in the organic top 10, according to Moz's analysis of 40,000 queries. The highest-leverage action for improving AI citation rates is earning media in trusted publications, which Muck Rack's AI citation research confirms as the dominant citation source type—accounting for over 80% of all AI-cited links across ChatGPT, Gemini, and Claude.

What are the AI citation factors that brands can actually control?

Brands can directly influence all three citation factors. Earned authority is built through editorial relationships and media placements in tier-1 publications, the work of a structured PR and earned media program. Entity clarity is built through consistent cross-platform naming, structured data markup, and cross-source category association, the work of brand governance and technical setup. Citation architecture is built through answer-first content structure, data density (12+ named statistics per long-form piece), FAQ sections, and comparison tables, the work of content strategy and editing. The Princeton/Georgia Tech GEO study found that structural improvements alone can increase AI visibility by 30–40%, but the research is consistent that earned authority is the factor with the highest ceiling.

Who coined Machine Relations?

Jaxon Parrott, founder of AuthorityTech, coined the term Machine Relations in 2024 to name the discipline of earning AI citations and recommendations for brands. He published the origin story and the five-layer MR stack at machinerelations.ai. Machine Relations extends Public Relations into a world where machines mediate how brands are surfaced, compared, and recommended, creating the machine as the primary reader of content that determines brand visibility and citation.

Is Machine Relations the same as SEO or GEO?

No. SEO optimizes for ranking algorithms, the goal is a top 10 position on a search results page. GEO (Generative Engine Optimization) optimizes content formatting for generative AI engines. Machine Relations addresses the full system: earned authority in trusted publications (Layer 1), entity clarity across the machine-readable graph (Layer 2), citation architecture in content (Layer 3), distribution across answer surfaces including GEO and AEO (Layer 4), and measurement via Share of Citation and Entity Resolution rate (Layer 5). GEO is a real and useful discipline, it is Layer 4 of the Machine Relations stack. The 6.82% overlap between ChatGPT citations and Google's top 10 results quantifies the gap between what SEO delivers and what AI citation requires.

Where do GEO and AEO fit inside Machine Relations?

GEO (Generative Engine Optimization) and AEO (Answer Engine Optimization) are distribution tactics within Layer 4 of the five-layer Machine Relations stack. GEO optimizes content formatting for generative AI engines; AEO optimizes content structure for answer boxes and featured snippets. Both tactics produce significantly better results when Layers 1–3 are already in place. A brand doing GEO without earned authority is optimizing the extractability of content that lacks the trust foundation needed for AI engines to cite it with confidence.

How is Machine Relations different from digital PR?

Digital PR optimizes for human journalists and editors, the success condition is a media placement that reaches human readers. Machine Relations uses the same mechanism (earned media in trusted publications) but structures content and entity signals so that machine readers can also parse, extract, and cite what has been earned. Digital PR was designed for human audiences. Machine Relations extends it to work for both audiences simultaneously, because AI engines are now doing the first cut of research that precedes human buyer decisions. The PR mechanism (earned media in respected publications) is correct. What Machine Relations adds is the entity and architecture layers that make that earned media produce AI citation, not just human readership.

What percentage of ChatGPT citations come from earned media?

Muck Rack's analysis of over one million links cited by ChatGPT found that 89% of citations originated from earned media, not owned blog content, paid placements, or social posts. This aligns with Ahrefs' analysis of ChatGPT's most-cited pages, which found that 65.3% of cited content comes from domains with a Domain Rating of 81 or higher—the threshold that corresponds almost exclusively to major editorial publications. These figures collectively establish that AI citation is overwhelmingly an earned media phenomenon, not a content marketing one.

Related Reading