AI search engines do not cite sources randomly. Analysis of 25,316 answer engine events across six platforms and 6,913 domains reveals systematic selection patterns: source role, cross-vertical reach, engine breadth, and temporal consistency predict citation probability far more reliably than domain authority or content recency alone. The data shows that market databases and analyst research firms earn disproportionate citation share — and the reasons are structural, not editorial.
What 25,000 Answer Engine Events Reveal About Source Selection #
The Machine Relations Index tracks citation events across six AI engines: Perplexity, ChatGPT, Gemini, Claude, Google AI Mode, and Google AI Overviews. Over a 30-day measurement window ending June 2026, the index recorded 25,316 citation events spanning 6,913 unique domains.
Of those 6,913 domains, only 354 scored high enough to earn a Machine Relations Index consensus rating — roughly 5% of all cited domains. The remaining 95% appear in citations sporadically, typically tied to a single query or engine.
This concentration pattern aligns with independent research. Zhang et al. (2026) analyzed 21,143 citations across 602 controlled prompts and found that citation influence is not evenly distributed — high-influence pages demonstrate specific structural attributes including extractable definitions, numerical facts, and comparison data. Similarly, Superlines research found that 85% of brand mentions in AI answers originate from third-party pages rather than brand-owned domains.
Source Roles Determine Citation Probability #
The most predictive variable in the MRI dataset is not domain authority or PageRank — it is source role. Domains that serve as structured databases of market information, analyst research, or industry benchmarks earn citation rates that dwarf general publishers.
| Source Role | Top Domain | MRI Consensus | 30-Day Citations | Engines | Verticals |
|---|---|---|---|---|---|
| Market database | G2 | 81.3 (Elite) | 192 | 6/6 | 10 |
| Market database | Crunchbase | 80.6 (Elite) | 147 | 6/6 | 10 |
| Analyst research | Gartner | 77.1 (Elite) | 205 | 5/6 | 10 |
| Analyst research | Deloitte | 77.8 (Elite) | 86 | 6/6 | 9 |
| Market database | Fortune Business Insights | 77.4 (Elite) | 72 | 6/6 | 10 |
Market databases like G2 and Crunchbase earn Elite MRI tier status because they provide structured, comparison-ready data that AI retrieval systems can extract directly. When a user asks "6sense vs Demandbase enterprise ABM platform comparison," G2 provides the structured review data, feature matrices, and pricing comparisons that retrieval pipelines need. Gartner and Deloitte earn analyst research citations for the same structural reason — they publish frameworks, quadrants, and categorized analysis that AI systems can decompose into answer components.
This pattern is consistent with API Serpent's analysis, which identifies information specificity and structural clarity as two of the seven primary citation selection factors. Pages with concrete data, statistics, and well-organized HTML with headings, lists, and tables increase citation likelihood over generic editorial content.
Each Engine Has Different Source Preferences #
The six AI engines in the MRI dataset do not cite the same sources equally. Engine-specific citation distribution reveals structural differences in retrieval architecture.
| Domain | Perplexity | ChatGPT | Gemini | Claude | Google AI Mode | Google AI Overviews |
|---|---|---|---|---|---|---|
| G2 | 46 | 9 | 54 | 22 | 53 | 8 |
| Crunchbase | 29 | 4 | 22 | 37 | 43 | 12 |
| Gartner | 0 | 21 | 72 | 24 | 68 | 20 |
| Deloitte | 27 | 6 | 16 | 12 | 20 | 5 |
Three patterns emerge from this data:
Gartner receives zero Perplexity citations but dominates Gemini and Google AI Mode. This suggests Perplexity's retrieval pipeline either deprioritizes gated content or routes around paywalled analyst sources. Gemini and Google AI Mode, which share Google's index infrastructure, surface Gartner heavily — likely because Gartner's structured reports are well-indexed and semantically dense.
Claude disproportionately favors Crunchbase (37 citations) relative to other engines. Claude's retrieval appears to weight primary-source databases higher than editorial or analyst content. Independent analysis confirms that Claude "prefers primary sources and original research over secondary summaries."
Google AI Overviews cites fewer sources overall but maintains broad domain coverage. AI Overviews generated only 8 G2 citations versus 53 from Google AI Mode, despite sharing the same underlying index. This reflects the architectural difference between AI Mode's conversational depth and AI Overviews' summary-oriented format.
Zhang et al.'s measurement framework distinguishes between citation selection (which sources get chosen) and citation absorption (how much of the source's content influences the generated answer). The MRI data confirms this distinction: Perplexity cites more sources per response (5-10) but with shallower extraction, while ChatGPT cites fewer (2-4) with deeper content absorption per source.
Cross-Vertical Reach Amplifies Citation Frequency #
Domains that earn citations across multiple industry verticals accumulate citation volume faster than single-vertical specialists. In the MRI dataset, Elite-tier domains are cited across 9 to 10 verticals out of 10 measured: cybersecurity, enterprise AI, fintech, healthtech, HR tech, and infrastructure/devtools among them.
G2 earns citations in 10 verticals because its review platform covers software across every enterprise category. When a fintech buyer asks about payment processing platforms and an HR leader asks about talent acquisition software, G2 provides the same structured format — verified user reviews, feature comparisons, and pricing data — adapted to each vertical.
This cross-vertical consistency is a measurable citation amplifier. A domain cited in one vertical might accumulate 20-30 citations monthly. A domain cited in 10 verticals, serving the same structural role in each, compounds to 150-200+ citations. The MRI data shows G2 at 192 total citations and Crunchbase at 147, both achieving this through cross-vertical structural consistency rather than editorial breadth.
The implication for brands seeking AI search visibility is direct: vertical-specific content that serves only one audience is structurally disadvantaged compared to content architectures that provide the same evidence type across multiple verticals.
Temporal Consistency Signals Structural Preference #
Citation frequency alone does not indicate whether a source is structurally favored or temporarily trending. The MRI dataset measures temporal consistency — the number of unique days a domain appears in citation results within the 30-day window.
| Domain | Days Cited (of 30) | Total Citations | Citations Per Active Day |
|---|---|---|---|
| G2 | 26 | 192 | 7.4 |
| Crunchbase | 23 | 147 | 6.4 |
| Gartner | 28 | 205 | 7.3 |
| Deloitte | 22 | 86 | 3.9 |
Gartner appears in citation results on 28 of 30 measured days — the highest temporal consistency in this cohort. This is not because Gartner publishes daily content updates. It indicates that retrieval systems have structurally embedded Gartner as a go-to source for certain query categories. The source is not being rediscovered each day; it is being recalled from a persistent retrieval preference.
BrightEdge research found that 57.1% of sources cited in AI Overviews rank outside Google's traditional top 10 organic results, confirming that AI citation systems operate on different selection logic than organic search. The MRI temporal consistency metric captures this divergence: a domain can rank poorly in traditional search but maintain daily citation presence in AI engines because it satisfies the structural requirements of retrieval-augmented generation.
The Machine Relations Index Measurement Framework #
The patterns described above are measured using the Machine Relations Index (MRI), a consensus scoring methodology that evaluates source authority across AI search engines. The MRI score comprises five weighted components:
- Engine breadth (40 points max): How many of the six measured engines cite the domain
- Query diversity (16.5 points max): The range of distinct queries that trigger citations
- Vertical spread (15 points max): How many industry verticals the domain serves
- Position quality (3.3 points max): Average citation position within AI-generated answers
- Temporal consistency (10 points max): Stability of citation presence across the measurement window
The consensus score — not any single component — determines tier placement. Elite-tier domains achieve consensus scores above 77 with confidence ratings of A or B. The methodology version (v1.1) measures across all six engines; earlier versions tracked fewer platforms.
The MRI does not measure content quality subjectively. It measures observable citation behavior: which domains appear, in response to which queries, across which engines, in which positions, and how consistently. This measurement-first approach distinguishes it from authority scoring systems that rely on proxy metrics like backlinks or domain age.
FAQ #
How many sources does a typical AI search engine cite per response? #
Citation count varies by platform. Perplexity typically cites 5-10 sources per response, Gemini cites 3-5, ChatGPT cites 2-4, and Claude cites 2-3 according to API Serpent's analysis. Google AI Overviews and AI Mode vary based on query complexity.
Why do market databases get cited more than news sites in AI search? #
Market databases like G2 and Crunchbase provide structured, comparison-ready data — review scores, feature matrices, pricing, and categorical rankings — that retrieval systems can extract directly into answer components. News sites provide narrative context but less extractable structured data, which reduces their utility for comparison and decision queries.
What is a good citation rate for AI search visibility? #
Industry benchmarks suggest a 15% citation rate across target keywords represents strong AI search performance, while 25% or higher is exceptional. The MRI measures this at the domain level: Elite-tier domains achieve citation presence across 73-93% of measured days in a 30-day window.
Does ranking well in Google organic search guarantee AI search citations? #
No. Research shows that 57.1% of sources cited in AI Overviews rank outside Google's traditional top 10 organic results. AI citation systems select sources based on structural factors — entity density, extractable evidence, cross-platform corroboration — that differ from traditional ranking signals.
Analysis based on Machine Relations Index data as of June 2026. MRI methodology v1.1 measures citation events across Perplexity, ChatGPT, Gemini, Claude, Google AI Mode, and Google AI Overviews. Total dataset: 25,316 events, 6,913 domains, 354 scored domains. Full methodology.
Additional source context #
- How AI Search Chooses What to Cite — Martech LLC Probe your page · free Answer-engine research12 min read2026-05-29 AI search engines rarely cite the page that ranks first. (How AI Search Chooses What to Cite — Martech LLC (martech.llc), 2026).
- How Gemini Chooses Sources: Google's AI Retrieval Pipeline Explained | The Searchless Journal # How Gemini Chooses Sources: Google's AI Retrieval Pipeline Explained If you want to understand why your brand appears, or disappears, inside Google's AI answers, yo (How Gemini Chooses Sources: Google's AI Retrieval Pipeline Explained | The Searchless Journal (searchless.ai), 2026).
- How AI Answer Engines Choose Sources: The 2026 Authority & Citation Framework ## Introduction Rankings once guaranteed visibility. (How AI Answer Engines Choose Sources: The 2026 Authority & Citation Framework (dubseo.co.uk), 2026).
- How AI Engines Choose Sources to Cite | Attrifast AI Search # How AI Engines Choose Which Sources to Cite: The 2026 Mechanics Vincent RuanFounder, Attrifast · May 26, 2026 · 26 min read The actual retrieval and training mechanics behind AI citations — the two (How AI Engines Choose Sources to Cite | Attrifast (attrifast.com), 2026).
- How AI Answer Engines Choose Sources: The 7 Signals | SolCrys provides external context for how AI search engines choose which sources to cite.
- How LLMs Decide Whom to Cite: 2026 Research Analysis provides external context for how AI search engines choose which sources to cite.
- How AI assistants decide which sources to cite provides external context for how AI search engines choose which sources to cite.
- How AI Search Decides What to Cite, and What It Ignores | Fahlout provides external context for how AI search engines choose which sources to cite.
- ChatGPT Citation Sources Decoded: What Actually Gets Cited in 2026 provides external context for how AI search engines choose which sources to cite.