AI answer engines do not cite the most popular websites. They cite sources that match the role the answer requires — market databases for product queries, analyst research for strategy questions, wire services for event-driven responses. Machine Relations Index data across 6,020 measured domains, 17,540 source events, and six engines shows that source type predicts citation rates more reliably than domain authority alone.
Why Source Type Matters More Than Domain Popularity #
Most AI citation studies rank domains by total citation count. LLM Pulse analyzed 5.3 million AI citations and found only 3.26% went to news and media outlets. SurfacedBy examined 127,198 citations across five engines and found 90.6% pointed to vendor, product, and review sites.
These numbers are accurate but incomplete. They describe where citations go without explaining why. The Machine Relations Index segments citations by source role — the functional category a domain serves in an AI-generated answer — and this segmentation reveals patterns that flat domain counts obscure.
A market database like G2 gets cited for a fundamentally different reason than an analyst publication like Gartner. Both may appear in the same answer, but the engine selects each for a distinct evidential function: structured comparison data from one, strategic interpretation from the other. Understanding these roles is the difference between measuring citation volume and understanding citation mechanics.
The Source Roles AI Engines Select From #
The Machine Relations Index classifies measured domains into source roles based on the function they serve when cited. Three roles dominate B2B and enterprise answer-engine citations:
Market and company databases — Platforms that maintain structured, queryable data about companies, products, pricing, and user reviews. G2, Crunchbase, Grand View Research, Fortune Business Insights, MarketsAndMarkets, and Mordor Intelligence all fall into this category. These sources provide the comparison-ready, extractable data that AI engines need for product and market queries.
Analyst and consulting research — Organizations producing proprietary analysis, forecasts, and frameworks. Gartner, Forbes (for its contributed analyst content), and Deloitte represent this role. AI engines cite these sources for strategic and evaluative questions where interpretation matters more than raw data.
Wire and press-release distribution — Services that aggregate and distribute official company announcements. PR Newswire is the primary example in this role. AI engines cite these sources for event-driven, time-stamped facts — funding rounds, product launches, leadership changes.
Market Databases: The Dominant Source Role #
Six of the ten most-cited domains in the MRI's enterprise query measurement window are market databases. This is not coincidence. Market databases offer what AI engines need most for purchase-intent and comparison queries: structured, multi-vendor data that can be extracted without editorializing.
Here is the per-engine citation breakdown for the top market databases measured across 17,540 source events:
| Source | Perplexity | ChatGPT | Gemini | Claude | Google AI Mode | AI Overviews | Total |
|---|---|---|---|---|---|---|---|
| G2 | 30 | 9 | 52 | 9 | 29 | 16 | 145 |
| Grand View Research | 18 | 17 | 15 | 5 | 19 | 10 | 84 |
| Crunchbase | 15 | 3 | 16 | 20 | 10 | 17 | 81 |
| Fortune Business Insights | 14 | 1 | 6 | 6 | 16 | 13 | 56 |
| MarketsAndMarkets | 12 | 0 | 12 | 4 | 13 | 8 | 49 |
| Mordor Intelligence | 9 | 1 | 12 | 11 | 8 | 5 | 46 |
The pattern is clear: market databases collectively account for the majority of enterprise-query citations across all six measured engines. But the per-engine distribution is far from uniform.
Analyst Research: The Strategic Citation Layer #
Analyst and consulting research sources serve a different function in AI answers. They are cited when the query requires interpretation, evaluation, or a framework rather than raw product data.
| Source | Perplexity | ChatGPT | Gemini | Claude | Google AI Mode | AI Overviews | Total |
|---|---|---|---|---|---|---|---|
| Gartner | 0 | 16 | 54 | 10 | 28 | 22 | 130 |
| Forbes | 5 | 9 | 29 | 2 | 10 | 10 | 65 |
| Deloitte | 16 | 7 | 11 | 4 | 5 | 7 | 50 |
Gartner is the second most-cited domain in the measured universe (130 total citations), yet Perplexity does not cite it at all. This is the sharpest single-engine blind spot in the dataset and illustrates why multi-engine measurement matters. A brand that optimizes solely for Perplexity visibility would miss the largest analyst citation source across the other five engines.
Wire Services: The Event-Driven Citation Role #
PR Newswire represents the wire and press-release distribution role in the measured data:
| Source | Perplexity | ChatGPT | Gemini | Claude | Google AI Mode | AI Overviews | Total |
|---|---|---|---|---|---|---|---|
| PR Newswire | 1 | 16 | 6 | 2 | 4 | 6 | 35 |
ChatGPT accounts for 45.7% of PR Newswire's total citations — a concentration far higher than any other engine shows for any other source. This suggests ChatGPT's retrieval model gives disproportionate weight to official announcement content, while other engines prefer to cite the downstream analysis rather than the original release.
Per-Engine Citation Preferences Across Source Roles #
Each AI engine exhibits distinct preferences for source types. These preferences are consistent across the measurement window and reflect differences in retrieval architecture, not random variation.
Gemini cites the highest volume per answer (11.0 sources per response on average, per SurfacedBy's measurement) and leans heavily toward structured databases. It is the dominant engine for both G2 (52 citations, 35.9% of G2's total) and Gartner (54 citations, 41.5% of Gartner's total). Gemini's retrieval favors sources that provide tabular, comparison-ready data.
Perplexity cites broadly across market databases but has a complete blind spot for Gartner (zero citations). It is the only engine that does not cite the second-most-cited domain in the measured universe. Perplexity also shows the strongest preference for Deloitte among the analyst sources (16 citations, 32% of Deloitte's total), suggesting it selects consulting content over analyst-framework content.
ChatGPT cites the fewest sources per answer (3.7 on average) and concentrates on established editorial brands and official announcements. It is the dominant engine for PR Newswire (16 citations, 45.7% of total) and the only engine to cite MarketsAndMarkets zero times while citing Grand View Research 17 times — two market databases in the same role that receive opposite treatment.
Claude shows the strongest preference for Crunchbase (20 citations, 24.7% of Crunchbase's total), a pattern no other engine replicates at that concentration. Claude cites the lowest volume overall (1.34 sources per answer per Foglift's measurement) but selects sources with higher specificity.
Google AI Mode distributes citations more evenly across market databases than any other engine. Its citation counts for Fortune Business Insights (16), MarketsAndMarkets (13), and Grand View Research (19) are the most balanced distribution in the dataset.
Google AI Overviews shows moderate concentration across all source roles, with a slight preference for Gartner (22 citations) and Crunchbase (17 citations). LLM Pulse found that AI Overviews relies least on journalism at 2.75% of citations, consistent with its preference for database and analyst sources over editorial content.
Cross-Engine Agreement Is Lower Than Expected #
The source-role segmentation reveals why cross-engine citation overlap is so low. Research across 412 queries and four engines found only 12% all-engine overlap, with 41% of citations unique to a single engine. SurfacedBy's five-engine analysis found 69.6% of cited domains appeared in only one engine.
Foglift's production-monitoring study of 1,373 AI answers added a critical nuance: engines agree on which brands to mention (85.5%–98.4% brand-mention agreement) but draw from different source domains to make those mentions. The ChatGPT-Claude citation-domain Jaccard overlap was just 0.027 — near zero — while Gemini and Google AI Overview reached 0.643.
This means a brand can be mentioned by every engine but cited through entirely different source domains in each. Visibility in one engine is not transferable to another. Source-role strategy must be engine-aware.
What This Means for Brands Measuring AI Visibility #
The practical implication of source-role segmentation is that brands need to measure citation rates per engine and per source role, not aggregate citation counts.
A brand that appears frequently on G2 has strong Gemini visibility for product-comparison queries but may have minimal ChatGPT visibility if ChatGPT selects from different source types for the same question. Conversely, a brand with strong Gartner coverage has broad visibility across five engines but is invisible to Perplexity for that analyst-sourced content.
The Machine Relations Index measures this at the source-role level across six engines, providing the segmentation needed to identify these gaps. Source-role citation rates — not total citation counts — are the actionable metric for brands trying to understand where they are visible and where they are not.
Source Type Ranking: Citation Rates by Role #
Based on the MRI measurement across 6,020 domains and six engines, source types rank by aggregate citation concentration as follows:
| Source Role | Domains in Top 10 | Total Citations | Share of Top-10 Citations | Primary Engine |
|---|---|---|---|---|
| Market and company databases | 6 | 461 | 53.2% | Gemini |
| Analyst and consulting research | 3 | 245 | 28.3% | Gemini |
| Wire and press-release distribution | 1 | 35 | 4.0% | ChatGPT |
Market databases hold six of the ten most-cited positions and over half the total citation volume. Analyst research holds three positions but captures nearly a third of citations, driven by Gartner's outsized citation rate. Wire services occupy one position with the smallest share, reflecting their narrow functional role.
The Machine Relations Measurement Approach #
Machine Relations measures source-level citation rates segmented by source role, buyer question type, and AI engine. This three-dimensional measurement reveals the mechanical basis for citation selection: AI engines do not choose "the best source" in the abstract. They choose the source that fits the evidential role the answer requires.
The MRI publishes these rates with confidence tiers (A, B, C, or collecting) and relative-signal ranks so that each domain's position is grounded in the volume of evidence behind it, not smoothed by composite scores. A domain with high citation rates on thin evidence carries a lower confidence grade than one with the same rates on deep evidence.
This approach separates signal from noise in a market where competing studies report citation counts without segmenting by source role, engine, or query type — producing numbers that describe the landscape without explaining the mechanism.
How to Use Source-Role Data #
For brands and operators working to improve AI visibility, source-role analysis produces three actionable outputs:
Identify role gaps. If your category is heavily cited through market databases but your brand is absent from G2 or Crunchbase, the gap is not "more content" — it is presence on the specific source type that engines select for your query class.
Measure per-engine exposure. A single-engine win is not cross-engine visibility. Perplexity's Gartner blind spot and ChatGPT's PR Newswire concentration mean that strategies optimized for one engine can fail silently on others.
Prioritize the source role that matches your query type. Product-comparison queries pull from market databases. Strategy queries pull from analyst research. Event-driven queries pull from wire services. Matching your presence to the source role your buyers' questions trigger is more effective than pursuing generic domain authority.
FAQ #
What source types do AI search engines cite most? #
Market and company databases (G2, Crunchbase, Grand View Research) account for the largest share of AI engine citations for B2B and enterprise queries. Analyst research (Gartner, Forbes, Deloitte) is the second most-cited source type. Wire services occupy a smaller but functionally distinct role for event-driven answers.
Why does Perplexity not cite Gartner? #
Perplexity shows zero citations for Gartner in the MRI measurement window despite Gartner being the second most-cited domain across the other five engines. The likely explanation is Perplexity's retrieval model, which emphasizes recency and open-web accessibility over paywalled analyst content. This makes Gartner-sourced information invisible to Perplexity users.
Do all AI engines cite the same sources? #
No. Research shows only 12% of citations are shared across all measured engines, and 69.6% of cited domains appear in only one engine. Engines agree on which brands to mention but cite different source domains to support those mentions.
How can brands identify which source types affect their AI visibility? #
The Machine Relations Index segments citation rates by source role, engine, and query type. Brands should compare their presence across the source types that engines select for their category's buyer questions — not just measure whether they appear in any AI answer at all.
What is the difference between citation count and citation rate? #
Citation count is the raw number of times a domain appears across AI answers. Citation rate is the proportion of measured answer runs in which a domain is cited for a specific query segment. Citation rates, segmented by source role and engine, are more actionable because they account for query volume and engine-specific selection behavior rather than conflating all citations into a single number.