Why AI Engines Cite Crunchbase: Source Authority in the Machine Relations Index

Crunchbase.com posted 81 citations across 6 AI engines — down 14.7% from 95 — while its MRI consensus score edged to 79.4, a new high. The nine-cycle trajectory now reads 186, 165, 141, 124, 89, 97, 97, 95, 81: five declines, a recovery, a two-cycle stabilization, a quality-consolidation dip, and now a volume contraction that broke below the 89–97 band that had held for four cycles. This analysis uses the latest Machine Relations Index data to examine what the volume break means for Crunchbase's citation durability, why the composite score kept climbing despite the sharpest single-cycle decline since cycle 4, and how G2's simultaneous recovery reshapes the competitive landscape.

Last updated: July 3, 2026

Crunchbase MRI Profile: 81 Citations Across 6 AI Engines #

The Machine Relations Index measures source citation authority across AI answer engines using a composite methodology (MRI Score v1.1, 6-engine). Crunchbase's current profile shows a source where volume is contracting faster than quality metrics can compensate.

MRI consensus score: 79.4 (Elite tier, B-confidence)

Crunchbase ranks #2 among 308 market databases tracked in the MRI, holding its position at the 99.7th percentile within that source role. Its weighted authority score fell sharply to 60.4 from 67.8 (-10.9%), the largest single-cycle weighted authority decline in its measurement history. The measurement covers 6,085 total domains and 17,860 source events.

The nine-cycle citation trajectory now reads: 186, 165, 141, 124, 89, 97, 97, 95, 81. The 14-citation decline from 95 to 81 (-14.7%) is the steepest single-cycle drop since the 141-to-124 decline in cycle 4 (-12.1%). More significantly, 81 falls below the 89–97 band that had held for four consecutive cycles — the recovery floor (89), the stabilization level (97, 97), and the consolidation dip (95). This is the first time since the initial recovery that Crunchbase has operated outside that band.

The consensus score gained 0.1 points (79.3 to 79.4), extending the consecutive improvement streak to four cycles (78.5 → 78.8 → 79.3 → 79.4). But the magnitude of the gain collapsed: the prior cycle added 0.5 points, this cycle added 0.1. The entire gain came from position quality, which surged from 3.8 to 4.3 as average citation position improved from 3.9 to 3.5 — Crunchbase's best recorded positioning. The improvement mechanism is concentration: with fewer total citations, the remaining ones cluster at higher positions in citation lists, lifting the per-citation quality metric even as the denominator shrinks.

Temporal consistency reversed its prior-cycle surge, falling from 7.7 to 7.3 as days cited dropped from 23 to 22. Query diversity held at 14.3 on the component score, but the underlying query count contracted from 30 to 29 — losing the context added in cycle 8. Vertical spread held flat at 13.5 across 9 verticals. The pattern is no longer quality consolidation on stable volume. It is volume contraction where one metric (position quality) improved enough to keep the composite marginally positive while two others (temporal consistency, query count) pulled back.

Weighted authority's 10.9% decline — from 67.8 to 60.4 — exposes the metric's sensitivity to volume at this scale. This metric amplifies both volume and position quality; the 14-citation volume drop overwhelmed the position improvement. The gap between consensus (up 0.1) and weighted authority (down 7.4 points) is the widest single-cycle divergence in Crunchbase's history, and it signals that the two metrics are now measuring fundamentally different things: consensus captures the quality of the remaining citations, while weighted authority captures the scale at which that quality operates.

The confidence rating holds at B. At 81 citations and 22 days cited, the volume contraction pushes Crunchbase further from the sustained-count threshold that A-confidence requires. The path to A-confidence now requires not just sustained improvement but volume recovery — a materially harder target than the steady-state counting the prior cycle suggested.

Citation Distribution by AI Engine #

The 30-day citation breakdown shows Claude's growth streak breaking while Google AI Mode's collapse accelerated to its steepest single-cycle decline.

Claude dropped from 25 to 20 citations, breaking a three-cycle growth streak (23 → 24 → 25 → 20). The loss of 5 citations in a single cycle is Claude's largest movement in either direction across Crunchbase's measurement history. Claude's share fell from 26.3% to 24.7% of total citations — still the largest single-engine share, but the concentration thesis from cycle 8 (Claude pulling away from the pack) is now tested. The decline may reflect changes in Claude's retrieval stack rather than Crunchbase-specific factors; research on AI citation divergence shows that engine-level retrieval preferences can shift when underlying model updates alter document ranking. Whether this is a one-cycle correction or a trend reversal will require cycle 10 data.

Google AI Mode fell from 16 to 10 citations (-37.5%), its steepest single-cycle decline and a continuation of the post-collapse trajectory: 40, 19, 19, 18, 16, 10. The cumulative decline from peak is now 75%, with no recovery cycle across six measurements. At 10 citations, Google AI Mode has moved from mid-pack to Crunchbase's second-weakest engine. Since Google AI Mode appears as a distinct retrieval surface from AI Overviews within Google search, its accelerating erosion suggests that the AI Mode retrieval stack is either deprioritizing structured market data or shifting toward different source types for the queries where Crunchbase was previously cited.

Google AI Overviews declined modestly from 18 to 17 (-5.6%), its second consecutive decline. Since AI Overviews appears directly in Google search results — the highest-traffic AI surface — its relative stability compared to AI Mode continues to confirm it as the more embedded retrieval path for Crunchbase data. Analysis of Google AI Overviews source selection has shown that pages with 15 or more Knowledge Graph entities exhibit 4.8x higher selection probability, a structural advantage Crunchbase's entity-dense pages retain.

Gemini dropped from 18 to 16 (-11.1%), breaking a three-cycle hold at 18. Perplexity held at 15 for a second consecutive cycle. ChatGPT held at 3 for a ninth consecutive cycle.

The distribution shifted downward across four of six engines. Only Perplexity and ChatGPT held steady — the two engines at opposite ends of the citation spectrum. The spread between the top engine (Claude, 20) and the fifth engine (Google AI Mode, 10) widened to 10 citations, but the cluster structure changed: instead of five engines between 15 and 25, there are now four between 15 and 20 with AI Mode dropping to 10. The mid-pack convergence that characterized prior cycles is fragmenting.

What Makes Crunchbase Citation-Eligible #

Crunchbase's structural properties remain intact through the nine-cycle trajectory. The question cycle 9 answers is whether volume contraction below the stabilization band signals retrieval-level displacement or a temporary contraction in the query environment where Crunchbase operates.

Structured entity data at scale #

Crunchbase maintains structured profiles for over 2 million companies, including funding rounds, acquisitions, leadership, and financial data. Its Entity Lookup APIs expose this data in machine-readable formats with consistent field schemas. In May 2026 alone, Crunchbase added more than 350,000 companies — one of its largest months of company growth — expanding coverage across Europe, North America, and Asia. The platform's Market Insights now cover over 5,100 micro-industries and 13.2 million products, backed by activity signals from more than 80 million users. Its predictive intelligence platform has generated over 455 million predictions as of March 2026.

AI engines retrieving answers to queries like "AI tools for public relations and enterprise communications" or "HR tech Series B funding announcements" find Crunchbase pages that contain exactly the structured, attributable data points the query demands. The SourceBench framework evaluates whether AI answers reference quality web sources and finds that structured, entity-rich pages with clear provenance consistently outperform unstructured narrative sources in retrieval-augmented generation systems.

Crunchbase is also expanding AI-native access paths beyond web retrieval. Its MCP Server for Finance enables AI agents to retrieve predictive intelligence directly through the Model Context Protocol. Glean expanded its financial services MCP ecosystem to include Crunchbase alongside FactSet, S&P Global, and CB Insights — positioning Crunchbase data inside enterprise AI agent workflows. The Snowflake Intelligence partnership delivers predictions within Snowflake environments. These programmatic access paths may sustain Crunchbase's structural retrievability even as web-retrieval citation counts fluctuate.

Query-to-answer alignment across verticals #

Crunchbase's 29 cited queries span 9 verticals: cybersecurity, enterprise AI, fintech, HR tech, infrastructure/devtools, and legal/compliance among them. The query count contracted from 30 to 29, reversing the prior cycle's expansion. Sample queries from the current MRI measurement include:

• "AI tools for public relations and enterprise communications"
• "HR tech Series B and growth-stage funding announcements"
• "HR tech acquisitions by enterprise software companies"
• "HR tech unicorns and high-growth workforce platforms"
• "HR technology market growth and investment activity"

Each query has a clear information need — specific companies, amounts, dates, deal terms — that Crunchbase's data pages answer directly. The Authority Signals Framework, which analyzed 10,038 citations across 542 sources, identifies query-answer alignment as one of the strongest predictors of citation selection. Research from the Omnius AI Search & GEO Report 2026 found that statistics and data points increase AI visibility by 22%, and expert attribution boosts it by 37% — both properties inherent to Crunchbase's structured company profiles. The query contraction from 30 to 29 is the first reversal in the metric's recent trajectory, and at one query it is within measurement noise, but it breaks the expansion pattern that supported the quality-consolidation thesis.

Position quality as the counter-signal #

Crunchbase's average citation position improved from 3.9 to 3.5 — its best ever. The position quality component surged from 3.8 to 4.3, the largest single-component gain this cycle and the only component that improved. When AI engines cite Crunchbase in cycle 9, they place it higher in citation lists than at any point in its measurement history.

The mechanism is worth understanding: as total citations drop, the remaining citations are the ones where Crunchbase's data most directly answers the query — the high-confidence, high-relevance retrievals that engines keep even as lower-confidence retrievals fall away. This is a filtering effect. A source losing marginal citations while retaining core citations will show improving position quality because the citations that survive are the ones where the source is most authoritative. Research on source attribution in retrieval-augmented generation applies Shapley-based attribution methods and finds that documents with structured, non-redundant factual content receive disproportionate attribution weight — exactly the kind of content that persists in Crunchbase's contracting citation set.

The question is whether this filtering stabilizes around a durable core or continues contracting. If position quality keeps improving while volume keeps falling, the source is being cited in an increasingly narrow set of high-confidence contexts. At some point, the remaining citations represent the irreducible core — the queries where no other source can serve the structured data need better. Whether 81 is near that core or still above it is what cycle 10 will reveal.

Source Role: Market Database Competitive Landscape #

Among 308 tracked market databases, the competitive relationship between the top two positions reversed course as G2 recovered sharply while Crunchbase contracted.

The consensus gap between G2 and Crunchbase widened from 0.8 to 1.3 points — the first expansion after three consecutive cycles of compression (2.5 → 1.7 → 0.8 → 1.3). The convergence narrative that defined the prior three cycles is now interrupted. G2's consensus gained 0.6 points (80.1 to 80.7) while Crunchbase's gained 0.1 points (79.3 to 79.4). For the first time in four cycles, both sources moved in the same direction on the composite metric, but G2 moved six times faster.

G2's volume trajectory reversed: 159, 140, 138, 148. After three consecutive declines, G2 recovered 10 citations (+7.2%), its first volume gain since the measurement series began tracking at this frequency. G2's weighted authority surged from 73.5 to 79.1 (+7.6%), partially recovering the erosion that had taken it from 87.2 three cycles earlier. Days cited recovered from 25 to 26. Average citation position improved from 7.8 to 7.7.

The volume ratio between G2 and Crunchbase expanded sharply to 1.83x (148/81), up from 1.45x (138/95) last cycle. This is the widest volume gap in the recent measurement series. In weighted authority, the gap widened from 5.7 points (73.5 vs. 67.8) to 18.7 points (79.1 vs. 60.4) — a dramatic reversal driven by G2's weighted authority recovery and Crunchbase's sharp decline.

G2 maintains and now extends advantages in query diversity (35 vs. 29 queries), temporal consistency (26 vs. 22 days), vertical coverage (10 vs. 9 verticals), and volume (148 vs. 81). G2's confidence rating holds at A while Crunchbase remains at B. Crunchbase retains its position quality advantage: its average citation position of 3.5 remains substantially better than G2's 7.7, meaning each Crunchbase citation carries more per-citation visibility — but with 81 citations versus 148, the total visibility gap is now heavily in G2's favor.

The divergence this cycle is instructive. G2's per-engine breakdown (Gemini 51, Google AI Mode 32, Perplexity 31, Google AI Overviews 16, ChatGPT 9, Claude 9) shows a fundamentally different engine distribution than Crunchbase's. Gemini alone gives G2 more citations (51) than Crunchbase receives from all six engines combined (81). G2's strength in Google AI Mode (32 vs. Crunchbase's 10) is particularly notable given that AI Mode is Crunchbase's fastest-declining engine. The two market databases appear to serve different retrieval niches within the same source role: G2's peer comparison format aligns with product-evaluation queries, while Crunchbase's company data format aligns with market-intelligence queries.

What Operators Can Learn from Crunchbase's Citation Trajectory #

Crunchbase's nine-cycle trajectory — five declines, a recovery, a two-cycle stabilization, a consolidation dip, and now a volume contraction below the stabilization band — is the longest continuously measured arc for any Elite-tier market database in the MRI.

1. Stabilization bands can break. The four-cycle band between 89 and 97 appeared to establish Crunchbase's post-recovery equilibrium. Cycle 9's drop to 81 breaks that band. For operators, this means apparent stability in citation metrics does not guarantee a permanent floor — external retrieval-environment changes can shift the band downward even when the source's structural properties are unchanged. The 350,000 companies Crunchbase added in May 2026 did not prevent the citation decline, confirming that content scale and retrieval selection are distinct dimensions.

2. Position quality can improve during volume contraction — but the mechanism matters. Crunchbase's best-ever position quality (average 3.5) came alongside its steepest recent volume drop. The mechanism — marginal citations falling away while core high-position citations persist — is a filtering effect, not an authority gain. Operators should distinguish between position quality improvements driven by expanding retrieval contexts (healthy) and those driven by contracting ones (diagnostic). In Crunchbase's case, the improving position quality is a signal about the durability of its core citations, not a sign that its overall citation authority is strengthening.

3. Engine-level growth streaks can reverse without warning. Claude's three-cycle growth with Crunchbase (23 → 24 → 25) ended with a five-citation drop to 20. No single-engine trend in AI citation markets is guaranteed to persist, because engine retrieval stacks change with model updates. Operators monitoring single-engine trends should weight cycle-over-cycle direction less than multi-engine structural presence. Crunchbase's continued 6-engine breadth (perfect 40/40) is more durable evidence than any single engine's trajectory.

4. Competitive convergence can reverse. Three cycles of consensus gap compression (2.5 → 1.7 → 0.8) suggested Crunchbase was on track to match or overtake G2. Cycle 9's reopening to 1.3 points, driven by G2's simultaneous recovery and Crunchbase's marginal gain, shows that convergence in citation markets is not monotonic. The competitive position depends on both sources' trajectories, and when one reverses while the other stalls, accumulated convergence can unwind in a single cycle.

5. Weighted authority and consensus can diverge to the point of measuring different realities. Cycle 9 produced the widest single-cycle divergence in Crunchbase's history: consensus up 0.1, weighted authority down 7.4 points. At this scale of divergence, the two metrics are no longer alternative views of the same authority — they are measuring different properties. Consensus says Crunchbase's citation quality is at an all-time high. Weighted authority says its citation influence is declining faster than at any previous point. Both statements are true simultaneously. Operators must choose which metric aligns with their strategic question: quality of remaining citations (consensus) or scale of citation impact (weighted authority).

6. Google AI Mode's erosion may be approaching a structural minimum. The trajectory from 40 to 19 to 19 to 18 to 16 to 10 now shows a 75% cumulative decline from peak with no recovery cycle. At 10 citations, AI Mode is approaching the level where the remaining citations represent only the highest-confidence retrievals. Whether it stabilizes near 10 or continues toward zero will indicate whether AI Mode still has a structural retrieval path for Crunchbase data or has effectively reclassified the source.

How This Connects to Machine Relations #

In the Machine Relations framework, citation authority is not a popularity metric. It is a measurement of whether a source is structurally legible, factually useful, and retrievable by machines making real-time decisions about what to include in an answer.

Crunchbase's nine-cycle trajectory reveals a pattern that only extended measurement can surface: volume contraction below a stabilization band, occurring while the composite quality score still improves, is the phase where citation authority separates from citation scale. The five-cycle decline took Crunchbase from 186 to 89. The recovery brought it to 97. The stabilization held it at 97 twice. The consolidation dipped to 95 while quality improved. And now the contraction drops to 81 while the consensus score reaches its highest point. Each phase reveals a different relationship between volume and quality, and the nine-cycle arc shows all of them in sequence.

The competitive reversal with G2 — consensus gap expanding from 0.8 to 1.3 after three cycles of compression — adds a second dimension. G2's recovery to 148 citations, driven primarily by Gemini (51 citations vs. Crunchbase's 16), reveals that different structured data formats serve different AI retrieval needs. G2's peer-comparison format and Crunchbase's company-data format are not in direct competition for the same retrieval slots; they are complementary source types that happen to share a source-role classification. The implication for practitioners building citation architecture is that competitive analysis within a source role must account for format-specific retrieval niches, not just aggregate scores.

The position quality signal is the most instructive finding of cycle 9. At 3.5 average citation position — meaning AI engines place Crunchbase citations in the top 3–4 slots when they cite it at all — Crunchbase demonstrates that structural data quality sets a position floor that persists even during volume contraction. Research from the Omnius AI Search Report found that 80% of AI-cited sources do not appear in Google's traditional top results, confirming that AI citation selection operates on different criteria than web search ranking. For sources with Crunchbase's structural properties — entity density, machine-readable schemas, consistent field formats — the citation position reflects inherent data quality rather than SEO-driven ranking signals.

The nine-cycle trajectory — 186, 165, 141, 124, 89, 97, 97, 95, 81 — now includes the longest continuous decline for any Elite-tier market database, the first recovery, the first post-recovery stabilization, the first quality consolidation, and the first below-band volume contraction. For operators, the Crunchbase case demonstrates that Elite-tier citation authority can persist through sustained volume loss: the consensus score has improved for four consecutive cycles while volume declined for two. The question cycle 10 must answer is whether 81 citations represents a new, lower stabilization band — or whether the contraction is accelerating toward a level where the remaining citations, however well-positioned, no longer constitute meaningful citation authority at scale.

FAQ #

What is Crunchbase's MRI score? #

Crunchbase.com has a Machine Relations Index consensus score of 79.4, its highest recorded, placing it in the Elite tier with B-confidence. It ranks #2 among 308 market databases tracked in the MRI, with 81 citations across 6 AI engines over a 30-day measurement period. The MRI methodology (v1.1, 6-engine) scores sources on engine breadth, query diversity, vertical spread, position quality, and temporal consistency.

Which AI engines cite Crunchbase most? #

Claude leads with 24.7% of Crunchbase's 30-day citations (20 of 81), followed by Google AI Overviews at 21.0% (17), Gemini at 19.8% (16), Perplexity at 18.5% (15), Google AI Mode at 12.3% (10), and ChatGPT at 3.7% (3). Claude's share declined from 26.3% as its citation count dropped from 25 to 20, breaking a three-cycle growth streak.

Why do AI engines cite market databases more than news sources? #

Market databases like Crunchbase provide structured, entity-attributed, verifiable data — the exact properties AI retrieval systems need to construct factual answers without hallucination risk. News articles contain analysis that AI engines may reference for framing, but when the query demands specific company data, funding amounts, or market figures, structured databases satisfy the information need more directly. Research on AI answer source quality confirms that structured sources with clear provenance outperform narrative sources in retrieval-augmented generation.

How is the Machine Relations Index calculated? #

The MRI (v1.1, 6-engine) measures citation authority across Perplexity, ChatGPT, Gemini, Claude, Google AI Mode, and Google AI Overviews. The consensus score combines five components: engine breadth (how many engines cite the source), query diversity (how many distinct queries trigger citations), vertical spread (industry coverage), position quality (where the source appears in citation lists), and temporal consistency (how many days the source is cited). The index currently tracks 6,085 domains across 17,860 source events. For methodology details, see What is Share of Citation.

Is Crunchbase gaining or losing AI citation authority? #

Crunchbase is in a volume contraction phase with a quality paradox: citation volume dropped 14.7% to 81 (the steepest recent decline and below the four-cycle stabilization band), but the consensus score reached a new high of 79.4 driven by best-ever position quality. Whether this represents a new lower stabilization band or accelerating contraction will be determined by cycle 10. Crunchbase remains in the Elite tier at the 99.7th percentile, but the competitive gap with G2 reopened from 0.8 to 1.3 consensus points as G2 recovered to 148 citations.