Sources cited across many different buyer questions maintain citation authority over time. Sources with high citation counts concentrated in one or two categories do not. Across 14,680 domains measured by the Machine Relations Index over 66 days, query diversity — being cited across different subject categories and question types — predicts citation stability, multi-engine presence, and confidence tier more reliably than raw citation volume.
What Query Diversity Measures and Why It Matters #
Most AI citation studies count how often a source gets cited. The Machine Relations Index v2 measures something additional: how many distinct buyer-question segments cite a source at rates that clear the evidence floor of at least 10 observations across at least 7 run dates.
A domain with 8 published strata is cited reliably across 8 different category-question combinations. A domain with 2 published strata gets cited often, but only when buyers ask about one or two subjects. This distinction matters because, as 5W Public Relations documented in their "Who AI Cites Now" report, AI retrieval ranks sources rather than pages — a fundamentally different mechanism from traditional search that rewards broad institutional authority over single-page optimization.
External research confirms the pattern. Ahrefs' study of 75,000 brands found that branded web mentions correlate with AI visibility at 0.664 — roughly 3x the correlation of backlinks (0.218). Content volume (number of site pages) showed almost no relationship (0.194). Brands mentioned across different contexts — blog posts, anchors, video transcripts, industry publications — are more likely to appear in AI responses than brands with deep but narrow link profiles.
The rotation problem compounds the diversity gap. SurfacedBy's analysis of 127,198 citations found that 69.6% of all cited domains appeared in only one engine's answers. Just 2.7% of domains — 309 out of 11,647 — were cited by all five engines they tracked. A source that earns citations across many buyer questions is more likely to survive engine-level rotation than one that dominates a single niche.
The MRI Data: 21x Citation Rate Gap Between High and Low Diversity #
The Machine Relations Index v2 tracks citation rates across source types for 14,680 domains, 6 AI engines (ChatGPT, Claude, Gemini, Google AI Mode, Google AI Overviews, and Perplexity), and 8 subject categories over a 66-day observation window ending July 17, 2026. Every domain is measured in every segment it appears in. Rates are published only after a segment clears the evidence floor.
Grouping domains by how many segments have published citation rates:
| Diversity Tier | Domains | Avg Citation Rate | Avg Temporal Consistency | Avg Engines | A-Tier Confidence |
|---|---|---|---|---|---|
| Medium-high (5-8 published strata) | 144 | 1.05% | 0.366 | 4.8 | 3.5% |
| Low (1-4 published strata) | 9,838 | 0.05% | 0.043 | 1.8 | 0% |
| None (still collecting) | 4,698 | 0.02% | 0.018 | 1.3 | 0% |
Sources in the medium-high diversity tier maintain citation rates 21 times higher than low-diversity sources. Their temporal consistency — the fraction of observation days on which they are cited — is 8.5 times higher. Every domain with A-tier confidence falls in the medium-high group. Not a single low-diversity domain earned A-tier confidence in this dataset.
High Volume Does Not Equal High Authority #
Some domains earn hundreds of citations but remain concentrated in one or two categories. The MRI data shows this concentration creates fragility:
| Domain | Citations | Published Strata | Temporal Consistency | Confidence |
|---|---|---|---|---|
| Gartner | 376 | 8 | 0.894 | A |
| 686 | 8 | 0.909 | A | |
| Medium | 548 | 8 | 0.970 | A |
| Palo Alto Networks | 144 | 2 | 0.485 | B |
| SentinelOne | 141 | 2 | 0.576 | B |
| Databricks | 116 | 3 | 0.591 | B |
Palo Alto Networks and SentinelOne each earn over 140 citations — more than many domains in the top 50. But their citations come almost entirely from cybersecurity questions. Their temporal consistency hovers near 0.5, meaning they appear in citation lists on roughly half the days they are measured. Gartner, cited across all 8 categories, appears on 89% of measured days.
The pattern holds at the engine level. TryProfound's analysis of 680 million citations across ChatGPT, Google AI Overviews, and Perplexity shows that Wikipedia earns 7.8% of ChatGPT citations but only 0.6% of Google AI Overviews citations. Reddit takes 6.6% of Perplexity citations but just 1.8% of ChatGPT. Presenc AI's study of news publisher citation share found a similar structural effect: publishers with AI licensing agreements earned measurably higher citation share (Reuters at 11.4%, Associated Press at 9.8%) compared to unlicensed outlets — demonstrating that structural access advantages, like query breadth, compound into durable citation positions.
Why Engines Reward Breadth Over Depth #
AI engines do not share a single retrieval index. Each one builds its own model of which sources answer which types of questions. GrackerAI's analysis of citation patterns documents the mechanism: ChatGPT Search pulls from a third-party search provider plus licensed media partners, Google AI Overviews runs on Google's core ranking systems, Claude grounds answers in documents provided to it, and Perplexity runs a retrieval-and-rerank pipeline. Each mechanism applies different trust signals.
LLM Pulse's study of 5.3 million AI citations from 470,380 answers confirms the divergence at scale: each engine has its own preferred outlets, and only 3.26% of all citations point to news and media publishers. The other 96.7% goes to Reddit, YouTube, product pages, review platforms, and company sites. Cross-engine citation agreement on which sources to cite for the same question remains low.
When a source demonstrates expertise across multiple categories, it signals something different to each engine's retrieval model than a source that dominates one category. A domain cited for cybersecurity, fintech, and HR technology questions provides evidence of broad analytical authority. A domain cited only for cybersecurity, regardless of citation count, provides evidence of narrow domain expertise. WhyIQ's AI Citability Playbook identifies statistical density and answer clarity as key citability drivers — but notes that brand mentions across the web are the strongest single predictor of AI citation, with a 0.664 correlation across 75,000 brands, roughly 3x the correlation of backlinks.
This distinction maps to how engines build retrieval trust. Conductor's research on how AI search engines choose sources documents that each platform weighs freshness, topical authority, and source type differently. Google's own AI optimization guide reinforces this by treating source reputation and topical authority as prerequisites for appearing in generative AI features. A source that clears the bar across many categories has effectively passed multiple independent authority checks — which is why earned media placements across different publications compound citation authority faster than content optimization within a single site.
What This Means for Brands Measuring AI Visibility #
The practical implication is that citation count alone is a misleading metric. A brand that earns 200 AI citations in one product category may appear dominant — until it checks the next quarter's numbers and finds its citations dropped 50% because one engine rotated its preferred sources for that category.
Three measurement shifts follow from the data:
Track category breadth, not just total count. Count how many distinct question types and subject areas cite your domain. A brand cited in 6 categories at moderate rates has more durable visibility than one cited 500 times in a single category.
Monitor temporal consistency. Citation count is a snapshot. Temporal consistency measures what fraction of days your domain maintains its citation presence. The MRI data shows high-diversity domains sustain 0.37 temporal consistency on average versus 0.04 for low-diversity domains — a gap that compounds over quarterly measurement cycles.
Watch for engine concentration risk. If 80% of your citations come from one engine, your AI visibility depends on that engine's retrieval preferences. SurfacedBy's data confirms the risk: nearly 70% of cited domains appear in only one engine's results. Diversified citation profiles survive engine-level changes better.
How Machine Relations Measures Query Diversity #
The Machine Relations Index v2 assigns each domain a confidence tier (A, B, C, or collecting) based on the volume and consistency of evidence behind its citation rates. The evidence floor requires at least 10 observations across at least 7 distinct run dates before a segment publishes a rate rather than showing "collecting."
Query diversity is embedded in the measurement architecture. A domain that clears the evidence floor in 8 segments has demonstrated repeatable citation authority across 8 buyer-question combinations. A domain that clears it in 2 segments has demonstrated authority in 2. The confidence tier reflects this breadth: every A-tier domain in the current dataset has published rates in 7 or 8 segments. No domain with fewer than 5 published segments has earned A-tier confidence.
The public dataset reports citation rates, rankings, evidence counts, and confidence tiers for 14,680 domains across 6 engines. It does not publish internal query identifiers or raw cited URLs.
FAQ #
Does query diversity matter more than citation rate? #
They measure different things. Citation rate is how often a domain gets cited within a segment. Query diversity is how many segments cite it at all. The MRI data shows that domains with high diversity maintain higher rates across time — temporal consistency for diverse sources averages 0.366 versus 0.043 for concentrated sources.
How many categories does a domain need to reach A-tier confidence? #
In the current MRI v2 dataset, every A-tier domain has published citation rates in at least 7 of 8 measured segments. No domain with published rates in fewer than 5 segments holds A-tier confidence. The threshold is not fixed — it reflects how much evidence exists behind the domain's rates.
Can a domain with high citations in one category still be authoritative? #
High citations in one category demonstrate narrow expertise. Engines will cite that domain for questions in that category. But the MRI data shows these domains have lower temporal consistency (0.485-0.576 for 2-strata domains with 140+ citations) compared to broadly cited domains (0.894-0.970 for 8-strata domains). Narrow authority is real but fragile.
What is the evidence floor in the Machine Relations Index? #
A segment publishes a citation rate only after accumulating at least 10 observed runs across at least 7 distinct dates. Below this floor, the segment shows "collecting" rather than a settled rate. The floor prevents thin data from producing misleading citation rates.
Analysis based on Machine Relations Index v2 data: 14,680 domains, 6 AI engines, 66-day observation window through July 17, 2026. Methodology: MRI Score. Last updated: July 18, 2026.