An Evidence Density Score measures how much evidence-rich, structurally complete content a document contains — the properties most strongly associated with AI citation probability and answer-level influence in published research. The score examines four dimensions: the presence of statistics, quotations, and citations (Evidence Richness); the density of definitional and comparative sentences (Content Signals); structural indicators like heading and paragraph count (Structural Richness); and Flesch-Kincaid readability as a fluency proxy. A higher score indicates content with stronger structural evidence properties — content that research associates with higher AI citation and absorption rates.

Evidence Richness (Dimension A):

RL-012: Statistics density and quotation presence both contribute 15 points to this dimension. Published research (Aggarwal et al. 2024) found quotations slightly outperformed statistics in citation probability effect (+41% vs. +31%) — the equal weighting in this tool is a design simplification, not a research-calibrated allocation.

RL-013: Citation and reference signals are detected using pattern-matching heuristics (e.g., 'according to', 'Source:'). This is the tool's operational approach — Aggarwal et al. (2024) established the directional effect of source attribution; the specific detection method is a PKA design choice.

Content Signals (Dimension B):

RL-014: Definition sentence detection is based on a page-level finding: Zhang et al. (2026, preprint) found pages with high definitional content showed approximately 57% higher absorption. This dimension applies that finding at document level — this is a directional inference from the research, not a directly measured sentence-level effect. Zhang et al. 2026 is not yet peer-reviewed.

RL-015: Comparative sentence detection is based on a page-level finding: Zhang et al. (2026, preprint) found comparative content was associated with approximately 55% higher absorption. This dimension applies that finding at document level — directional inference, not a measured sentence-level effect. Zhang et al. 2026 is not yet peer-reviewed.

Structural Richness (Dimension C):

RL-016: Heading count thresholds are calibrated from Zhang et al. (2026, preprint), which found high-influence pages had 12.5× more headings than low-influence pages. The specific cutoff values are the tool's internal calibration — not research-specified thresholds.

RL-017: Paragraph count thresholds are calibrated from Zhang et al. (2026, preprint), which found high-influence pages had 5.69× more paragraphs. The specific cutoffs are the tool's internal calibration.

Readability Proxy (Dimension D):

RL-018: The Grade 6–12 target zone is derived from the Flesch-Kincaid readability formula (Kincaid et al. 1975). Aggarwal et al. (2024) found fluency-optimised content associated with approximately +28% citation probability — this tool treats FK Grade 6–12 as a proxy for that fluency standard. The mapping between FK score and Aggarwal's fluency measure is an inference, not a directly tested relationship.