Public Methodology
Fact density is GEO's measure of how many discrete, verifiable attribute-value pairs a page exposes in a form an LLM can extract without inference. The production analyzer lives in packages/sfe/src/geo_sfe/analyzers/fact_density.py.
Absolute count: total extracted attributes on the page.
Attributes per 1,000 words: normalized density so short pages and long pages can be compared.
Category coverage: diversity across facts such as pricing, integrations, compliance, founding year, language support, customer count, and other verifiable dimensions.
The structural score rubric consumes this signal in packages/sfe/src/geo_sfe/structured_data.py.
Erlin's 2026 industry dataset, summarized in AI Brand Visibility Tracking: A Complete Guide (2026), reports that brands with 9+ structured, extractable facts average 78% AI coverage versus 9% for brands with 0-2 facts. That is directionally consistent with the product goal of fact densification, but it is industry benchmark data rather than a peer-reviewed causal estimate.
When fact density is low, RAID prioritizes fact_densification, attribute_faq_insertion, comparison_table_insertion, and, when entity clarity is weak, entity_anchor_addition. The aim is to move ambiguous copy toward extractable claims, not to make the page longer for its own sake.