Metric: Structure Score

How it's calculated

Structure Score is derived from a raw structural density ratio:

RawStructure = (li_count + tr_count + th_count) / word_count × 100

This ratio measures how many structural elements (list items, table rows, table headers) exist relative to the word count. It's then normalised via a square-root curve with diminishing returns:

StructureScore = sqrt(clamp(RawStructure, 0, 30) / 30) × 100

The sqrt curve means early structural elements have a large impact, but adding more has diminishing returns. The cap at 30% prevents gaming.

Calibration

• 0–2% raw: Poor — wall of text with no semantic structure.
• 2–8% raw: Good — healthy use of lists and tables.
• 8%+ raw: Excellent — highly structured, ideal for AI extraction.

Pages with >500 words and any list items get a minimum score of 5 (to avoid false zeros on long-form content).

What counts as structure

• List items (<li>) — bullet points, numbered lists, feature lists
• Table rows (<tr>) — comparison tables, pricing tables, data tables
• Table headers (<th>) — column headings that label table data

Note: headings (h1–h6) are evaluated separately in the Heading Structure dimension of the Answer Format tool.

Why it matters

AI systems extract information hierarchically. A structured page with clear lists and tables is much easier for RAG systems to chunk, index, and retrieve than a wall of prose. Structure also helps AI generate accurate citations and featured snippets.