Structure only: rewriting how a page is built lifted citations 17.3% across six engines
Document architecture, information chunking and visual emphasis were the only things changed — the wording stayed identical.
Key numbers
- The study splits content structure into three levels: macro (document architecture), meso (information chunking) and micro (visual emphasis), then measures each against citation probability.
- Across six mainstream generative engines, structural rewriting raised citation rate by 17.3% and subjective quality by 18.5% — with semantic content held constant.
- The authors propose architecture-aware optimization strategies and predictive models, and conclude structure belongs in the foundations of GEO rather than being optional polish.
Method and sample
arXiv paper “Structural Feature Engineering for Generative Engine Optimization” (2026-03-31): the GEO-SFE framework decomposes structure into three levels and evaluates them across six generative engines, reporting relative gains in citation rate and quality.
What it means for brands
Beyond sounding authoritative and carrying data, structure is itself a variable: how headings are nested, how information is chunked and how emphasis is placed all change citation odds — and structure is far cheaper to change than meaning.
How SourceSight handles it
The site audit covers exactly this layer (H1 and heading hierarchy, schema, language declaration, link patterns), and the content factory rebuilds each page as answer-first sections with extractable blocks and an FAQ. Then the same prompts are re-run to compare.
Limits
The gains are relative lifts measured in a research setting, not a guarantee for any given site; engine mixes, prompt distributions and real user queries differ, so results vary by industry.
References
Run it on your own project
What the same prompts return across 10 engines is only knowable by running them.
Related research
Same engine, different surface, different answer: test app, web and API separately
On the same platform and the same questions, app and web cite systematically different source sets — and the API shows something else again.
Methods · 2026-10-07How to design a retest you can trust: five moves so noise does not pass as progress
AI visibility is a distribution, not a point: fix the prompts, repeat the runs, add controls, paraphrase, validate by hand — or you cannot attribute the change.