This lesson explains why content written to be read from top to bottom can fail when a machine retrieves one section of it on its own, and how to structure sections so each one survives being lifted out. It draws on Beatrice Gamba’s work in the AI Search Optimisation & Agentic SEO course, with credit, and tests it on this site’s own blog.
What this lesson covers
- The retrieval pipeline and where pages fail in it
- Chunks, and why context inside them matters
- The proposition-first section
- What the research says gets content used
- Four habits that break extraction
- A five-check audit you can run on any page
Retrieval systems split pages into passages and judge each passage on its own. The lesson shows one of the site’s posts the way a retriever sees it, then sets out the evidence that this matters: Anthropic’s contextual retrieval study cut top-20 retrieval failures from 5.7 to 1.9 per cent by giving chunks their missing context, and research presented at EMNLP 2024 found self-contained statements retrieved better than whole passages. The fix is the proposition-first section: a heading phrased as the question it answers, the answer in the first sentence with the subject named in full, one scoped and sourced claim, then what it means.
The lesson then names four long-form habits that undo this — the narrative build-up, the definition spread across a page, the table without context and the paragraph with several claims — and the GEO study’s finding that adding quotations, statistics and sources raised visibility while keyword stuffing scored below the baseline. Finally it runs five checks over the demo site’s blog: generic headings, back-references such as “as mentioned above”, sections opening with a bare pronoun, unsourced statistics and oversized sections. Forty-six of ninety-nine posts fail at least one, and the lesson rewrites one section to show the fix is a section at a time, not a page at a time.
This lesson continues Module 5, Remediation: Fixing Entity Representation, in The AI-powered Entity Audit, and is followed by a practical that automates the five checks. It suits writers, editors and SEOs preparing content for AI search without rewriting everything they have.