When an LLM retrieves a passage to cite in an answer, it selects passages where the key fact is clearly identifiable and extractable in a short span of text. Facts that require reading three paragraphs to understand in context are less extractable than facts that are stated clearly in a single sentence with their source. Formatting choices that surface key facts improve extraction rates significantly.
Key Fact Surfacing Techniques
Bold the key claim in each paragraph ("RAG reduces hallucination rates by 60–80%"). Use callout boxes for the most important statistics in the article. Bullet lists for multiple related facts ("Key findings: 78% of newsrooms use AI for research; 45% for fact-checking; 23% for drafting"). Definition boxes for new terms (structured as: [Term]: [Definition in one sentence with citation]). TL;DR sections at the top of long articles provide the most extractable summary for LLMs that scan before retrieving.