Heading format is one of the most underrated LLMO variables. An H2 heading formatted as a question ("Why Does Multi-Engine Fact-Checking Outperform Single-Engine?") signals to LLM retrieval systems the exact query this section answers. An H2 formatted as a vague topic label ("Background") provides no retrieval signal at all.
The Retrieval Signal Mechanism
When an LLM retrieval system searches for evidence to answer a user query, it scores document passages by semantic similarity to the query. Question-format headings explicitly encode a query topic — making the semantic similarity calculation more precise. An H2 that reads "How does RAG reduce hallucinations?" will score higher for the query "how does RAG reduce hallucinations" than an H2 that reads "RAG and Accuracy," even if the body text under each heading is identical.
Rewriting Existing Headings
Converting existing content headings to question format takes 5–10 minutes per article and can significantly improve LLMO performance. Method: identify the implicit question each section answers, then rewrite the heading as that question explicitly. "Introduction" → "What Is AI-Assisted Journalism and Why Does It Matter?" "The Problem" → "What Are the Biggest Risks of AI-Generated News?" Most heading rewrites follow the Who/What/When/Where/Why/How pattern.