Claim
"Relevance Engineering" reframes Generative Engine Optimization as an IR engineering discipline: entity graphs, structured chunking, citation-worthiness. Investment that compounds in AI-native PMM is entity-first, not keyword-first. A single AEO skill should be entity-first by default.
Mechanism
LLM retrieval scores chunks against entities and relations, not keyword density. Pages structured as discrete passages with clean entity references become more retrievable; long undifferentiated copy becomes invisible regardless of word count.
Conditions
Holds when: the brand owns a defined entity graph (products, features, problems, audiences) and can express it in schema.
Fails when: the offering is ambiguous or pre-PMF — entities are still in flux.
Evidence
Paraphrased from Mike King's framing: treat GEO as IR engineering, not content marketing. Entities, structured chunks, citation worth are the levers.
— Mike King, iPullRank, April 2026
Signals
- AEO investment going into schema and entity graphs, not blog volume.
- Passage-level citations outperforming page-level rankings in LLM citation traces.
- Content briefs shaped around entities and questions rather than keyword clusters.
Counter-evidence
For new categories with no shared entity vocabulary yet, content has to do double duty — define the entities first, then engineer relevance around them.
Cross-references
- (none in current corpus)