AEO relevance engineering playbook

Treat answer-engine optimization as an information-retrieval engineering discipline, not content production. The unit of citation is the passage, not the page. Each pass emits four artifacts side-by-side so the doc ships AI-ready, not just human-ready.

Source synthesis: Mike King (iPullRank, "relevance engineering" frame), Aleyda Solis (per-vertical/per-country shape; three-layer measurement, Measure AI search on three layers: Presence, Readiness, Business Impact), Lily Ray (YouTube + LinkedIn = highest-cited AI Overview sources; A single seeded fake claim can self-confirm in AI Overviews), Amanda Natividad (precision over breadth), Kevin Indig (citation rate vs. mention rate gap, Citation rate and mention rate are different metrics; comparative content closes the gap).

When to use

Per-vertical AEO/GEO pass on a single source artifact (vertical playbook, battle card, pricing page, integration page, help doc). Run before publishing the page; rerun whenever schema.org vocabulary shifts or competitor citation profiles change.

Inputs

• One target artifact path
• Vertical anchor (real-estate / solar / home-services / finserv / dental / horizontal, per-vertical shape varies; one motion does not fit all)
• ICP anchor

Process

1. Read substrate. Positioning, ICP-anchor card, brand-voice, product-knowledge for any product entities in the target.
2. Extract canonical entities. Product names, integrations, competitors, ICP role titles, pricing tiers, capability claims. Output an entities.json with type tags compatible with schema.org Product / Service / Organization.
3. Produce 8–12 stand-alone passages, 40–90 words each. Each answers ONE buyer question. Each is self-contained, LLMs cite passages, not pages (Mike King's frame). Single buyer question per passage.
4. Generate FAQPage + HowTo schema blocks for the passages where the question/answer shape fits. Validate against the current schema.org spec.
5. Map each passage to a YouTube + LinkedIn long-form mirror. Lily Ray: those two surfaces are the highest-cited AI Overview sources. Output a mirror plan with title, opening hook, source passage.
6. Score on Aleyda's three layers (Measure AI search on three layers: Presence, Readiness, Business Impact):

- Presence: 10–15 buyer-prompt coverage list, recommendation-rate target, citation-rate target.

- Readiness: 10-trait checklist (accessible, transactable, navigable, factual, …) with current state per trait.

- Business Impact: keep three confidence layers separate, observed referrals / modeled lift / attributed deals. Never combine.

Outputs

• entities.json, typed entity list.
• passages.md, 8–12 cite-ready passages.
• schema.json, FAQPage + HowTo blocks.
• mirror-plan.md, YouTube + LinkedIn mirror plan per passage.
• aeo-scorecard.md, three-layer scorecard.

Quality gates

• Substrate-cited. Every entity and claim cites a substrate path.
• Passage discipline. 40–90 words. Self-contained. One buyer question per passage.
• Schema validity. Validates against current schema.org spec.
• Voice-enforce. Passages obey brand-voice rules.
• No fabricated entities. Entity must be present in product-knowledge or competitive-data-bank.
• Vertical declared. Per-vertical shape varies; one motion does not fit all.
• Three layers separate. Never bleed modeled lift into observed referrals (Aleyda's failure-mode warning). Never combine into a single AEO number.

Human checkpoints

• Canonical-entity owner: signs off on which entities and capability claims are canonical enough to expose to LLM crawlers.
• Content + distribution owner: approves the YouTube/LinkedIn mirror plan; decides which docs cross over to long-form.
• Positioning + measurement owner: sets the Presence prompt set per vertical; decides which Business Impact layer rolls into the growth narrative.

Refusal patterns

• Refuses if the brief / substrate index is missing or stale.
• Refuses if target doc has no frontmatter or no substrate cite.
• Refuses if vertical anchor missing.

Common failure modes

• Optimizing whole pages instead of passages. Passage is the unit of citation.
• Single AEO score that mixes Presence, Readiness, and Business Impact.
• Modeled lift presented as observed lift.
• Schema generated without validation.
• Skipping YouTube + LinkedIn mirror plan; web pages alone underperform on AI Overview citations.
• Ignoring per-vertical shape, same passages across verticals.
• Mention rate up but citation rate flat (Citation rate and mention rate are different metrics; comparative content closes the gap), comparative content closes the gap.
• Seeded fake claims that self-confirm in AI Overviews (A single seeded fake claim can self-confirm in AI Overviews).