Claim
Replace rank-tracking metrics with a three-layer AI search measurement stack — Presence (where the brand appears in answers), Readiness (why it looks that way), and Business Impact (what it produces) — with each layer kept in separate observed/proxy/modeled confidence buckets.
Mechanism
Ranks measured a single observable: the SERP. AI answers are generated from many invisible inputs (chunked passages, embeddings, model memory, citation gates), so a single metric collapses signal. Splitting the stack lets each layer get its own instrument: Presence is observed via prompt sweeps across surfaces; Readiness is diagnosed by inspecting page traits (schema, llms.txt, entity density, paragraph shape); Impact is split into observed referral traffic, proxy signals (branded search lift, direct visits), and modeled estimates (assisted conversions). Keeping confidence buckets separate stops modeled numbers from contaminating observed reporting.
Conditions
Holds when:
- The team accepts that AI search needs a different measurement contract than SERP SEO.
- Operators can run prompt sweeps across at least two AI surfaces (ChatGPT, Claude, Perplexity, Google AI Mode).
- Page-level audit tooling exists (or can be built — the Readiness layer runs offline against saved HTML).
Fails when:
- Stakeholders demand a single AEO number; the framework's value is in the separation.
- The org has no first-party referral analytics — Impact collapses to vanity metrics.
- Modeled estimates leak into observed reporting and the framework loses its anti-overclaim function.
Evidence
"Presence tells you where the brand appears, Readiness tells you why it looks that way, and Business Impact tells you whether that visibility creates measurable value."
"Measured AI referral traffic is the floor, not the ceiling, of AI's contribution."
Companion piece on global AI search:
"Your global competitor set is not your AI search competitor set in each market."
— Aleyda Solis, https://www.aleydasolis.com/en/ai-search/, 2026-04-23
Signals
- Readiness scores rise on owned pages (llms.txt present, JSON-LD blocks valid, paragraphs in chunker-friendly 200–1200 char range).
- Presence sweeps show comparative-win-rate and recommendation-rate trends, not just citation counts.
- Impact reporting has visible confidence labels (observed vs proxy vs modeled) and stakeholders learn to read them.
Counter-evidence
The framework is young and the Presence layer requires either browser automation against AI surfaces (hard to maintain) or paid AI-search analytics tooling (expensive, varying quality). Indig's ghost-citation finding (Citation rate and mention rate are different metrics; comparative content closes the gap) shows even Presence has a hidden split: cited-but-not-mentioned. Operators who measure citations as a single metric get a misleading floor.
Cross-references
- Citation rate and mention rate are different metrics; comparative content closes the gap — Indig's data on the cited-vs-mentioned gap; informs the Presence layer's sub-metrics.
- The unit of optimization is the passage, not the page — Mike King's framework supplies the Readiness diagnostic vocabulary.
- AEO is a GTM capability, not an SEO experiment — Maja Voje's argument for who should own this stack.