Relevance engineering in AI search: my playbook after Mike King’s chat with Sean

I watched “Chat With Mike King | AI Search, Relevance Engineering & The Future Of SEO” from Search With Sean (https://www.youtube.com/watch?v=DptW_u5qXEw). Worth talking about because it reframes SEO as engineering relevance for answer engines, not just ranking blue links. 🎯

Here’s my filter: what can a team ship in 2–4 weeks that moves AI search, not just organic SERPs. Mike’s framing lines up with how I work this in the field.

What Mike King gets right about AI search

• The game is shifting from documents to answers assembled across entities. If you’re not modeling entities, relationships, and source credibility, you’re background noise. 🤖
• Evidence matters. AI systems weight verifiable, attributed claims higher than marketing prose. Think citations, provenance, and first-party data that can be checked. 🔬
• Schema is table stakes, not a cheat code. Good schemas clarify meaning; they don’t manufacture it.
• Distribution is not just pages—it’s APIs, feeds, and data surfaces models can crawl, call, or quote.

If you want a deeper primer, see my pieces on what is a relevance engineer and what is a relevance architect.

Turn “relevance engineering” into an operating cadence

Here’s the cadence I use when teams ask “what do we do on Monday?” 🛠️

1. 01
   Map entities and claims

   Inventory your core entities (brand, products, people) and the claims you want answer engines to repeat.

2. 02
   Establish evidence

   Back every claim with a source: studies, policies, docs, or first-party metrics that can be cited.

3. 03
   Structure everything

   Add precise schema, stable IDs, and canonical claim pages with section-level anchors.

4. 04
   Expose programmatically

   Ship a lightweight content API or JSON feeds for docs, FAQs, pricing, and policies.

5. 05
   Instrument visibility

   Track where AI answers cite or summarize you; log changes after releases.

I’d phase this over two sprints. Sprint 1: entity/claim map, top-10 claim pages, schema. Sprint 2: API/feeds, citations audit, measurement hooks.

Instrumentation that actually helps

• Use my AI Visibility Checker to spot where models quote you or ignore you across AI surfaces. 🔗
• Run a tight schema diff on every deploy. Small schema regressions break machine understanding disproportionally.
• Maintain a claims ledger: claim -> evidence URL -> last verified date -> owner. No orphaned claims.
• Log where you control first-party proof: policy docs, pricing, support articles, technical specs.
• If your house isn’t in order yet, start with an SEO audit. It’s cheaper than guessing.

Where teams overreach (and how to avoid it)

• “We’ll crank out more content.” Quantity without entity clarity just dilutes authority. Fewer, canonical claim pages beat a hundred tangents.
• “Let’s sprinkle more schema.” Bad schema creates contradictions. Only publish what you can prove.
• “RAG will fix it.” RAG needs a clean knowledge base. If your docs are inconsistent, you’ll just accelerate confusion.
• “We’ll be fine without APIs.” You’re making models scrape and guess. Give them a contract instead.

A simple 14‑day ship list

• Day 1–3: Build entity and claim map. Identify proof sources.
• Day 4–7: Launch or refactor 10 canonical claim pages with evidence, anchors, and precise schema.
• Day 8–11: Stand up a minimal JSON feed for FAQs, policies, specs.
• Day 12–14: Add visibility checks, baseline AI answer snapshots, and a schema regression test. ⏱️

My take

Mike’s right to push beyond “rankings.” The durable edge is operational clarity: clean entities, provable claims, structured delivery, measurable outcomes. That scales across updates and UI changes.

The takeaway

AI search rewards teams who make understanding easy and verification trivial. Model your entities, publish provable claims, structure the page, and ship a feed or API. Then measure ruthlessly. That’s how you protect and grow visibility across organic, local, and AI answers—exactly where modern SEO/GEO/AI practice converges. 🧭