Retrieval systems don't read pages the way people do. They chunk them, embed them, compare the pieces against a question, and quote the fragment that matches best. A page can be excellent for a human and nearly opaque to that process — buried answers, drifting sections, headings that promise one thing over paragraphs about another.
A relevance designer works on single pages the way a relevance architect works on the whole library: making each page's subject, claims, and answers unambiguous to the machines deciding what gets surfaced and synthesized.
- Relevance design is page-levelone page, made unambiguous to retrieval — the architect handles how pages relate.
- Machines read in chunksevery section needs to survive being quoted alone, stripped of surrounding context.
- Answers come firstdirect claims near the top and under matching headings are what extraction rewards.
- This is not keyword stuffingthe work is disambiguation and structure, and it usually improves the page for humans too.
What the job actually consists of
- 01Answer placement
The direct answer within the first lines of its section, elaboration after — extraction reads top-down.
- 02Heading honesty
Each heading an actual question or claim its section fully delivers, because chunk boundaries follow structure.
- 03Self-contained sections
Passages that make sense in isolation, since that is how they will be quoted.
- 04Entity precision
Naming things fully and consistently — pronouns and internal nicknames dissolve when a chunk stands alone.
- 05Claim hygiene
One clear position per section, stated as a sentence a system could safely lift.
Write every section as if it will be read by something that never sees the rest of the page — because increasingly, it will be.
Why the page level needs its own owner
Library-level structure decides which page should win a question; page-level design decides whether that page is quotable once retrieved. Teams that only do the first get surfaced and then paraphrased out of the answer, because the model found the right page and couldn't lift a clean sentence from it.
How the work gets measured
- Presence in AI answers and featured snippets for the page's target questions.
- Whether the system quotes you or merely lists you — quoted framing means the page was liftable.
- Accuracy of AI summaries of the page: ask five assistants what the page says and count the errors.
- Long-tail question coverage a single well-sectioned page picks up.
- The human numbers holding steady or improving — proof the page wasn't sacrificed to the machines.
Relevance designer versus relevance architect?
Scope. The architect arranges the library so the right page owns each question; the designer makes each page parseable and quotable. Same discipline, different altitude.
Is this just on-page SEO renamed?
It inherits from on-page SEO but targets a different reader — extraction and synthesis systems, not ranking algorithms. Passage self-containment and claim hygiene barely mattered to classic SEO and are central here.
Can one person do both design and architecture?
On most teams, yes — they are layers of one job. The split matters on large libraries where page work and map work compete for the same hours.
How these roles fit together
One library, different layers. Content design makes a single page work for the person reading it. Content architecture decides what pages exist and how they relate. Content engineering makes producing them repeatable. Relevance design and relevance architecture make the result legible to the retrieval systems deciding who gets surfaced. Resonance architecture is the layer those all serve: whether any of it lands with an actual human once it is retrieved.
Nobody hires all six. Small teams get one person wearing several of these hats badly until the seams show. The names matter less than noticing which layer is failing, because the fixes live in different places.
