Answer engine optimizer

answer-engine-optimizer

The citation feedback loop. Where seo-performance-tracker measures Google rankings, this optimizes for being the source an LLM quotes — which, for a tool category whose buyers research inside ChatGPT and Perplexity, is where a lot of the demand now decides.

It reuses the suite's spine: read the brand workspace, respect the grounding tier, tag every value, and feed results back into the map and calendar. It layers onto the on-page map — same nodes, hardened — it does not replace it.

Read first: ../../framework/answer-engine-optimization.md (the method + the honesty rules), then ../../framework/macro-micro-semantics.md (the writing tactics it scores).

Preconditions

• entity-profile.json + topical-map.json exist (run seo-brand-foundation / topical-map-builder first).
• Drafts to score live in brands/<slug>/drafts/. With no drafts yet, the skill still produces the hardening spec and the live-answer probe.
• Live-answer probing needs grounding.sources.web_search: true (T1). Without it, do the offline scoring only and say the probe was skipped — do not guess citations.

Workflow

1. Score citation-readiness (T0, offline). For each draft:

   python ../../scripts/aeo_score.py --draft brands/<slug>/drafts/<slug>.md \
     --schema-dir brands/<slug>/data/schema --json
   

   Run it after validate_draft.py is clean — AEO is advisory, fabrication is a gate. Collect score, grade, and the specific fixes (DEF / QA / TLDR / LIFT / BREV / SELF / SCHEMA). Scores are measured (mechanical), the recommended rewrites are asserted.

2. Probe live answer engines (T1, web_search). For the highest-value target queries (core-section, especially comparison/alternative nodes), query them answer-style and record, per query + engine + date: is the brand named? cited with a link? which competitor sources are quoted instead? This is a dated spot check (n=1 per probe), labelled measured — not a rank tracker. Never aggregate it into a visibility %.

3. Write the AEO report → brands/<slug>/audits/<date>-aeo.md:

   • Readiness table — per node: AEO score, grade, top fixes (measured + asserted).
   • Live citations — per probed query: brand cited? competitors cited? (measured, dated, with the query text; honest about the tiny sample).
   • Hardening queue — nodes <70, ranked, with the concrete edits.
   • If web_search is off: state the probe was skipped; emit only the readiness table.

4. Feed the loop.

   • Nodes scoring <70 → mark needs-update in the map; push up calendar.md.
   • Apply hardening to drafts via semantic-draft-writer; ensure JSON-LD via linking-and-schema. Re-score to confirm the lift.
   • Queries where competitors are cited and the brand isn't → a hardening task on the owning node and a signal for off-page authority (link-opportunities).
   • New questions found while probing → query-network additions via topical-map-builder.

Definition of done

• Every existing draft scored; a dated AEO report with the three sections written.
• Hardening queue fed back into map statuses + calendar.
• No invented visibility number anywhere — citations are dated, per-query observations or they are absent. If web_search was off, the report says so.

Grounding ladder

• T0: offline aeo_score.py readiness scoring + hardening spec. Fully useful alone.
• T1 (web_search): + live answer-engine spot checks (measured, dated, per query).
• T2: no paid dependency; SERP-feature data from DataForSEO (if on) can corroborate which queries trigger AI Overviews, labelled measured.