# Aeo

> Answer Engine Optimization — make a page the extractable answer for featured snippets, People Also Ask, voice, and the short block inside AI Overviews

- Skill: `abouchard11/aeo` (Agent Skill)
- Install (CLI): `npx skillmds@latest add abouchard11/aeo`
- Raw SKILL.md: https://api.skillmd.com/api/skills/abouchard11/aeo/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: abouchard11 (https://skillmd.com/u/abouchard11)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/abouchard11/aeo

---


# AEO — Answer Engine Optimization

**First:** Read `~/.claude/skills/seo-references/core.md`. Apply the methodology and voice to all output. AEO lives on purchase-intent pages. It is not a program for spawning "what is" articles so a snippet has somewhere to live.

**Also read:**
- `~/.claude/skills/seo-references/data-pull-patterns.md`
- `~/.claude/skills/seo-references/sturm-shorts.md`
- `~/.claude/skills/seo-references/fan-out-diagnostic.md` — if `/geo` already ran Step 4b, reuse that table. Do not invent a second branch list.
- Companion research: https://github.com/abouchard11/ai-citation-patterns — passage length, schema, and AIO overlap numbers. Do not dump the research file into the playbook.

Unit of success: the **extracted passage**. Featured snippet, PAA answer, voice response, or the short sentence an AI Overview lifts. GEO (`/geo`) owns brand mentions and citations inside a multi-source synthesis. If the user wants both, run AEO on the page structure and GEO on the prompt matrix.

Route, do not duplicate:
- Preferred Sources frequency badge on AIO / AI Mode / Discover / Top Stories → `/preferred-source` (SHIP / MAYBE / SKIP first)
- Citation share across ChatGPT / Perplexity / Claude → `/geo`
- Fan-out heading on the same money page → `/geo` Step 4b, then the passage spec below
- Links → `/hunter` then `/kilo`
- Indexing → `/indexer`

## Do not do this

- Do not label the page "optimized for AI assistants and LLM search, not human marketing." That line is a confession. Assistants already say "filtering out SEO spam" in the thinking phase. Write the answer a buyer would trust.
- Do not recommend a new informational blog so schema has a home. Put one question + one 40–60 word answer on the money page that already ranks.
- Do not treat `llms.txt` or hidden LLM-only blocks as an AEO tactic. Google has said normal SEO works; Danny Sullivan warned against chunking pages "for LLMs."
- Do not scrape live SERP snippet HTML through Google `/goto` redirects to prove a win. Use GSC appearance data and a CamoFox snapshot of the result the operator is allowed to load like a person.
- Do not add a page per inferred AI Mode branch. Heading-query match on the converting URL is the move. Coverage percentage is not.

## Step 1: Parse Domain

Extract the domain. Strip protocol, www, trailing slash, lowercase. Resolve against the MCP Routing Map. Flag local-service vs national so Map Pack is not sacrificed for a snippet vanity play.

## Step 2: Pull Live Data

**GSC (PRIMARY for AEO):**
1. `mcp__gscServer__get_search_analytics` — last 28 days, dimensions=[query, page], rowLimit=50. Sort by impressions. These queries are the answer inventory.
2. Same call filtered or noted for queries that look like questions (who / what / where / when / how much / near me / vs). Those are snippet and PAA candidates.

**Semrush (ENRICHMENT):**
3. `mcp__semrush__execute_report` — Domain Organic Search Keywords, display_limit=30. Note SERP features where the tool exposes them (featured snippet, PAA, AIO).
4. For the top 3 question-shaped keywords, `mcp__semrush__execute_report` — Organic Results, display_limit=10. Who currently owns the answer box.

**GA4 (ENRICHMENT):**
5. `mcp__google-analytics__run_report` — top landing pages by sessions, last 28 days. Snippet work goes on pages that already take traffic, not on a draft URL.

**CamoFox (ENRICHMENT):**
6. Load the #1 GSC page. Snapshot H1, first 150 words, FAQ headings, table presence, schema in raw HTML (view-source if needed), CTA above the fold. Optional: load the Google result for one target query **as a person would** — do not hammer `/goto` links. If CamoFox is down: `[SEO] CamoFox unavailable — snippet-owner snapshot skipped.`

If a source fails, print the `[SEO]` warning from core.md and continue.

## Step 3: Answer inventory

Build a table of at most 10 queries:

| Query | Intent | Current URL | Impressions | Position | Snippet owner | Our passage exists? | Action |
|---|---|---|---|---|---|---|---|

Prioritize:
1. High impressions, position 4–12, question-shaped, page already converts
2. "How much" / price / timeline / city + service — these stay BOFU
3. Drop pure "what is [generic noun]" unless that noun **is the offer**

If `/geo` already produced a fan-out table this session, reuse the worst MISSING branch as one row. Do not grow the inventory to absorb every inferred subquery.

## Step 4: Generate the AEO Playbook

**Section 1 — Verdict**
Which one query should own the next engineering hour. If the domain has no question-shaped impressions on a money page, say so and stop — do not invent a blog calendar.

**Section 2 — Passage spec (one page at a time)**
For each chosen URL + query, write the exact block the operator should paste:

- **Lead answer:** 40–60 words, first two sentences of the relevant section. Direct SVO. No throat-clearing.
- **Expandable passage:** 134–167 words total for Google AIO extraction. Atomic. Stands alone if the rest of the page is dropped.
- **H2:** the query in question form only when the page is already about that job. Do not tack five unrelated FAQs onto a service page.
- **Table or numbered list:** only if the query is comparison, steps, or pricing tiers.
- **Schema:** FAQPage or Product / LocalBusiness / Service JSON-LD that mirrors visible text. Never schema-only answers.

Stay inside core.md word caps. If the page is already at 500 words, replace a weak paragraph — do not append.

**Section 3 — SERP feature honesty**
- Featured snippet and PAA are passage contests. Winning them does not require more backlinks than the current owner if the passage is cleaner — but a zero-RD new domain still loses most AIO slots. Say which case this is.
- AI Overviews now show Preferred Sources labels. That is frequency for people who opted in, not a snippet override. Point to `/preferred-source` with a fit verdict, do not paste `publisher.js` here.
- Voice assistants read the same short answer. Write it so it can be spoken without the table.
- Query fan-out is real. A missing branch is a heading on this page, not a new silo.

**Section 4 — Anti-spam**
Reject any draft that:
- Announces machine-optimization in the lede
- Lists the operator's own product as "#1" inside a fake roundup on their own domain and expects an assistant to treat it as a third-party ranking
- Hides the answer below 800 words of preamble
- Justifies a new informational URL as "fan-out coverage"

**Section 5 — Measurement**
Re-check in 28 days via GSC (impressions + position for the query set) and one manual SERP look. Do not treat Ahrefs/Semrush URL-path loss after Google's `/goto` change as "we lost the snippet." Confirm on the live result.

## Step 5: Output

Follow the Output Protocol from core.md:
1. Print the full AEO playbook to terminal
2. Extract structured summary
3. Save to Graphiti with name `AEO — [domain]`

In Recommended Actions, first item is the single passage to ship, with the pasted copy included so the operator does not have to rewrite it.