magnus919_agent-skills/seo/SKILL.md

name, description, license, compatibility, metadata

name	description	license	compatibility	metadata
seo	Audit and improve website discoverability across traditional search, answer engines, and generative search. Use for technical SEO, on-page content, structured data, question and entity architecture, AI citations, crawler controls, agent-readable content, and reproducible visibility measurement. Do not use for only copy-editing, writing, CMS operations, or generic AI marketing claims without a defined search surface and verification plan.	MIT	Requires access to the target site or content for implementation and verification; bundled scripts use Python 3.9+ standard library only.	scope aliases search-answer-generative-engine-optimization SEO, AEO, GEO, LLMO, AI-search optimization

SEO

A full-spectrum search visibility skill. It treats traditional SEO, Answer Engine Optimization (AEO), and Generative Engine Optimization (GEO) as overlapping work across different search and answer surfaces, not as separate collections of ranking hacks.

Operating model

1. Scope the surface and outcome. Name the provider, product or search surface, audience, entity, questions, business outcome, and exclusions. Define whether success means crawl access, index eligibility, retrieval, mention, citation, citation correctness, share of voice, referral, or conversion.
2. Inspect and research. Audit the public/rendered page and technical delivery. Read current first-party provider guidance and authoritative subject sources. Treat practitioner claims and vendor studies as hypotheses unless methods and scope support more.
3. Map intent to canonical content. Use topics and question clusters, but avoid manufacturing near-duplicate pages. Assign one canonical answer location, entity ownership, evidence, freshness owner, and internal links.
4. Improve people-first content and structure. Put a concise answer near the relevant heading, then supporting evidence, qualifications, and useful detail. Preserve natural prose, distinct point of view, accessibility, and human value.
5. Implement only supported controls. Fix crawlability, indexability, metadata, internal links, page experience, textual content, structured-data parity, sitemaps, freshness, and provider-specific crawler or preview controls. Optional files such as llms.txt or Markdown representations are provider-scoped proposals, not universal requirements.
6. Measure at the correct boundary. Freeze prompts and versions, capture exact answers and citations, score citation correctness, and use provider-native reports where available. Separate implementation evidence from observed visibility and causal claims.
7. Verify and learn. Recheck the rendered/public boundary, validate structured data, inspect search-console or provider evidence, and run bounded one-variable experiments. Do not declare success from HTTP 200, parseable JSON-LD, a single answer, or a third-party score.

Terminology and boundaries

• SEO is the umbrella: improving a site's eligibility, discoverability, interpretation, and useful visibility in search systems.
• AEO is a stakeholder label for answer-oriented work, including direct answers, snippets, knowledge surfaces, and answer-engine inclusion.
• GEO is a stakeholder label for visibility in generated answers, especially being selected, cited, or factually absorbed into a synthesized response. The term originated in the 2024 KDD paper by Aggarwal et al.; it does not establish a universal algorithm.
• LLMO and AI-search optimization are overlapping labels. Preserve the target organization's term, then define the measurable outcome and provider scope.
• These terms do not guarantee ranking, inclusion, citation, traffic, recommendation, or conversion. A mention is not a citation, a citation is not proof of correctness, and a citation is not a click.

Evidence rules

Use these labels in plans and reports:

• Primary documentation: provider, standards body, schema vocabulary, or tool owner describes its own behavior.
• Observed: recorded crawl, rendered page, provider dashboard, exact answer, URL, or reproducible local result.
• Independent study: disclosed method and dataset, with generalization limits stated.
• Vendor-reported: useful for hypothesis generation, not a universal rule.
• Inference: reasoned interpretation beyond direct observation.
• Unresolved: provider-dependent, stale, contradicted, or not verified.

Reject claims such as “FAQ schema guarantees citations,” “short paragraphs are required by all engines,” “llms.txt is a search standard,” or “allowing a crawler guarantees inclusion.” Record source URL, access date, provider scope, exact support, what it does not prove, confidence, and refresh trigger for every material claim.

Reference routing

Need	Read
Crawlability, indexability, robots, sitemaps, performance, canonicals, mobile, HTTPS	references/technical-seo.md
Titles, descriptions, headings, content quality, links, images	references/onpage-seo.md
Schema.org, JSON-LD, rich-result eligibility, visible parity	references/schema-markup.md and references/structured-data.md
Topics, question clusters, entities, answer blocks, evidence architecture	references/content-and-entity-architecture.md
Full answer/generative implementation sequence and completion gate	references/implementation-playbook.md
Provider guidance, crawler identities, robots and preview controls	references/platform-guidance.md and references/discovery-and-freshness.md
llms.txt, Markdown delivery, content negotiation, provider support	references/agent-readable-content.md
Outcome definitions, confidence, and rejected claims	references/evidence-boundaries.md
Prompt sets, citation logs, metrics, experiments, and confounders	references/measurement-and-experimentation.md
Ghost metadata and injection	references/ghost-metadata.md
Content strategy, topic clusters, keywords, gaps, SERP features	references/content-strategy-seo.md
Source URLs, access dates, authority tiers, and refresh notes	references/source-index.md

Scripts and templates

Run from the skill directory:

python3 scripts/aeo_audit.py <page.html-or-URL> --json
python3 scripts/build_prompt_matrix.py <topics.json> --output prompt-matrix.json
python3 -m pytest scripts/test_aeo_scripts.py

The scripts are read-only and use the Python standard library. They inspect source HTML; they do not execute JavaScript, call an LLM, publish, submit URLs, or change crawler policy. Templates cover implementation plans, question clusters, citation observations, optional llms.txt, and crawler-policy decisions.

Audit output

Use assets/audit-report-template.md and distinguish:

• Observed findings: what the inspected page, response, dashboard, or answer actually shows.
• Recommended changes: proposed actions with owner, risk, expected mechanism, and verification.
• Provider scope: which engine or search surface the evidence applies to.
• Status: implemented, verified, observed, inferred, or unresolved.

When not to use

Do not use this skill alone for only mechanical copy-editing, ordinary article writing, CMS administration, or a generic request to “rank better” without a defined target, evidence boundary, or measurable outcome. Route those to the relevant writing, copy-editing, CMS, or product skill.

Portability

Use the host agent's normal mechanisms to load references, templates, and scripts. Do not assume a particular profile system, orchestrator, memory service, CMS, search console, or provider API.