The complete rule book of AI search optimization.

How to invoke this skill

This skill is triggered when the user asks any of:

• "Run a GEO/AEO audit on [URL]"
• "How do I get cited by ChatGPT/Claude/Perplexity?"
• "Make my site citable by AI search"
• "Generate llms.txt for [URL]"
• "Why is my site not showing up in AI Overviews?"
• "Improve my AI search visibility"
• "Fix my robots.txt for AI bots"

When invoked, it defaults to running audit on the provided URL and presenting a confidence-labeled findings list. Recommendations without confidence labels are never presented.

Architecture overview

┌────────────────────────────────────────────────────────────────────┐
│                        best-aeo-skill                              │
├────────────────────────────────────────────────────────────────────┤
│                                                                    │
│  User intent                                                       │
│     ↓                                                              │
│  CLI dispatcher (audit | fix | monitor | compare)                  │
│     ↓                                                              │
│  Orchestrator                                                      │
│     ↓                                                              │
│  ┌─────────────┬─────────────┬─────────────┬─────────────┐         │
│  │ Technical   │ Citability  │ Schema      │ Entity      │         │
│  │ agent       │ agent       │ agent       │ agent       │         │
│  └─────┬───────┴─────┬───────┴─────┬───────┴─────┬───────┘         │
│        ↓             ↓             ↓             ↓                 │
│       33 evidence collectors (parallel execution)                  │
│        ↓             ↓             ↓             ↓                 │
│  ┌───────────────────────────────────────────────────────┐         │
│  │       Composite Scorer (4-vector weighted sum)         │        │
│  │  Profile-adaptive weights · Confidence labeling        │        │
│  └────────────────────────────┬───────────────────────────┘        │
│                               ↓                                    │
│  Ranked findings + projected score impact + auto-fix paths         │
│                                                                    │
└────────────────────────────────────────────────────────────────────┘

The Composite GEO Score

A single 0–100 number computed as a weighted sum of four vectors. Default weights:

Score bands

Interpretation rules

• A score below 36 always indicates a technical or schema problem, not a content problem. Fixing prose first is the wrong order.
• A score 68–85 is the typical post-fix state for a well-built site. Reaching 86+ requires sustained content investment.
• Scores fluctuate ±3 due to crawl variance. Treat changes within ±3 as noise.

Adaptive weights per business profile

Business profiles re-weight the four vectors:

Pass --profile saas (or other) to switch.

Sub-skills reference

audit — diagnose

Compute the GEO Score. Return ranked findings with confidence labels.

bestaeo audit --url https://yoursite.com --profile saas --format markdown

• Inputs: --url, --sitemap, --profile, --format [json|markdown|html|sarif|junit]
• Outputs: Composite GEO Score, per-vector breakdown, ranked findings, projected score impact, recommended next sub-skill

fix-content — rewrite

Adds expert quotes (+41% citations), inserts statistics (+40% visibility), embeds source attributions (+115% citation likelihood). Idempotent. Preserves voice. Never invents facts.

bestaeo fix-content --url URL --apply

fix-schema — generate JSON-LD

Auto-creates FAQPage (highest AI citation surface), Article, Organization, HowTo, Speakable, Product, BreadcrumbList, Person schemas. Validates with Schema.org. Outputs ready-to-paste blocks or auto-injects.

bestaeo fix-schema --types faq,article,org

fix-llmstxt — AI catalog

Generates a llmstxt.org-compliant catalog: H1, summary, primary pages, optional reference. Pulls from sitemap or crawl. Anthropic honors this for ClaudeBot.

bestaeo fix-llmstxt --base-url URL

fix-robotstxt — patch robots

Adds explicit Allow: rules for 27 AI bots: GPTBot, ChatGPT-User, ClaudeBot, anthropic-ai, Claude-Web, PerplexityBot, Google-Extended, OAI-SearchBot, Applebot-Extended, CCBot, +17 more. Diff-aware.

bestaeo fix-robotstxt --url URL

compare — benchmark

Side-by-side GEO Score vs 2–5 competitors. Detects per-vector deltas; generates a delta-action list — what to ship next to leapfrog them.

bestaeo compare --you URL --them URL1,URL2

monitor — track

Stores GEO Score history; detects regressions, alerts via webhook/Slack/email. Detects content decay (3.2× citation drop after 30 days). CI/CD-ready.

bestaeo monitor --url URL --threshold 80

Specialist agents

The skill delegates to 5 specialist agents in parallel during full audits. Each owns one vector + monitor for longitudinal:

Evidence collectors (33 total)

Every finding traces back to one or more of these. Output is what determines confidence label.

Technical (9)

1. robots_check — parses robots.txt; verifies User-agent rules
2. ai_bot_access — tests fetch as GPTBot, ClaudeBot, PerplexityBot
3. js_render — compares HTML before/after JS execution
4. cdn_blocking — detects Cloudflare/Akamai bot challenges
5. response_codes — 200/301/404/500 distribution
6. sitemap_check — XML sitemap presence + freshness
7. http2_check — HTTP/2 or HTTP/3 support
8. mobile_render — viewport, touch targets, mobile-first
9. lazyload_check — confirms above-fold content not lazy-loaded

Citability (10)

10. statistic_density — counts numeric claims per 100 words
11. quote_extractor — counts and validates quoted passages
12. citation_check — counts inline links to external sources
13. freshness_check — extracts dateModified, computes content age
14. readability — Flesch-Kincaid + sentence variance
15. passage_score — passage-level extractability (chunk-readiness for RAG)
16. fluency_check — detects awkward AI-rewrite patterns
17. hedge_density — counts hedging language ("may", "might", "could")
18. claim_verifier — flags factual claims without citation
19. rag_chunk_score — 250-word chunks ready for retrieval

Schema (7)

20. schema_validate — Schema.org compliance
21. faq_check — FAQPage presence + structure
22. article_check — Article required fields
23. jsonld_lint — JSON-LD syntax validity
24. speakable_check — Speakable Spec markup
25. product_check — Product + AggregateRating + offers
26. breadcrumb_check — BreadcrumbList navigation

Entity (7)

27. entity_extractor — NER for Person, Org, Place, Product
28. author_check — author markup + credentials
29. knowledge_graph — sameAs links to Wikidata, Wikipedia
30. nap_consistency — Name/Address/Phone matching across surfaces
31. brand_signal — brand mention frequency in content
32. sameas_links — verifies all sameAs URLs resolve
33. expertise_signals — credentials, affiliations, bylines

4 scoring frameworks bundled

CORE-EEAT — 80-item content quality publish-gate

Adopted from aaron-he-zhu/seo-geo-claude-skills.

• Experience (20): firsthand examples, real screenshots, original photos, dated experiences, personal voice, specific use cases, original tests, hands-on insights, primary research…
• Expertise (20): author credentials, named contributors, technical depth, specific terminology, methodology disclosure, source code, primary research links, edge case coverage…
• Authority (20): citing primary sources, linking to authoritative domains, credential markup, organization markup, awards, press mentions, peer-reviewed citations, .edu/.gov sources, sameAs to KG…
• Trust (20): HTTPS, contact info, privacy policy, transparent author bio, last-updated dates, fact-check links, error corrections published, no AI-generated boilerplate…

Pass threshold: 60/80 = pass. 50–59 = revise. <50 = block.

CITE — 40-item domain authority benchmark

• Backlinks (10): editorial inbound links, anchor diversity, follow ratio, referring domain growth, link velocity, contextual placement, .edu/.gov links
• Authority (10): topical authority depth, content cluster coverage, internal linking density, hub-and-spoke architecture, freshness average, expert contribution
• Brand (10): branded search volume, social mention velocity, Wikipedia entry, founder authority, podcast guest appearances, owned media
• Trust (10): historical domain age, ownership transparency, regulatory compliance pages, third-party audits, certifications, customer logos, case studies with metrics

Princeton GEO Tactics — 9 research-validated methods

From Aggarwal et al. KDD 2024.

Confidence Rubric — 3 labels

Anti-hallucination labeling. Detailed below at Confidence rubric.

The 100 Rules of AEO/GEO Optimization

Organized by category. Each rule is research-validated, traceable to a citation, and actionable.

Category A: AI Crawler Access (Rules 1–10)

Category B: Content Citability (Rules 11–35)

Category C: Structured Data (Rules 36–55)

Category D: Entity & Brand (Rules 56–75)

Category E: Multi-engine optimization (Rules 76–90)

Category F: Anti-patterns (Rules 91–100)

View / download the full SKILL.md (40 KB) →

Confidence labeling rubric

Every finding output by this skill MUST carry exactly one label.

Confirmed

Directly observed by ≥1 evidence collector. Examples:

• HTTP response code present in trace
• DOM element absent in parsed HTML
• JSON-LD parser returned syntax error
• Author byline element not found

When in doubt about confidence: prefer Likely over Confirmed.

Likely

Strong inference from ≥2 evidence collectors that agree, OR ≥1 collector with high specificity that maps to a known anti-pattern. Examples:

• schema_validate returned no FAQPage AND quote_extractor found Q&A patterns
• statistic_density returned 0.2/100 AND citability_score ≤ 50
• robots.txt allows GPTBot but CDN blocks GPTBot fetch

Hypothesis

LLM judgment, speculative recommendation, or single weak signal. Examples:

• Tone analysis suggesting voice mismatch
• Recommendations for future-proofing emerging engines
• Recommendations relying on user-only-knowable context

NEVER present a Hypothesis as a fix path without flagging human review needed.

Output formats

The skill produces output in 5 formats simultaneously when requested.

Markdown (default)

Human-readable report with score, findings, and projected impact.

JSON

{
  "url": "https://example.com",
  "geoScore": 64,
  "scoreBand": "Foundation",
  "vectors": {
    "technical": 72,
    "citability": 58,
    "schema": 48,
    "entity": 78
  },
  "findings": [
    {
      "id": "F-001",
      "severity": "high",
      "confidence": "Confirmed",
      "category": "schema",
      "rule": "Rule 36",
      "message": "Missing FAQPage schema",
      "evidence": ["schema_validate", "quote_extractor"],
      "projectedImpact": 18,
      "fixCommand": "bestaeo fix-schema --types faq"
    }
  ]
}

Other formats

• HTML — Visual dashboard with interactive elements. Auto-opens in browser.
• SARIF — Static Analysis Results Interchange Format. For GitHub Code Scanning integration.
• JUnit — For CI/CD pipelines. Each finding as a test case; severity maps to pass/fail.

Common workflows

Workflow 1: First-time audit

bestaeo audit --url https://yoursite.com --profile saas --format markdown

Read top 3 fixes. Apply manually or via:

bestaeo fix --url https://yoursite.com --apply

Workflow 2: CI/CD gate

bestaeo audit --url $DEPLOY_URL --format junit > geo-report.xml
bestaeo monitor --url $DEPLOY_URL --threshold 80 --fail-on-drop

Workflow 3: Competitor analysis

bestaeo compare \
  --you https://yoursite.com \
  --them https://competitor1.com,https://competitor2.com \
  --format markdown

Workflow 4: Site-wide audit

bestaeo audit \
  --sitemap https://yoursite.com/sitemap.xml \
  --max-urls 100 \
  --profile publisher \
  --format html

Anti-patterns & pitfalls

• Don't fix prose first if technical/schema is below 60. Prose rewriting is high-effort, low-leverage when foundations are missing.
• Don't trust llms.txt as a foundational signal. 0.1% of AI bots fetch it as of 2026.
• Don't ship synthetic FAQ content. AI engines detect mass-generated Q&A.
• Don't auto-fix on production without backup. fix --apply should be tested on staging first.
• Don't optimize for one engine in isolation. Optimizing only for ChatGPT loses Perplexity and Claude.
• Don't ignore citation decay. A 90-day-old article with declining citation needs refresh, not "more backlinks".

CLI reference

Global flags

• --profile — business profile (default, saas, ecommerce, publisher, local, agency, devtools, academic)
• --format — output format (json, markdown, html, sarif, junit)
• --output — write to file
• --engine — target a specific engine (chatgpt, claude, perplexity, gemini, ai-overviews, all)
• --verbose — full evidence dump
• --no-color — for piping
• --ci — CI/CD mode (machine-readable, exit codes)

Sub-commands

bestaeo audit          # diagnose
bestaeo fix            # apply fixes
bestaeo fix-content    # content rewrites only
bestaeo fix-schema     # schema generation only
bestaeo fix-llmstxt    # llms.txt only
bestaeo fix-robotstxt  # robots.txt only
bestaeo compare        # head-to-head
bestaeo monitor        # track over time
bestaeo init           # initialize config in current directory
bestaeo --version
bestaeo --help

Citations & references

This skill is built on peer-reviewed research and empirical analysis.

Primary research

1. Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., Deshpande, A. (2024). GEO: Generative Engine Optimization. Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD '24). arxiv:2311.09735 · DOI
2. AutoGEO Authors (2026). AutoGEO: Automatic Optimization for Generative Engine Citation. International Conference on Learning Representations (ICLR 2026).
3. C-SEO Bench Authors (2025). Conversational SEO Benchmark: Empirical Analysis of Content Manipulation Effectiveness in AI Search.

Empirical analysis

4. 199-biotechnologies team (2025). 41M-citation corpus analysis.
5. OtterlyAI (2026). llms.txt Adoption Study.
6. SE Ranking (2026). llms.txt Audit of 300,000 Domains.

Open-source projects studied (predecessors)

11. AgriciDaniel/claude-seo (5.8k ⭐)
12. aaron-he-zhu/seo-geo-claude-skills (1.4k ⭐)
13. Bhanunamikaze/Agentic-SEO-Skill (458 ⭐)
14. Auriti-Labs/geo-optimizer-skill (328 ⭐)
15. 199-biotechnologies/claude-skill-seo-geo-optimizer (28 ⭐)

best-aeo-skill v1.0.0 · MIT · 2026 · bestaeoskill.com