claude-code-arcane 1.0.0

package/skills/ai-seo/SKILL.md

@@ -0,0 +1,172 @@
+---
+name: ai-seo
+description: "Optimize content to get cited by AI search engines — ChatGPT, Perplexity, Google AI Overviews, Claude, Gemini, Copilot. GEO / AI SEO."
+argument-hint: ""
+category: "marketing-seo"
+user-invocable: true
+allowed-tools: Read, Glob, Grep, Bash, Write, Edit
+---
+
+# AI SEO
+
+You are an expert in generative engine optimization (GEO) — the discipline of making content citeable by AI search platforms. Your goal is to help content get extracted, quoted, and cited by ChatGPT, Perplexity, Google AI Overviews, Claude, Gemini, and Microsoft Copilot.
+
+This is not traditional SEO. Traditional SEO gets you ranked. AI SEO gets you cited. Those are different games with different rules.
+
+## Before Starting
+
+**Check for context first:**
+If `marketing-context.md` exists, read it. It contains existing keyword targets, content inventory, and competitor information — all of which inform where to start.
+
+Gather what you need:
+
+### What you need
+- **URL or content to audit** — specific page, or a topic area to assess
+- **Target queries** — what questions do you want AI systems to answer using your content?
+- **Current visibility** — are you already appearing in any AI search results for your targets?
+- **Content inventory** — do you have existing pieces to optimize, or are you starting from scratch?
+
+If the user doesn't know their target queries: "What questions would your ideal customer ask an AI assistant that you'd want your brand to answer?"
+
+## How This Skill Works
+
+Three modes. Each builds on the previous, but you can start anywhere:
+
+### Mode 1: AI Visibility Audit
+Map your current presence (or absence) across AI search platforms. Understand what's getting cited, what's getting ignored, and why.
+
+### Mode 2: Content Optimization
+Restructure and enhance content to match what AI systems extract. This is the execution mode — specific patterns, specific changes.
+
+### Mode 3: Monitoring
+Set up systems to track AI citations over time — so you know when you appear, when you disappear, and when a competitor takes your spot.
+
+---
+
+## How AI Search Works (and Why It's Different)
+
+Traditional SEO: Google ranks your page. User clicks through. You get traffic.
+
+AI search: The AI reads your page (or has already indexed it), extracts the answer, and presents it to the user — often without a click. You get cited, not ranked.
+
+**The fundamental shift:**
+- Ranked = user sees your link and decides whether to click
+- Cited = AI decides your content answers the question; user may never visit your site
+
+This changes everything:
+- **Keyword density** matters less than **answer clarity**
+- **Page authority** matters less than **answer extractability**
+- **Click-through rate** is irrelevant — the AI has already decided you're the answer
+- **Structured content** (definitions, lists, tables, steps) outperforms flowing narrative
+
+But here's what traditional SEO and AI SEO share: **authority still matters**. AI systems prefer sources they consider credible — established domains, cited works, expert authorship. You still need backlinks and domain trust. You just also need structure.
+
+See [references/ai-search-landscape.md](references/ai-search-landscape.md) for how each platform (Google AI Overviews, ChatGPT, Perplexity, Claude, Gemini, Copilot) selects and cites sources.
+
+---
+
+## The 3 Pillars of AI Citability
+
+Every AI SEO decision flows from these three:
+
+### Pillar 1: Structure (Extractable)
+
+AI systems pull content in chunks. They don't read your whole article and then paraphrase it — they find the paragraph, list, or definition that directly answers the query and lift it.
+
+Your content needs to be structured so that answers are self-contained and extractable:
+- Definition block for "what is X"
+- Numbered steps for "how to do X"
+- Comparison table for "X vs Y"
+- FAQ block for "questions about X"
+- Statistics with attribution for "data on X"
+
+Content that buries the answer in page 3 of a 4,000-word essay is not extractable. The AI won't find it.
+
+### Pillar 2: Authority (Citable)
+
+AI systems don't just pull the most relevant answer — they pull the most credible one. Authority signals in the AI era:
+
+- **Domain authority**: High-DA domains get preferential treatment (traditional SEO signal still applies)
+- **Author attribution**: Named authors with credentials beat anonymous pages
+- **Citation chain**: Your content cites credible sources → you're seen as credible in turn
+- **Recency**: AI systems prefer current information for time-sensitive queries
+- **Original data**: Pages with proprietary research, surveys, or studies get cited more — AI systems value unique data they can't get elsewhere
+
+### Pillar 3: Presence (Discoverable)
+
+AI systems need to be able to find and index your content. This is the technical layer:
+
+- **Bot access**: AI crawlers must be allowed in robots.txt (GPTBot, PerplexityBot, ClaudeBot, etc.)
+- **Crawlability**: Fast page load, clean HTML, no JavaScript-only content
+- **Schema markup**: Structured data (Article, FAQPage, HowTo, Product) helps AI systems understand your content type
+- **Canonical signals**: Duplicate content confuses AI systems even more than traditional search
+- **HTTPS and security**: AI crawlers won't index pages with security warnings
+
+---
+
+## Mode 1: AI Visibility Audit
+
+Three-step audit: (1) check robots.txt for AI bot access (GPTBot, PerplexityBot, ClaudeBot, Google-Extended), (2) test target queries on each platform and document who gets cited, (3) score content extractability against the 7-point checklist.
+
+See [references/ai-search-landscape.md](references/ai-search-landscape.md) for the complete bot list, robots.txt templates, platform testing guide, and extractability checklist.
+
+---
+
+## Mode 2: Content Optimization
+
+> See [references/optimization-playbook.md](references/optimization-playbook.md) for the 6 content patterns that get cited (definition block, numbered steps, comparison table, FAQ, statistics, expert quotes), rewriting for extractability, and schema markup for AI discoverability.
+
+See also [references/content-patterns.md](references/content-patterns.md) for ready-to-use templates.
+
+---
+
+## Mode 3: Monitoring
+
+AI search is volatile. Citations change. Track weekly via manual query testing on Perplexity and ChatGPT, plus Google Search Console AI Overviews filter.
+
+> See [references/monitoring-guide.md](references/monitoring-guide.md) for the full tracking setup, signal-by-platform frequency table, citation drop diagnosis, and templates.
+
+---
+
+## Proactive Triggers
+
+Flag these without being asked:
+
+- **AI bots blocked in robots.txt** — If GPTBot, PerplexityBot, or ClaudeBot are blocked, flag it immediately. Zero AI visibility is possible until fixed, and it's a 5-minute fix. This trumps everything else.
+- **No definition block on target pages** — If the page targets informational queries but has no self-contained definition in the first 300 words, it won't win definitional AI Overviews. Flag before doing anything else.
+- **Unattributed statistics** — If key pages contain statistics without named sources and years, they're less citable than competitor pages that do. Flag all naked stats.
+- **Schema markup absent** — If the site has no FAQPage or HowTo schema on relevant pages, flag it as a quick structural win with asymmetric impact for process and FAQ queries.
+- **JavaScript-rendered content** — If important content only appears after JavaScript execution, AI crawlers may not see it at all. Flag content that's hidden behind JS rendering.
+
+---
+
+## Output Artifacts
+
+| When you ask for... | You get... |
+|---|---|
+| AI visibility audit | Platform-by-platform citation test results + robots.txt check + content structure scorecard |
+| Page optimization | Rewritten page with definition block, extractable patterns, schema markup spec, and comparison to original |
+| robots.txt fix | Updated robots.txt with correct AI bot allow rules + explanation of what each bot is |
+| Schema markup | JSON-LD implementation code for FAQPage, HowTo, or Article — ready to paste |
+| Monitoring setup | Weekly tracking template + Google Search Console filter guide + citation log spreadsheet structure |
+
+---
+
+## Communication
+
+All output follows the structured standard:
+- **Bottom line first** — answer before explanation
+- **What + Why + How** — every finding includes all three
+- **Actions have owners and deadlines** — no "consider reviewing..."
+- **Confidence tagging** — 🟢 verified (confirmed by citation test) / 🟡 medium (pattern-based) / 🔴 assumed (extrapolated from limited data)
+
+AI SEO is still a young field. Be honest about confidence levels. What gets cited can change as platforms evolve. State what's proven vs. what's pattern-matching.
+
+---
+
+## Related Skills
+
+- **content-production**: Use to create the underlying content before optimizing for AI citation. Good AI SEO requires good content first.
+- **content-humanizer**: Use after writing for AI SEO. AI-sounding content ironically performs worse in AI citation — AI systems prefer content that reads credibly, which usually means human-sounding.
+- **seo-audit**: Use for traditional search ranking optimization. Run both — AI SEO and traditional SEO are complementary, not competing. Many signals overlap.
+- **content-strategy**: Use when deciding which topics and queries to target for AI visibility. Strategy first, then optimize.

package/skills/ai-seo/references/ai-search-landscape.md

@@ -0,0 +1,191 @@
+# AI Search Landscape
+
+How each major AI search platform selects, weights, and cites sources. Use this to calibrate your optimization strategy per platform.
+
+Last updated: 2026-03 — this landscape changes fast. Verify platform behavior with manual testing before making major decisions.
+
+---
+
+## The Fundamental Model
+
+Every AI search platform follows the same broad pipeline:
+
+1. **Index** — Crawl and store web content (or use a third-party index)
+2. **Retrieve** — For a given query, retrieve candidate documents
+3. **Extract** — Pull the most relevant passages from those documents
+4. **Generate** — Synthesize an answer, often citing the sources
+5. **Present** — Show the answer to the user, with or without sources visible
+
+Your leverage points are steps 1-3. By the time generation happens, you've either been selected or you haven't.
+
+---
+
+## Platform-by-Platform Breakdown
+
+### Google AI Overviews
+
+**What it is:** AI-generated answer boxes appearing above organic search results. Rollout expanded globally in 2024-2025.
+
+**How it selects sources:**
+- Uses Google's own index (you must rank in traditional Google search first — this is NOT optional)
+- Strongly prefers pages that already rank in the top 10 for the query
+- Favors content with structured data (FAQPage, HowTo schemas)
+- The featured passage is typically lifted from a page's most extractable paragraph — usually a definition or a direct answer near the top
+- Recency matters more here than elsewhere for news-adjacent queries
+
+**Citation behavior:**
+- Shows 3-7 source links typically
+- Cited sources don't always correlate with position 1-3 in organic results
+- Pages that had featured snippets before AI Overviews launched tend to appear in AI Overviews
+
+**What to prioritize for Google AI Overviews:**
+1. Rank in traditional search first (prerequisite)
+2. Add FAQPage schema
+3. Put a direct answer in the first 200 words
+4. Get backlinks from high-authority sites (still matters)
+5. Set `Google-Extended` to Allow in robots.txt
+
+**Monitoring:** Google Search Console → Performance → Search type: AI Overviews
+
+---
+
+### ChatGPT (with Browsing / Search)
+
+**What it is:** OpenAI's ChatGPT has web browsing capability (via Bing) plus its own live search product. When users ask factual questions or enable browsing, it retrieves and cites web sources.
+
+**How it selects sources:**
+- Uses Bing's index (Microsoft partnership) — Bing crawl and indexing quality matters
+- GPTBot also crawls independently for training data (distinct from search citations)
+- For search-backed answers: pulls several sources, synthesizes, cites inline
+- Prefers authoritative domains — news outlets, Wikipedia, academic sources, established company blogs
+- Content with clear, extractable answers wins over dense narrative
+
+**Citation behavior:**
+- Inline citations in the answer ("according to [Source]")
+- Sources panel at the bottom
+- Not all cited sources get equal weight in the synthesis
+
+**What to prioritize for ChatGPT:**
+1. Ensure Bing has indexed your pages (submit to Bing Webmaster Tools)
+2. Allow `GPTBot` in robots.txt
+3. Structure content with explicit definition and step patterns
+4. Author attribution with credentials helps — include author bylines
+5. Original data and research get preferential citation
+
+**Bing indexing check:** Bing Webmaster Tools → URL Inspection
+
+---
+
+### Perplexity
+
+**What it is:** AI-native search engine built on real-time web retrieval. Every answer cites sources with a numbered reference panel. Among the most transparent about citation.
+
+**How it selects sources:**
+- Has its own crawler (PerplexityBot) plus access to third-party indexes
+- Real-time retrieval for every query — very current
+- Strongly rewards structural clarity: numbered lists, definition blocks, tables
+- Tends to pull from multiple perspectives on a query (shows variety in citations)
+- Recency bias is strong — old content competes poorly against recent content on current topics
+
+**Citation behavior:**
+- Numbers every cited source
+- Shows the exact passage it pulled from (if you inspect carefully)
+- Citations appear inline and in a source panel
+
+**What to prioritize for Perplexity:**
+1. Allow `PerplexityBot` in robots.txt (critical)
+2. Use numbered lists, definition blocks, and tables extensively
+3. Keep content current — update pages when information changes
+4. For competitive topics, publish comprehensive pieces that cover the query more completely than alternatives
+5. Include specific data with dates ("In Q1 2025, X% of...") — Perplexity responds strongly to timestamped specifics
+
+**Tracking:** Perplexity doesn't offer a publisher dashboard. Manual testing is the only method currently.
+
+---
+
+### Claude (Anthropic)
+
+**What it is:** Claude.ai now has web search capability. When users ask questions that require current information, Claude retrieves and cites sources.
+
+**How it selects sources:**
+- Uses a third-party search index (search partnership)
+- ClaudeBot crawls for training purposes — separate from search citations
+- Prefers clearly structured, credible content
+- High-authority domains get preference
+- Technical and expert-authored content performs well given Claude's user base
+
+**Citation behavior:**
+- Inline citations in responses
+- Source list at end of response
+- Tends toward fewer, higher-quality citations vs. showing many sources
+
+**What to prioritize for Claude:**
+1. Allow `ClaudeBot` and `anthropic-ai` in robots.txt
+2. Focus on content quality and accuracy — Claude's users are often technical and will notice errors
+3. Expert authorship and institutional credibility matter here
+4. Long-form, well-researched pieces tend to perform better than thin listicles
+
+---
+
+### Google Gemini
+
+**What it is:** Google's AI assistant, separate from Google Search but increasingly integrated. Uses Google's web index.
+
+**How it selects sources:**
+- Full access to Google's index
+- Similar selection criteria to Google AI Overviews but different interface
+- Schema markup influences what Gemini can understand about your content type
+- Prefers content that directly answers conversational queries
+
+**What to prioritize for Gemini:**
+- Same fundamentals as Google AI Overviews (they share the index)
+- Conversational phrasing in your content helps — Gemini handles voice/chat-style queries
+- Allow `Google-Extended` in robots.txt (covers both AI Overviews and Gemini)
+
+---
+
+### Microsoft Copilot
+
+**What it is:** Microsoft's AI assistant integrated into Bing, Windows, Office 365, and Edge. Uses Bing's index.
+
+**How it selects sources:**
+- Bing index (same as ChatGPT browsing)
+- Integrated into productivity contexts — Office documents, business queries
+- B2B and professional content performs particularly well
+- Bing's relevance signals apply
+
+**Citation behavior:**
+- Source cards in the Copilot interface
+- Inline citations in longer answers
+
+**What to prioritize for Copilot:**
+- Ensure strong Bing indexing (submit to Bing Webmaster Tools, build Bing-friendly signals)
+- For B2B companies: professional tone and industry-specific expertise matters more here
+- FAQ and definition patterns work well for business query types
+
+---
+
+## Cross-Platform Summary
+
+| Signal | Google AI Overviews | ChatGPT | Perplexity | Claude | Copilot |
+|---|---|---|---|---|---|
+| Must rank in traditional search | ✅ Yes | Bing only | No | No | Bing only |
+| Bot to allow | Google-Extended | GPTBot | PerplexityBot | ClaudeBot | (via Bing) |
+| Schema markup impact | High | Medium | Low | Medium | Medium |
+| Content recency weight | High | Medium | Very high | Medium | Medium |
+| Original data advantage | High | High | High | High | High |
+| FAQ pattern extraction | Very high | High | High | Medium | High |
+| Numbered steps extraction | High | High | Very high | High | High |
+| Author attribution impact | Medium | High | Low | High | Medium |
+
+---
+
+## What No Platform Does (Yet)
+
+Things that are widely assumed but not confirmed:
+
+- **Direct "opt-in to citations" programs**: None of the major platforms have a verified publisher program that guarantees citation
+- **Predictable citation ranking**: Even with perfect structure, citations are non-deterministic — the same query on the same platform can produce different citations on consecutive days
+- **Real-time citation tracking**: No platform offers publishers a dashboard showing when they're cited and for which queries (Google Search Console for AI Overviews is the closest, and it's limited)
+
+Plan your AI SEO strategy for influence, not for guaranteed outcomes. Maximize your signal quality, then track and iterate.

package/skills/ai-seo/references/content-patterns.md

@@ -0,0 +1,276 @@
+# Content Patterns for AI Citability
+
+Ready-to-use block templates for each content pattern that AI search engines reliably extract and cite. Copy, adapt, and embed in your pages.
+
+---
+
+## Why Patterns Matter
+
+AI systems don't read pages the way humans do. They scan for extractable chunks — self-contained passages that can be pulled out and quoted without losing meaning.
+
+The patterns below are structured to be self-contained by design. If the AI pulls paragraph 3 without paragraph 2, the citation should still make sense.
+
+---
+
+## Pattern 1: Definition Block
+
+**Used for:** "What is X" queries — the most common AI Overview trigger.
+
+**Requirements:**
+- First sentence: direct definition
+- Second sentence: why it matters or how it works
+- Third sentence (optional): example or context
+- Placed in first 300 words of the page
+
+**Template:**
+```markdown
+**[Term]** is [precise definition — what it is, what it does, who uses it]. 
+[One sentence on why it matters or what problem it solves]. 
+[Optional: one sentence example — "For example, a SaaS company might use X to..."].
+```
+
+**Example:**
+```markdown
+**Churn rate** is the percentage of customers who cancel or stop using a service within a given period, typically measured monthly or annually. It directly impacts recurring revenue — a 5% monthly churn means losing over half your customer base each year. For subscription SaaS, a healthy monthly churn rate is typically below 2%.
+```
+
+**Tips:**
+- Bold the term on its first use
+- Don't start with "In the world of..." or "When it comes to..."
+- The definition should work even if the reader knows nothing about the topic
+
+---
+
+## Pattern 2: Numbered Steps (How-To)
+
+**Used for:** "How to X" and "How do I X" queries.
+
+**Requirements:**
+- Numbered list (not bulleted)
+- Each step starts with an action verb
+- Each step is self-contained (can be cited alone)
+- 5-10 steps maximum
+- Pair with HowTo schema markup
+
+**Template:**
+```markdown
+## How to [Task]
+
+1. **[Verb phrase]** — [1-2 sentence explanation of this specific step]
+2. **[Verb phrase]** — [1-2 sentence explanation]
+3. **[Verb phrase]** — [1-2 sentence explanation]
+4. **[Verb phrase]** — [1-2 sentence explanation]
+5. **[Verb phrase]** — [1-2 sentence explanation]
+```
+
+**Example:**
+```markdown
+## How to Reduce SaaS Churn
+
+1. **Define your activation event** — Identify the specific action that signals a user has experienced core product value. For Slack, it's 2,000 messages sent. For Dropbox, it's saving the first file.
+2. **Instrument the activation funnel** — Add event tracking from signup to activation. Find the step where most users drop off — that's your highest-leverage point.
+3. **Build a customer health score** — Combine login frequency, feature adoption, and support ticket volume into a single score. Customers below 40 get proactive outreach.
+4. **Segment churn by cohort** — Not all churn looks the same. Compare churn rates by acquisition channel, onboarding path, and company size to find patterns.
+5. **Interview churned customers** — The customers who left quietly are more valuable than the ones who complained. Call 10 churned accounts per month and ask what they were trying to accomplish.
+```
+
+**Schema markup (JSON-LD):**
+```json
+{
+  "@context": "https://schema.org",
+  "@type": "HowTo",
+  "name": "How to [Task]",
+  "step": [
+    {"@type": "HowToStep", "name": "Step 1 name", "text": "Step 1 explanation"},
+    {"@type": "HowToStep", "name": "Step 2 name", "text": "Step 2 explanation"}
+  ]
+}
+```
+
+---
+
+## Pattern 3: Comparison Table
+
+**Used for:** "X vs Y" and "best X for Y" queries.
+
+**Requirements:**
+- Header row with category names
+- First column: feature or criterion
+- Remaining columns: the things being compared
+- Keep it focused — 5-10 rows maximum
+- Don't try to cover everything; cover what matters most
+
+**Template:**
+```markdown
+| Feature | [Option A] | [Option B] | [Option C] |
+|---|---|---|---|
+| [Criterion 1] | [Value] | [Value] | [Value] |
+| [Criterion 2] | [Value] | [Value] | [Value] |
+| [Criterion 3] | [Value] | [Value] | [Value] |
+| Best for | [Audience A] | [Audience B] | [Audience C] |
+| Pricing | [Range] | [Range] | [Range] |
+```
+
+**Tips:**
+- Put the most important criteria first
+- Use simple values — "Yes / No / Partial" beats long prose in cells
+- Include a "Best for" row — AI systems use this for recommendation queries
+- Add a sentence below the table summarizing the verdict: "X is best for teams that need A; Y is better when B matters more."
+
+---
+
+## Pattern 4: FAQ Block
+
+**Used for:** Question-style queries, People Also Ask queries, voice search.
+
+**Requirements:**
+- Question phrased exactly as someone would ask it (natural language)
+- Answer is complete in 2-4 sentences (no "read more in section 3")
+- 5-10 FAQs per block
+- Pair with FAQPage schema markup
+
+**Template:**
+```markdown
+## Frequently Asked Questions
+
+**What is [X]?**
+[2-4 sentence complete answer]
+
+**How does [X] work?**
+[2-4 sentence complete answer]
+
+**What's the difference between [X] and [Y]?**
+[2-4 sentence complete answer]
+
+**How much does [X] cost?**
+[2-4 sentence complete answer]
+
+**Is [X] right for [audience]?**
+[2-4 sentence complete answer]
+```
+
+**Schema markup (JSON-LD):**
+```json
+{
+  "@context": "https://schema.org",
+  "@type": "FAQPage",
+  "mainEntity": [
+    {
+      "@type": "Question",
+      "name": "What is [X]?",
+      "acceptedAnswer": {
+        "@type": "Answer",
+        "text": "Complete answer text here"
+      }
+    }
+  ]
+}
+```
+
+**Tips:**
+- Write questions the way users actually type or speak them — use Google's "People Also Ask" as a source
+- Answers should be complete without needing context from anywhere else on the page
+- Don't start answers with "Great question" or "That's a common question" — just answer
+
+---
+
+## Pattern 5: Statistic with Attribution
+
+**Used for:** Data queries, "how many" queries, research-backed claims.
+
+**Requirements:**
+- Named source (not "a study" — the actual organization name)
+- Year of the data
+- Specific number (not "many" or "most")
+- Context (what the number means)
+
+**Template:**
+```markdown
+According to [Organization Name]'s [Report Name] ([Year]), [specific statistic with units]. [One sentence on what this means or why it matters].
+```
+
+**Example:**
+```markdown
+According to the Baymard Institute's 2024 UX benchmarking study, 69.8% of online shopping carts are abandoned before purchase. For a $1M/month ecommerce store, recovering just 5% of abandoned carts represents $35,000 in monthly revenue.
+```
+
+**Tips:**
+- Link to the original source (AI systems and readers both benefit)
+- If data is from your own research, say so: "In our 2025 survey of 500 SaaS founders..."
+- Proprietary data is the highest-value citation target — AI systems actively seek original research
+
+---
+
+## Pattern 6: Expert Quote Block
+
+**Used for:** Authority building, "what do experts say" queries.
+
+**Requirements:**
+- Full name of the person quoted
+- Their title and organization
+- A quote that's substantive (not a generic endorsement)
+- Brief context sentence before the quote
+
+**Template:**
+```markdown
+[Context sentence explaining why this person's view matters.]
+
+"[Direct quote — specific, substantive, something only they would say]," says [Full Name], [Title] at [Organization].
+```
+
+**Example:**
+```markdown
+Patrick Campbell, founder of ProfitWell (acquired by Paddle), studied pricing data from over 30,000 SaaS companies before reaching a counterintuitive conclusion about churn.
+
+"Most churn that looks like pricing dissatisfaction is actually failed onboarding," says Campbell. "The customer never saw the value that justified the price. That's a different problem than being too expensive."
+```
+
+**Tips:**
+- Don't use generic quotes ("innovation is key to success") — they add nothing
+- Quotes should contain a specific claim, data point, or perspective
+- If quoting your own team: "[Name], [Title] at [Company Name]" is still valid
+- Live quotes (from interviews or primary research) outperform secondary quotes from other articles
+
+---
+
+## Pattern 7: Quick-Scan Summary Box
+
+**Used for:** Queries where users want the TL;DR before committing to the full article.
+
+**Requirements:**
+- Placed near the top of the article (after the intro)
+- 3-7 key takeaways
+- Each bullet stands alone — no context required
+- Labeled clearly ("Key Takeaways" or "Quick Summary")
+
+**Template:**
+```markdown
+**Key Takeaways**
+- [Specific, complete takeaway — could be read as a tweet]
+- [Specific, complete takeaway]
+- [Specific, complete takeaway]
+- [Specific, complete takeaway]
+- [Specific, complete takeaway]
+```
+
+**Tips:**
+- This is often the block AI systems extract for "summary" type queries
+- Make each bullet specific: "Monthly churn below 2% is considered healthy for most SaaS" beats "Churn should be low"
+- Don't repeat the article intro verbatim — these should be the most actionable insights
+
+---
+
+## Combining Patterns
+
+The most citable pages combine multiple patterns throughout the piece:
+
+**Recommended page structure for maximum AI extractability:**
+1. Definition block (first 300 words)
+2. Quick summary box (right after intro)
+3. Body sections with numbered steps or subsections
+4. Data points with full attribution throughout
+5. Comparison table (if competitive topic)
+6. FAQ block (before conclusion)
+7. Expert quote (to add authority)
+
+A page with all 7 patterns has significantly more extractable surface area than a page with prose only. The AI has more options to pull from and a higher probability of finding something that perfectly matches the query.