borghei/Claude-Skills / marketing/aeo
aeo
VerifiedAnswer Engine Optimization (AEO): optimize content to be cited by LLMs (ChatGPT, Claude, Perplexity, Gemini) in their answers. Use when designing content for LLM citation, auditing citability, or structuring Q&A schema.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 13.8 KB |
| references/aeo-fundamentals.md | markdown | 10.4 KB |
| references/citation-tracking-and-measurement.md | markdown | 10.8 KB |
| references/llm-content-structuring.md | markdown | 12.4 KB |
| scripts/aeo_content_auditor.py | script | 11.1 KB |
| scripts/citation_extractor.py | script | 9.0 KB |
| scripts/schema_qa_generator.py | script | 7.4 KB |
references/llm-content-structuring.md
View on GitHub# LLM Content Structuring
Reference for structuring content to maximize LLM citation: the 5 citable content patterns in depth, FAQ / QAPage / HowTo schema, citation hooks, voice-search optimization, structure markers.
---
## The 5 citable content patterns (deep)
### Pattern 1: Definitional content
LLMs respond to "What is X?" queries with definitions, then cite the source for the definition. To be the cited source:
**Format:**
```
[Term] is [crisp definition in 1-2 sentences].
[Elaboration with 1-3 paragraphs: history, context, who uses it, why it matters].
## How [Term] works
[Mechanism / process explanation]
## When to use [Term]
[Use cases]
## When NOT to use [Term]
[Anti-patterns / limitations]
## [Term] vs [Related concept]
[Comparative content]
```
**LLM extraction patterns:**
- First 1-2 sentences become the definition LLM reproduces
- Subsequent sections answer follow-up questions ("how does X work")
- "When NOT to use" content helps LLM nuance the answer
**Anti-patterns:**
- Burying the definition behind narrative intro
- Generic definitions ("X is a method...")
- Definitions that require reading 3 paragraphs
### Pattern 2: Comparative tables
LLMs love tables. When users ask "X vs Y" or "best X," tables get cited.
**Format:**
```markdown
| Feature | Product A | Product B | Product C |
|---------|-----------|-----------|-----------|
| Price | $X/mo | $Y/mo | $Z/mo |
| Speed | 50ms | 75ms | 30ms |
| Best for | Small teams | Enterprise | Solo |
| Free tier | Yes (limited) | No | Yes |
| Support | Email | 24/7 chat | Self-serve |
| Integration count | 50+ | 200+ | 20 |
```
**LLM extraction patterns:**
- Reproduces table verbatim
- Cites your page as the comparison source
- Sometimes generates "Source: [your site]" attribution
**Anti-patterns:**
- Tables in images (not extractable)
- Tables with inconsistent columns
- Tables with marketing-claim columns ("amazing", "best in class")
- Single-column tables (not actually comparisons)
### Pattern 3: Step-by-step procedural content
"How to [task]" content. LLMs reproduce steps; cited as authoritative.
**Format:**
```
# How to [task]
[1-sentence intro: what this accomplishes, prerequisites]
## Step 1: [Action]
[Detail; 1-3 sentences explaining what + why]
## Step 2: [Action]
[Detail]
## Step 3: [Action]
[Detail]
[Continue for all steps]
## What to do if [common error]
[Troubleshooting]
```
**LLM extraction patterns:**
- Reproduces numbered steps
- Cites your page as procedure source
- Sometimes condenses if very long
**Anti-patterns:**
- Steps with implicit ordering ("first you'll need to..." in prose)
- Missing step numbers
- Steps that combine multiple actions
- Procedures behind login walls
### Pattern 4: Statistics + data with attribution
"According to [your study], X% of Y..." LLMs cite sources for specific statistics.
**Format:**
```
[Specific statistic with context]. Source: [your research / study / data analysis], conducted [date], based on [methodology].
For example:
"54% of B2B buyers research vendors via LLMs in 2026, up from 12% in 2024. (Source: [study name] by [your brand], published Feb 2026, based on survey of 1,200 B2B buyers across 12 industries.)"
```
**LLM extraction patterns:**
- Statistic reproduced verbatim
- Source attribution included if explicit
- Higher citation rate for data with credible methodology
**Anti-patterns:**
- Statistics without source ("studies show...")
- Methodology not specified
- Old data (LLMs prefer recent)
- Statistics in images / charts (not extractable without alt text)
### Pattern 5: Lists with explanations
"Top N [things]" content. LLMs reproduce lists when users ask comparative or enumeration questions.
**Format:**
```
# Top [N] [things] for [purpose]
[Intro: 1-2 paragraphs about the category + selection criteria]
## 1. [Item Name]
[2-3 sentences: what it is, why it ranks #1, who should use]
## 2. [Item Name]
[Same structure]
[Continue for N items]
## How we evaluated
[Methodology — adds credibility]
```
**LLM extraction patterns:**
- Reproduces top 3-5 items
- Cites your page as source of ranking
- Sometimes filters / re-orders based on user-specific context
**Anti-patterns:**
- Lists without explanations (just names)
- Inconsistent rationale per item
- Lists without selection criteria
---
## FAQ / QAPage / HowTo schema (JSON-LD)
Structured data can help machines classify Q&A content, but treat it as optional and low impact (as of September 2026):
- Google retired HowTo rich results (Sept 2023) and FAQ rich results (May 2026); the markup is still valid schema.org but earns no Google search feature.
- Google states that no special schema.org markup is required to appear in AI Overviews or AI Mode ([AI features and your website](https://developers.google.com/search/docs/appearance/ai-features)).
- No major LLM provider documents FAQ/HowTo markup as a citation signal. The visible Q&A and step structure is what gets extracted.
If you still add it, keep it accurate and matching visible content. Three relevant schemas:
### FAQPage schema
For pages with a list of questions and answers (e.g., FAQ page, knowledge base article with FAQ section).
```json
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
{
"@type": "Question",
"name": "What is X?",
"acceptedAnswer": {
"@type": "Answer",
"text": "X is..."
}
},
{
"@type": "Question",
"name": "How does X work?",
"acceptedAnswer": {
"@type": "Answer",
"text": "X works by..."
}
}
]
}
```
### QAPage schema
For pages that are primarily one Q&A (e.g., Stack Overflow-style question pages, individual help articles).
```json
{
"@context": "https://schema.org",
"@type": "QAPage",
"mainEntity": {
"@type": "Question",
"name": "How do I [task]?",
"acceptedAnswer": {
"@type": "Answer",
"text": "To accomplish this, ..."
}
}
}
```
### HowTo schema
For procedural content with steps.
```json
{
"@context": "https://schema.org",
"@type": "HowTo",
"name": "How to [task]",
"description": "Procedure for [task].",
"step": [
{"@type": "HowToStep", "name": "Step 1", "text": "First, ..."},
{"@type": "HowToStep", "name": "Step 2", "text": "Then, ..."}
]
}
```
### Where to place schema
- In `<head>` as `<script type="application/ld+json">`
- One schema per page (combining is allowed but cleaner separate)
- Don't duplicate the visible content meaning in schema; describe the structure
### Schema verification
Use Google's Rich Results Test (or Schema.org validator) to verify JSON-LD parses correctly. Errors are silent failures (LLMs / search ignore broken schema).
---
## Citation hooks (in-content patterns)
Specific in-content patterns that increase citation likelihood:
### Hook 1: Explicit "according to [source]" pattern
```
According to [your study], 73% of users prefer X.
```
When LLM cites this content, the "according to..." pattern transfers as attribution.
### Hook 2: Bold key terms / claims
```
The recommended approach is **X over Y because of Z**.
```
Bold makes claims extraction-friendly.
### Hook 3: Pull-quotes / callouts
```
> Key insight: 80% of [outcome] depends on [factor].
```
LLMs often extract callouts directly.
### Hook 4: Inline citations to authoritative sources
```
This pattern is documented in [authoritative source link].
```
When LLM cites you, the citation chain helps establish your authority.
### Hook 5: Update dates visible
```
**Last updated: 2026-05-01**
```
Recency increases citation rate for time-sensitive topics.
---
## Voice-search and conversational query optimization
LLM queries are often conversational ("How do I...", "What's the difference between..."). Optimize content to match conversational query patterns:
### Match conversational query templates
| Query template | Content structure |
|----------------|-------------------|
| "What is X?" | Definitional content |
| "How does X work?" | Mechanism / process explanation |
| "How do I X?" | Step-by-step procedure |
| "Why does X happen?" | Causal / explanatory content |
| "When should I use X?" | Use-case content |
| "X vs Y" | Comparative table + explanation |
| "Best X for Y" | List with criteria |
| "Top N X" | Ranked list |
Title pages / sections matching these patterns directly.
### Long-tail conversational queries
Traditional SEO: target high-volume keywords.
AEO: also target long-tail conversational queries (3-7 word patterns).
Example:
- SEO target: "data quality" (high volume)
- AEO target: "how do I improve data quality in my data warehouse" (specific, conversational)
Long-tail conversational has lower volume but higher LLM-citation rate.
---
## Structure markers LLMs use
LLMs use specific markers to extract content. Use them consistently:
### Headings (H1, H2, H3, ...)
- One H1 per page (the main title)
- H2 for major sections
- H3 for sub-sections
- Don't skip levels (H1 → H3 without H2)
LLMs use heading hierarchy to understand content structure.
### Ordered vs unordered lists
- Ordered (numbered): use for sequential steps, ranked items
- Unordered (bullets): use for parallel items without order
LLMs preserve the distinction.
### Tables
- Use Markdown table syntax (or HTML `<table>`) — extractable
- First row = headers
- Consistent columns per row
- One table per concept (avoid sprawling tables)
### Code blocks
- Use fenced code blocks with language hints (```python)
- LLMs reproduce code from properly-fenced blocks
- Inline code with backticks for short references
### Definition lists (Markdown via `:` or HTML `<dl>`)
Underused but powerful for definitional content:
```html
<dl>
<dt>Term</dt>
<dd>Definition</dd>
</dl>
```
LLMs extract dt/dd pairs efficiently.
---
## Content length
| Content type | Optimal length |
|--------------|----------------|
| Definitional content | 500-1500 words |
| How-to guide | 800-2500 words |
| Comparative analysis | 1500-3500 words |
| Statistics report | 1000-3000 words |
| Listicle (top N) | 1500-3500 words |
Too short = thin content; too long = LLM may truncate or miss key sections.
---
## Multimedia and LLM extraction
LLMs (as of 2026) extract from text well; multimedia partially:
| Media | LLM extraction | Recommendation |
|-------|----------------|----------------|
| Plain text | Excellent | Use as primary content |
| Markdown / HTML tables | Excellent | Use freely |
| Bullet / numbered lists | Excellent | Use freely |
| Headings | Excellent | Use for structure |
| Code blocks | Excellent (with language hints) | Use freely |
| Images | Limited (depends on alt text + nearby caption) | Always include alt text |
| Charts / data visualizations | Limited | Pair with data table + alt text |
| Videos | Limited (some LLMs can transcribe; not for citation) | Pair with transcript |
| Embeds (Twitter, YouTube) | Variable | Don't rely on embed content for citation |
| PDF content | Limited (some LLMs read PDFs; many can't) | Publish key content in HTML, not PDF-only |
| Tables in images | Cannot extract | Always render tables in HTML / Markdown |
---
## Updating content for AEO
Existing content audit checklist:
- [ ] Does the page answer a specific question?
- [ ] Is there a 1-2 sentence definition / direct answer at the top?
- [ ] Are there headings (H2 / H3) for sub-topics?
- [ ] Are key claims bolded?
- [ ] Are there extractable tables, lists, or step-by-steps?
- [ ] Are statistics attributed?
- [ ] Is there an FAQ section?
- [ ] (Optional) Is FAQ schema valid and matching visible Q&A?
- [ ] Is the last-updated date visible?
- [ ] Is author attribution visible?
- [ ] Are images alt-texted?
- [ ] Are PDFs paired with HTML equivalents?
- [ ] Does the URL include relevant keywords?
Run through this for each priority page.
---
## Cheat sheet
| Question | Answer |
|----------|--------|
| Most impactful single change? | Add 1-2 sentence definition / direct answer at top of page |
| Add FAQ schema everywhere? | No — optional and low impact; only where visible Q&A exists. Visible structure matters more |
| Length sweet spot? | 1500-2500 words for most content types |
| Bullet vs numbered? | Numbered for sequential; bullet for parallel |
| Tables in images? | Never. Always HTML / Markdown |
| PDF-only content? | Don't. HTML or HTML+PDF |
| Update dates visible? | Yes; especially for time-sensitive content |
| Inline keywords still matter? | Less than 5 years ago, but still relevant |
| Long-tail or head terms? | Both; long-tail favored for AEO |
| Match conversational query templates? | Yes; title pages / sections directly |
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |