Project Manager

Answer Engine Optimization (AEO)

End-to-end practice of optimizing content to be cited by LLMs when they generate answers. Covers the technical foundations (how LLMs select sources), content structuring patterns (Q&A schema, citation-worthy patterns), measurement (which content gets cited, by which LLM, how often), and the strategic positioning that differentiates AEO from traditional SEO and from AI-SEO.

This skill is provider-aware but provider-agnostic: works for content optimized for ChatGPT, Claude, Perplexity, Gemini, Copilot, and emerging AI surfaces.

---

When to use this skill

---

AEO vs SEO vs AI-SEO

Three distinct (but overlapping) practices. Confusing them leads to wasted investment.

Strategic positioning

For most B2B brands:

• Traditional SEO: still 50-70% of organic traffic. Don't abandon.
• AI-SEO: emerging 10-20% of search-driven engagement. Growing fast.
• AEO: 5-15% of LLM-mediated user discovery. Largest growth potential.

Optimize content for all three simultaneously; the techniques substantially overlap.

---

The AEO funnel

Users find brands through LLMs in a different funnel than search:

Traditional search:           AEO funnel:
1. User types query           1. User asks LLM a question
2. SERPs show ~10 results     2. LLM generates answer
3. User clicks one            3. LLM cites N sources (1-10)
4. User reads page            4. User reads answer; may click cited source
5. User converts              5. User attributes answer to LLM (less so to cited brand)

Key implications:

• Citation is the new click. When LLM cites your content, you don't always get a visit — but you get attribution.
• Brand-as-source becomes the goal. Even without click, being cited builds brand association.
• Quality > volume. LLMs cite a small number of sources; quality of citation matters more than ranking position.
• Trust signals matter more. LLMs avoid citing low-authority sources.

See references/aeo-fundamentals.md for the deep mechanics of how LLMs select sources, the citation models per provider, and the trust signals that drive selection.

---

The 5 content patterns that get cited

After analysis of LLM citation behavior, five content patterns dominate:

Pattern 1: Definitional content with clear claims

LLMs cite sources for definitions, facts, and short claims. Pages that answer "What is X?" with a clean 2-3 sentence definition followed by elaboration get cited often.

Structure:

[Term] is [crisp definition in 1-2 sentences].

[Elaboration with context and nuance — 1-3 paragraphs].

[Related concepts / scope / boundaries — optional].

Pattern 2: Comparative tables

LLMs use tables to extract comparisons. Markdown tables in published content (or HTML equivalents) get cited when users ask "X vs Y."

| Feature | Product A | Product B |
|---------|-----------|-----------|
| Price | $X | $Y |
| Speed | Z ms | W ms |
| Support | 24/7 | Business hours |

Pattern 3: Step-by-step procedural content

"How to [task]" content with explicit numbered steps. LLMs reproduce procedural steps; the cited source becomes the authoritative reference.

Pattern 4: Statistics + data with sources

LLMs cite content that provides numerical facts with attribution. "According to [your study], X% of [thing] does Y" is repeatable and citable.

Pattern 5: Lists with explanations

"Top N approaches to X" with each item explained gets cited when users ask comparative or enumeration questions.

See references/llm-content-structuring.md for deep patterns including FAQ schema, citation hooks, voice-search optimization, and LLM-readable structure markers.

---

Clarify First

Before generating, confirm these inputs. If any is unknown or vague, ASK — do not assume:

• [ ] Target queries — the actual questions customers ask LLMs about your category (drives which content to audit and restructure)
• [ ] Your brand name — exact wording to track in answers vs competitors (drives citation extraction)
• [ ] Target LLM surface — ChatGPT / Claude / Perplexity / Gemini (citation behavior and trust signals differ per provider)
• [ ] Canonical page/content — the high-value page to be the authoritative source (drives schema generation + pattern restructuring)

Stop rule: ask only the 2-3 that most change the output. If the user says "just draft it," proceed and list your assumptions at the top of the artifact.

Quick start

1. Audit existing content: python3 scripts/aeo_content_auditor.py --path ./content 2. Add Q&A schema to high-value pages: python3 scripts/schema_qa_generator.py --content article.md 3. Track citations from competitors: python3 scripts/citation_extractor.py --query "What is X?" --brand "Your Brand" 4. Iterate: monthly content review with AEO scoring

---

End-to-end workflows

Workflow: AEO content strategy from scratch

1. Identify target queries — what questions do potential customers ask LLMs about your category? 2. Audit competitor citations — which brands get cited for those queries? scripts/citation_extractor.py 3. Audit your existing content — score current content for AEO patterns: scripts/aeo_content_auditor.py 4. Prioritize 10-20 high-value pages — those that should be the canonical source 5. Restructure per AEO patterns — definitional content, tables, step-by-step, statistics 6. Add structured data — scripts/schema_qa_generator.py generates FAQ schema 7. Build authority signals — backlinks, citations, mentions 8. Monitor monthly — track citation rate trend

Workflow: Audit individual content piece

1. Run scripts/aeo_content_auditor.py --path article.md --format markdown 2. Review per-pattern scoring (5 patterns above) 3. Identify gaps: missing definition, no table, no clear steps, no stats, no list 4. Restructure to add 2-3 missing patterns 5. Add FAQ schema with scripts/schema_qa_generator.py 6. Re-audit to confirm improvements

Workflow: Competitive citation analysis

1. Identify 10-20 key queries in your category 2. Query each LLM (ChatGPT, Claude, Perplexity, Gemini) with those questions 3. Record citations + brands mentioned 4. Analyze: which brands dominate? what content do they have? 5. Identify white-space queries (no clear dominant source yet) 6. Prioritize content creation for white-space queries

Workflow: Measure AEO performance

1. Citation rate: % of queries where your brand is cited (target: 30%+ for category leaders) 2. Brand mention rate: % of queries where your brand is mentioned (cited or not) 3. Source quality: are you cited as primary source or supporting? 4. Click-through from citations: traffic attributable to LLM citations (requires source tracking) 5. Voice tracking: how is your brand characterized (positive / neutral / negative attributes)

See references/citation-tracking-and-measurement.md for measurement methodologies, attribution challenges, and competitive benchmarking.

---

Common AEO failures

• Optimizing only for Google SERP: misses the LLM citation surface entirely
• Generic content without specific claims: LLMs prefer specific, factual content over generic explanation
• No structure markers (headings, lists, tables): LLMs can't extract specific information
• No FAQ schema: missed opportunity for Q&A surfacing in AI Overviews
• Stuffed keyword content: LLMs prefer natural language with clear meaning
• No authority signals: LLMs avoid citing low-trust sources
• Outdated content: LLMs prefer recent, current content
• Hidden behind paywalls: LLMs can't cite what they can't access
• No structured data: missed opportunity for richer extraction
• Brand-first content: LLMs prefer informational content over promotional

---

LLM-by-LLM citation behavior

Different LLMs have different citation behaviors:

Optimize content with structure markers (headings, lists, tables) and authority signals (links, citations, expert attribution) — works across all of these.

---

Tooling

---

References

• aeo-fundamentals.md — how LLMs select sources; citation mechanisms per provider; trust signals
• llm-content-structuring.md — content patterns; Q&A schema; voice-search; structure markers
• citation-tracking-and-measurement.md — measurement methodologies; attribution; benchmarking

---

Related skills

• marketing/ai-seo — AI search engine ranking (Perplexity, Google AI Overviews); complementary to AEO
• marketing/seo-specialist — traditional SEO (Google rankings); foundational; still 50-70% of organic
• marketing/seo-audit — technical SEO audit
• marketing/programmatic-seo — scaled content production with SEO patterns
• c-level-advisor/cs-cmo-advisor — strategic AEO investment decisions

AEO Fundamentals

Deep reference on how LLMs select sources to cite, citation mechanisms per provider, the trust signals that drive selection, and the strategic positioning of AEO relative to traditional SEO and AI-SEO.

---

How LLMs select sources

Modern LLMs use one of three citation mechanisms:

1. Training-time citation (implicit)

Sources mentioned in training data become part of the model's parametric knowledge. The model can reproduce facts without explicitly citing source. Brand mentions accumulated over years of training-data presence shape what the model knows.

Implications:

• Long-tail brand presence in high-quality sources (Wikipedia, major publications, academic) matters even without per-query citation
• New brands have a cold-start disadvantage
• Content authored in 2024+ may not appear until next training cycle (typically 6-18 month lag)

2. Retrieval-time citation (RAG / web-enabled)

When the LLM has web access (ChatGPT browsing, Claude with web search, Perplexity always, Gemini with grounding), it retrieves current sources and cites them in the answer.

Selection signals (varies by provider):

• Search ranking (uses upstream search engine: Google / Bing typically)
• Source authority (domain reputation)
• Content recency
• Content relevance to the specific query
• Structured data presence (helps extraction)
• Citation quality from prior interactions

Implications:

• Traditional SEO matters — LLMs use search results as input
• Content structure matters — extractable content gets cited more
• Authority matters — established domains preferred over new

3. Tool-mediated citation

When LLMs use specific tools (Anthropic's Computer Use, OpenAI function calls, etc.), citations come from tool outputs. Less common for typical user queries.

---

What makes content citable

Beyond search ranking, specific content properties increase citation rate:

Property 1: Specificity over generality

LLMs cite specific claims. "The recommended ratio is 1:7" is more citable than "the ratio depends on context."

Property 2: Extractable structure

LLMs can extract:

• Numbered lists
• Tables with clear headers
• Q&A pairs
• Definitions with clear subject-verb-object structure
• Step-by-step procedures

LLMs cannot reliably extract:

• Narrative prose with implicit structure
• Information embedded in images without alt text
• Content in PDFs (most LLMs can't read PDFs natively)
• Tweet / social-media-style fragments

Property 3: Citation-friendly URL structure

LLMs cite URLs. Good URL structure:

• Permanent / stable
• Human-readable (slug includes topic)
• Accessible (no login wall; no JS-only rendering)
• Indexed (in Google's index = likely in LLM's reach)

Property 4: Author + publication date visible

LLMs prefer cite-able sources with author attribution + date. Avoid:

• Anonymous content
• Hidden dates
• "Last updated" without specific date

Property 5: Authority signals

• Domain authority (Ahrefs DR, Moz DA — proxies)
• Inbound links from other authoritative domains
• Wikipedia presence
• Citations in academic / news sources
• Industry recognition (analyst reports referencing brand)

---

Trust signals (what LLMs avoid)

LLMs (their training and grounding pipelines) systematically de-prioritize:

• Spammy domains
• Aggressive ads
• Thin content (low word count, low information density)
• AI-generated content without human review (yes, ironic)
• Content with high engagement-bait patterns
• Sources flagged for misinformation
• Hate speech / extreme content
• Affiliate-link farms
• Outdated content for time-sensitive topics
• Content behind paywalls (mostly; some LLMs handle some paywalls)

---

Citation mechanisms per provider

ChatGPT (OpenAI)

Without browsing: Cites primarily from training-time knowledge. Citations are implicit (brand mentions appear in answers, but no source links).

With browsing/search: Cites with inline links when answer requires recent / specific information. Browsing typically engaged for time-sensitive or specific factual queries.

SearchGPT (when available): More explicit citation, similar to Perplexity model.

Claude (Anthropic)

Native: Tends to avoid citing without evidence; less likely to provide specific URL citations than ChatGPT.

With tools / grounding: Cites sources when web search / retrieval is part of the response.

Quality bias: Strong preference for high-quality, evidence-based sources. Lower tolerance for low-authority citations.

Perplexity

Citation-first: Every answer includes source citations prominently displayed. Multiple sources per answer (typically 4-10).

Search-grounded: All answers run through search before answering. Heavy use of recent web content.

Implications for AEO: Perplexity is the most-amenable to AEO investment. If you rank in their search, you get cited.

Google Gemini / AI Overviews

AI Overviews (in Google Search): Inline citations to source pages. Selected for queries where AI Overview is shown (mostly informational queries).

Gemini conversational: Cites sources when relevant; web-grounded by default in many use cases.

Implications: Google AI Overviews are heavily SEO-driven. If you rank top 10 in Google, you have a shot at AI Overview citation.

Microsoft Copilot

Bing-grounded: Most answers grounded in Bing search results. Citations prominently displayed.

SharePoint / enterprise: Internal sources cited for enterprise Copilot.

Meta AI

Limited transparency: Citations less consistent. Brand presence in training data more important than retrieval-time.

---

Citation distribution patterns

Observed patterns in LLM citations:

---

Strategic AEO positioning

When AEO matters most

When NOT to invest in AEO

• Pre-PMF brands without informational content: AEO requires content investment that may not return
• Hyperlocal businesses: Google Maps still dominates local queries
• Pure transactional commerce: "Buy X" queries less LLM-mediated than "Research X" queries
• Highly regulated content that can't make specific claims: AEO benefits from specificity

AEO investment levels

---

How LLMs evolve (and what to track)

LLM citation behavior changes:

• New training cycles → updated parametric knowledge
• New search providers → different retrieval-time citation
• New product features → AI Overviews, Perplexity Pro, etc.

Track:

• LLM market share (which to optimize for)
• New citation surfaces (AI Overviews, Perplexity Spaces, etc.)
• Provider policy changes (training data opt-outs, paywall handling)
• Industry citation patterns in your category

---

Common AEO misconceptions

Misconception 1: "AEO = AI-generated content"

False. AI-generated content is often de-prioritized. AEO is about content STRUCTURE and AUTHORITY, not authorship.

Misconception 2: "AEO is just SEO with new keywords"

False. AEO requires structure (tables, lists, Q&A), specificity (claims, statistics), and authority (trust signals) beyond keyword optimization.

Misconception 3: "If we rank #1 in Google we'll be cited by LLMs"

Mostly true for Perplexity + Google AI Overviews. Less true for ChatGPT / Claude where training-time mentions matter.

Misconception 4: "AEO replaces SEO"

False. SEO is still 50-70% of traffic. AEO adds another channel.

Misconception 5: "We can measure AEO precisely like SEO"

False. AEO measurement is harder. Citations don't always link back; brand mentions in LLM answers are hard to detect; provider attribution varies.

Misconception 6: "Just add Q&A schema and we're optimized"

Partly. Q&A schema helps extraction; doesn't substitute for content structure + authority.

---

AEO and brand voice

AEO content has a tension with brand voice:

• AEO content prefers: factual, structured, specific
• Brand voice often: personal, narrative, distinctive

Balance:

• Brand pages (homepage, about) stay brand-voice
• Knowledge content (blogs, guides, docs) goes AEO-structured
• Product pages mix (top: brand; bottom: structured specs / FAQ)

---

Cheat sheet

Citation Tracking and Measurement

Reference for measuring AEO performance: citation rate, brand mention rate, sentiment, attribution challenges, competitive benchmarking, dashboard design.

---

What to measure

Primary AEO metrics

Secondary metrics

---

Citation extraction methodology

Manual method (small scale, qualitative)

1. Identify target queries — list of 20-50 queries in your category 2. Query each LLM — submit each query to ChatGPT, Claude, Perplexity, Gemini, Copilot 3. Capture full response — text + citations (where shown) 4. Tag for analysis — brand mentions, source URLs, sentiment

Pros: Captures full context; sentiment analysis possible Cons: Doesn't scale; sample-based; LLM responses vary across queries

Semi-automated method (medium scale)

1. API-based querying — use LLM APIs to programmatically query 2. Parse responses — extract citations + brand mentions 3. Store in DB — query / LLM / response / citations 4. Dashboard — citation rate over time

Pros: Scales to hundreds of queries; consistent capture Cons: API costs; doesn't capture consumer-LLM behavior (no API access for ChatGPT-app, only API responses)

Third-party tracker tools

Emerging category of tools that track brand mentions in LLMs:

• Profound, Otterly, Athena, others
• Subscribe-based; track your brand vs competitors
• Provide dashboards + alerts

Best for: ongoing tracking without DIY infrastructure.

---

Attribution challenges

AEO attribution is harder than SEO attribution:

Challenge 1: No referrer

When a user clicks a citation in an LLM response, the referrer is often the LLM domain (chat.openai.com, claude.ai, perplexity.ai) — not preserving the originating query.

Workaround:

• UTM parameters on cited URLs (if you control them)
• Cohort analysis: traffic from LLM domains
• Survey users ("How did you hear about us?" → "AI assistant / ChatGPT")

Challenge 2: No-click citations

Most LLM citations don't generate a click. Brand mention exists; web traffic doesn't change.

Workaround:

• Measure brand-mention rate (not just citation-driven traffic)
• Brand-awareness lift survey (annual)
• Direct-traffic increase as proxy

Challenge 3: LLM response variability

Same query, multiple responses, different citations. Statistical significance requires multiple samples per query.

Workaround:

• Query each LLM 3-5x per target query
• Use median citation rate, not single-instance

Challenge 4: Training-time vs retrieval-time

When LLM cites without web grounding, the brand mention comes from training data. Hard to attribute to specific content investment.

Workaround:

• Track brand mentions in long-term training-data-proxy sources (Wikipedia, major publications)
• Distinguish "with browsing" vs "without browsing" citations

Challenge 5: Sentiment / characterization variability

LLM may cite you but characterize you neutrally or negatively. Citation alone insufficient.

Workaround:

• Capture full context, not just URL
• Sentiment classification per citation
• Track narrative shifts over time

---

Sampling methodology

Query selection

Pick queries representative of your category:

• Definitional queries: "What is X?"
• Comparison queries: "X vs Y"
• Best-of queries: "Best X for Y"
• How-to queries: "How to X"
• Recommendation queries: "Should I use X?"
• Brand-specific queries: "Is [brand] worth it?", "[brand] review"

Aim for 50-200 queries per category.

Sampling frequency

• Daily: for time-sensitive industries (news, finance)
• Weekly: standard
• Monthly: for slow-moving categories

LLM sampling

Sample across LLMs in proportion to your audience's usage:

Adjust per actual audience usage (verify via survey).

---

Competitive benchmarking

Share of voice (SOV) in citations

SOV (your brand) = your citations / total citations across all brands × 100%

For each target query:

• List all brands cited
• Calculate share for each
• Aggregate across queries

Healthy SOV by stage:

• Category leader: 25-40% SOV
• Top 3: 10-25%
• Growing: 5-15%
• Emerging: < 5%

Competitor citation analysis

For competitors that out-cite you:

• Which queries do they win?
• What content do they have for those queries?
• What's the structure / format?
• What's their authority profile?

Compete by:

• Building equivalent content (better structured)
• Targeting white-space queries (no clear leader yet)
• Building authority through earned media

Share of citations by content type

Track which content types drive your citations:

• Blog posts (informational)
• Documentation (technical)
• Product pages (transactional)
• Research / reports (data-driven)
• Customer stories (social proof)

Helps allocate content investment.

---

Sentiment analysis

For each citation, classify sentiment:

Detection methods

• Manual: human reviewers tag citations (small samples)
• LLM-assisted: use LLM to classify sentiment of citations (medium scale)
• Automated: NLP-based sentiment analysis (large scale; less accurate for nuance)

Target: > 80% positive/neutral; investigate negative citations (often based on outdated info or specific incidents).

---

Dashboard design

Daily / weekly view

• Citation rate trend (line chart, last 90 days)
• Top 10 queries cited for (table)
• New negative citations (alert)
• LLM-by-LLM distribution (bar chart)

Monthly view

• Citation rate by category
• Share of voice trend
• Content type contribution
• Sentiment distribution
• Competitive comparison

Quarterly view

• AEO ROI estimate
• Content investment vs citation lift
• Strategic recommendations

---

Tooling stack (recommended)

---

Common measurement failures

Failure 1: Measuring only what's easy

E.g., only counting Perplexity citations (most explicit) and ignoring ChatGPT / Claude (less explicit).

Failure 2: Single-LLM bias

Optimizing for one LLM (often ChatGPT) and missing where audience actually queries.

Failure 3: Conflating mention with citation

Brand mentioned in LLM response = mention. Brand-source URL referenced = citation. Different metrics.

Failure 4: Ignoring sentiment

Citation count up, but negative? Worse than no citation. Always measure sentiment.

Failure 5: Not segmenting by query intent

Branded queries (asking about your brand) ≠ category queries (asking about your space). Citations from each have different value.

Failure 6: Over-relying on direct-traffic proxy

Direct traffic increase has many causes; LLM attribution requires more rigorous methodology.

---

ROI estimation

Estimate AEO ROI:

AEO ROI = (Estimated brand-awareness lift × Brand value per impression × Citation count)
          - AEO investment cost

Approximations:

• Brand-awareness lift per citation: typically equivalent to a low-quality social media impression (~$0.01-$0.10 brand value)
• Or: citation in high-trust context (LLM-cited as authoritative) ~ $0.50-$2.00 brand value per impression
• Aggregate across thousands of citations per month

Specifics vary; require survey-based validation.

Budget allocation guide

Quarterly AEO investment by stage:

---

Cheat sheet

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:

| 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 helps LLMs (and search engines) extract Q&A content from your pages. Three relevant schemas:

FAQPage schema

For pages with a list of questions and answers (e.g., FAQ page, knowledge base article with FAQ section).

{
  "@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).

{
  "@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.

{
  "@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

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:

<dl>
  <dt>Term</dt>
  <dd>Definition</dd>
</dl>

LLMs extract dt/dd pairs efficiently.

---

Content length

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:

---

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?
• [ ] Is FAQ schema added?
• [ ] 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