---
name: aeo-content-patterns
slug: aeo-content-patterns
description: This skill should be used when the user asks to "write content for AI search", "write for AI extraction", "format content for ChatGPT", "structure sentences for AI engines", "write extractable content", "optimize writing for AI search", "content patterns for AEO", "write for Perplexity", or any variation of writing, formatting, or structuring content at the sentence and paragraph level for extraction by AI answer engines.
category: general
---

# AEO Content Patterns

AI engines extract answers at the sentence level. They scan your page, identify the most relevant sentences, and lift them into a generated answer. The sentences they choose follow specific patterns: declarative, factual, self-contained, and specific. Sentences that don't match these patterns get skipped — even if the page ranks #1 in Google.

This skill covers the writing patterns that AI engines extract most reliably. Apply them to every page targeting AI search.

## The Extractability Principle

An extractable sentence is one an AI engine can lift from your page and drop into an answer without modification. It makes sense on its own, contains a complete thought, and includes enough specificity to be useful.

**Extractable:**
> HubSpot CRM starts at $0/month for the free plan and $20/month per seat for the Starter plan.

**Not extractable:**
> Pricing varies depending on your needs and the features you select. Contact our team to learn about the options that work best for your organization.

The first sentence can be lifted verbatim into an AI answer. The second says nothing that could be cited.

---

## The 8 Patterns AI Engines Extract

### 1. The definition pattern

**Format:** `[Term] is [category] that [key function/distinction].`

AI engines extract definitions more than any other pattern. Every page should contain at least one clean definition in the first 50 words.

| Bad | Good |
|-----|------|
| "Revenue intelligence has become an increasingly important part of the modern sales tech stack." | "Revenue intelligence is a category of sales software that captures buyer interaction data to surface deal insights and improve forecast accuracy." |
| "At its core, ABM is about focusing on accounts." | "Account-based marketing (ABM) is a B2B strategy that targets specific high-value accounts with personalized campaigns across multiple channels." |

**Rules:**
- Subject + "is" + category + "that" + function
- Under 35 words
- No hedging ("essentially," "basically," "in simple terms")
- Include the full term and abbreviation on first use

### 2. The comparison pattern

**Format:** `[X] differs from [Y] in [number] ways: [A], [B], and [C].`

Comparison queries are the highest-volume AI search queries for SaaS. Structure comparisons as extractable sentences, not just tables.

| Bad | Good |
|-----|------|
| "There are quite a few differences between these tools that are worth exploring in more detail." | "Notion differs from Confluence in three ways: Notion offers a flexible block-based editor, supports personal wikis alongside team spaces, and prices per user without seat minimums." |
| "Both tools have their strengths and weaknesses depending on your needs." | "HubSpot is better for small marketing teams. Salesforce is better for enterprise sales orgs with complex deal cycles and custom objects." |

**Rules:**
- Name both entities explicitly
- State the specific differences (not "there are differences")
- Under 40 words per comparison sentence
- Use direct verdicts: "[X] is better for [use case]. [Y] is better for [other use case]."

### 3. The quantified claim pattern

**Format:** `[Subject] [verb] [specific outcome] [with qualifier].`

AI engines cite factual claims with numbers over unquantified opinions.

| Bad | Good |
|-----|------|
| "Our customers see significant improvement in their results." | "Teams using Acme reduce SDR ramp time by 40%, based on data from 500+ customers." |
| "It helps companies save a lot of time." | "Acme saves sales teams an average of 12 hours per rep per week by automating CRM data entry." |

**Rules:**
- Include a number (percentage, time saved, dollar amount, count)
- Add a qualifier for credibility ("based on," "according to," "across X customers")
- Never use vague quantifiers ("many," "significant," "a lot of")

### 4. The list pattern

**Format:** `The [number] [things] are: 1. [item], 2. [item], 3. [item].`

AI engines extract numbered and bulleted lists readily. Use them for any content with multiple items.

| Bad | Good |
|-----|------|
| "There are several key steps involved in the process that teams typically follow." | "Setting up lead scoring requires four steps: 1. Define scoring criteria, 2. Assign point values, 3. Set MQL thresholds, 4. Configure routing rules." |

**Rules:**
- State the count upfront ("four steps," "three types," "five criteria")
- Name each item concisely (under 10 words per item)
- Use inline numbered lists for 3-5 items. Use bullet lists for 6+ items
- Each item must be independently understandable

### 5. The conditional pattern

**Format:** `Use [X] when [condition]. Use [Y] when [other condition].`

AI engines answer "when should I use X?" questions by extracting conditional statements.

| Bad | Good |
|-----|------|
| "It really depends on your specific situation and what you're trying to accomplish." | "Use Notion when your team is under 50 people and wants a single workspace for docs and projects. Use Confluence when you need deep Jira integration and structured documentation for engineering." |

**Rules:**
- Specific conditions, not vague ("when you need flexibility")
- Include a use case or team profile in the condition
- Direct recommendation — "use X when" not "you might consider X if"

### 6. The cost pattern

**Format:** `[Product] costs [amount] per [unit] on the [plan name] plan.`

Pricing queries are extremely common in AI search. Structure pricing as extractable sentences.

| Bad | Good |
|-----|------|
| "We offer competitive pricing that scales with your business." | "HubSpot Marketing Hub costs $800/month on the Professional plan, which includes up to 2,000 marketing contacts, automation workflows, and A/B testing." |

**Rules:**
- State the actual price. "Contact us for pricing" is never extractable
- Include the plan name
- Include what's included at that price point
- State the billing unit (per month, per user, per seat)

### 7. The process pattern

**Format:** `To [accomplish X], [do step 1], then [step 2], then [step 3].`

AI engines extract process descriptions for "how to" queries.

| Bad | Good |
|-----|------|
| "Getting started is a simple and straightforward process that your team can complete easily." | "To set up SPF for cold email, add a TXT record to your DNS with the value 'v=spf1 include:_spf.google.com ~all', wait 24-48 hours for propagation, then verify with MXToolbox." |

**Rules:**
- Start with "To [goal]" — matches how users phrase questions
- 3-5 steps per process sentence
- Include specific details (actual values, tool names, timeframes)
- For longer processes, use the list pattern instead

### 8. The negation pattern

**Format:** `[Common belief] is wrong. [Correct fact].`

AI engines value corrections of common misconceptions. These get cited as authoritative clarifications.

| Bad | Good |
|-----|------|
| "There are some common misunderstandings about this topic." | "AEO does not replace SEO. Google still drives 80%+ of search traffic. AEO is an additional channel that optimizes for AI-generated answers, which currently represent 10-15% of B2B research queries." |

**Rules:**
- State the misconception directly
- Correct it with a specific fact
- Include a number or comparison to anchor the correction

---

## Paragraph Structure for AEO

Beyond individual sentences, paragraph structure affects extractability.

### The inverted pyramid

Every paragraph should follow inverted pyramid structure: most important information first, supporting detail second.

```
Sentence 1: The answer / key claim (extractable standalone)
Sentence 2: Supporting evidence or detail
Sentence 3: Nuance, caveat, or context (optional)
```

AI engines weight the first sentence of each paragraph more heavily. If your key claim is in sentence 3, it may not be extracted.

### Paragraph length

- 2-4 sentences per paragraph. Never more than 5
- One idea per paragraph. Never combine two distinct claims in one paragraph
- Short paragraphs are more extractable than long ones

### Paragraph openers to avoid

| Never open with | Why | Instead |
|----------------|-----|---------|
| "It's worth noting that..." | Filler. Delays the point | State the point directly |
| "Interestingly, ..." | Editorializing. AI skips it | State the fact |
| "As mentioned above, ..." | Depends on context. Not self-contained | Repeat the key fact |
| "In today's fast-paced world..." | Generic filler. AI engines skip entire paragraphs that open this way | Cut it. Start with the claim |
| "Let's dive into..." | Conversational filler | State the topic: "[Topic] works by..." |
| "When it comes to..." | Windup before the pitch | Name the subject directly |

---

## Page-Level Content Rules

### First 50 words

The first 50 words of any page are the highest-value real estate for AEO. AI engines weight the opening heavily.

**Rules:**
- The first sentence must directly answer the page's primary question
- No introductions, context-setting, or "welcome to our guide"
- Include a definition (pattern #1) or a comparison verdict (pattern #2) or a quantified claim (pattern #3)
- Include the primary keyword naturally

### Headers (H2s and H3s)

Headers are extraction anchors. AI engines use headers to locate relevant sections.

**Rules:**
- H2s must be question-shaped: "What is X?", "How does X work?", "How much does X cost?"
- H3s can be declarative: "Pricing by plan", "Key differences"
- Never use clever or branded headers: "The Acme Advantage" tells an AI engine nothing
- Match header language to how buyers actually phrase questions

### Tables

Tables are the most extractable content format after Q&A pairs.

**Rules:**
- Use HTML tables, not images of tables
- Include a header row with clear column names
- Keep cell values short (1-15 words per cell)
- Use tables for any comparison, criteria list, or multi-attribute data
- Always pair a table with a summary sentence above it (AI engines extract both)

---

## Pre-Publish Checklist

Before publishing any page targeting AI search:

- [ ] First 50 words contain a direct answer using one of the 8 patterns
- [ ] At least 3 of the 8 extractable patterns appear on the page
- [ ] Every paragraph follows inverted pyramid (claim first, detail second)
- [ ] No paragraph exceeds 5 sentences
- [ ] All H2s are question-shaped
- [ ] At least one table on the page
- [ ] No filler openers ("It's worth noting," "Let's dive in," "In today's...")
- [ ] All claims include specific numbers, names, or comparisons
- [ ] No hedging language ("might," "could potentially," "it depends")
- [ ] Every comparison names both entities and gives a direct verdict
- [ ] Pricing is stated in actual numbers, not "contact us"
- [ ] Page reads coherently when only the first sentence of each paragraph is extracted

---

## Anti-Pattern Check

- Opening with 3 paragraphs of context before the answer → AI engines extract the top of the page. If the answer is in paragraph 4, it won't be cited. Move the answer to sentence 1
- Using hedged language throughout ("might help," "could improve," "it depends") → AI engines skip hedged statements for confident, declarative ones. Replace every "might" with a direct claim backed by a number
- Comparisons that avoid verdicts ("both tools have their strengths") → Useless for AI extraction. State the verdict: "[X] is better for [use case]. [Y] is better for [other use case]"
- All prose, no tables → Tables are extracted 2-3x more reliably than equivalent prose. Convert every comparison, criteria list, and multi-attribute dataset into a table
- Sentences that reference other parts of the page ("as mentioned above") → Not self-contained. AI engines extract individual sentences. Every sentence must make sense in isolation
- Paragraphs with 8-10 sentences → Too long. AI engines may extract only the first sentence, missing important detail. Break into 2-3 shorter paragraphs with the key claim leading each one