AEO and GEO: The Complete Guide to Getting Traffic from AI Search in 2026

AI search is eating traditional search. ChatGPT has 900 million weekly active users processing 2 billion prompts per day. Google AI Overviews appear on 16-25% of US queries. Perplexity, Claude, and Copilot are growing fast. The question is no longer whether AI search matters — it is whether your site gets cited when these systems answer questions about your domain.

The data is compelling: AI referral traffic grew 527% year-over-year across 19 GA4 properties in early 2025, and converts at 4.4-5x the rate of traditional organic search. One B2B portfolio of 42 websites saw AI-driven sessions increase 240% while traditional organic clicks dropped 18%. A peer-reviewed paper from Georgia Tech and Princeton proved that targeted optimization can boost AI visibility by 30-40%.

This guide covers everything you need to implement AEO (Answer Engine Optimization) and GEO (Generative Engine Optimization) on your site — backed by research papers, industry data, and real case studies. If you want a deeper dive on llms.txt specifically, I wrote a complete guide to creating llms.txt files.

Table of Contents

• What Are AEO and GEO?
• The Data: Why AI Search Traffic Matters
• What the Research Says: The Georgia Tech GEO Paper
• How Each AI Platform Chooses Sources
• JSON-LD Structured Data for AI Citations
• Configuring robots.txt for AI Bots
• Implementing llms.txt
• Content Structure That Gets Cited
• Semantic HTML for AI Crawlers
• Off-Site Brand Mentions: The New Backlinks
• Measuring AI Traffic
• What the Experts Say
• Implementation Roadmap
• Quick Reference
• Conclusion

What Are AEO and GEO?

Answer Engine Optimization (AEO) is the practice of optimizing your website to be cited as a source by AI-powered answer engines — ChatGPT, Perplexity, Google AI Overviews, Microsoft Copilot, and Claude. Unlike traditional SEO where you optimize for ranking positions, AEO focuses on getting your content extracted, summarized, and linked by AI systems that synthesize answers from multiple sources.

Generative Engine Optimization (GEO) is the academic term for the same discipline. The name was coined in the landmark paper "GEO: Generative Engine Optimization" published at ACM SIGKDD 2024 by researchers from Georgia Tech, Princeton, IIT Delhi, and the Allen Institute for AI.

The key difference from traditional SEO: in AI search, there are no "10 blue links." There is one synthesized answer, potentially citing 3-8 sources. Either you are one of those sources, or you are invisible.

AEO vs SEO vs GEO

Lily Ray, VP of SEO at Amsive Digital, frames it well: AEO/GEO "represents an expansion of the digital marketer's toolkit, not a paradigm shift." Strong Google rankings remain foundational because every URL surfaced in an LLM response is pulled from a live search index.

The Data: Why AI Search Traffic Matters

AI referral traffic is small in absolute terms but growing at a rate that demands attention. Here are the numbers from the most comprehensive industry studies available.

Traffic Volume and Growth

Conversion Rates by Platform

This is the most interesting finding: AI traffic converts dramatically better than traditional organic.

Source: Superlines AI Search Statistics 2026

A paired t-test by Amsive found the conversion rate difference was not always statistically significant across all datasets, so treat these numbers as directional rather than absolute. The likely explanation: users who click through from an AI answer are further along in their decision journey than someone browsing search results.

Case Studies with Real Numbers

Mentimeter (EdTech via Siege Media): 124,000 ChatGPT-referred sessions and 3,400 conversions in a single month. Strategy: content structured for citability with clear definitions, data in tables, and step-by-step frameworks.

B2B SaaS Portfolio (42 websites, Q4 2025 - Q1 2026): AI traffic converted at 14.2% vs 2.8% for traditional organic — a 5x advantage. Traditional organic clicks dropped 18%, but AI-driven sessions increased 240%.

E-commerce Water Supply Retailer: Product pages with "Use Cases" sections attracted 90% of AI visits with a 5% conversion rate vs 4% from organic search.

14-Month LLM Revenue Engagement: One client achieved a 4,900% revenue increase and 2,622% traffic growth from LLM-referred sources over 14 months (HubSpot case study).

Industry Investment

The market is betting heavily on AEO:

• 56% of CMOs made significant AEO investments in 2025
• 94% plan to increase AEO investment in 2026
• 97% reported positive impact from their AEO efforts

Source: Conductor State of AEO/GEO CMO Report

What the Research Says: The Georgia Tech GEO Paper

The foundational academic research on GEO is "GEO: Generative Engine Optimization" by Pranjal Aggarwal et al., published at ACM SIGKDD 2024. The paper is a joint effort from IIT Delhi, Princeton University, Georgia Tech, and the Allen Institute for AI. It introduced GEO-bench, a benchmark of 10,000 diverse queries, and tested nine optimization strategies.

The Three Methods That Work

Each of these strategies improved visibility by 30-40% on the Position-Adjusted Word Count metric:

1. Cite Sources — Adding inline citations and references to authoritative sources
2. Add Statistics — Including specific data points, percentages, and numbers
3. Add Expert Quotations — Including quotes with attribution from recognized experts

What Does NOT Work

Keyword stuffing offers little to no improvement in generative engine responses. This is a major departure from traditional SEO, where keyword density still influences rankings. LLMs evaluate semantic relevance, not keyword frequency.

Domain-Specific Results

The paper found that different domains respond to different strategies:

Quantitative Impact

• Basic unoptimized content scored 19.3 in visibility metrics
• Adding authoritative citations, statistics, and fluency improvements increased scores above 40 — over 100% improvement
• Tables earn approximately 2.5x more AI citations than the same information presented as prose
• Direct answers in the first 40-60 words of a section significantly increase citation probability
• Maintaining statistics every 150-200 words optimizes fact density

Source: arXiv:2311.09735, ACM Digital Library

How Each AI Platform Chooses Sources

One of the most important insights from the research: only 11% of domains are cited by both ChatGPT and Perplexity. Each platform has different retrieval systems, biases, and citation patterns.

Platform Comparison

ChatGPT Citation Patterns

ChatGPT dominates AI referral traffic at 87.4% market share. Key findings from Ahrefs analysis of 26,283 source URLs:

• 67% of top 1,000 citations are "off-limits" to marketers (Wikipedia, .gov, reference sites)
• 43.8% of cited page types are "best X" blog lists
• Only 12% of AI-cited URLs rank in Google's top 10 for the original prompt
• Favors structured content with clear H2/H3 headers framed as questions

Perplexity Citation Patterns

Perplexity uses a three-layer source selection process:

1. Keyword and semantic embedding matching
2. Cross-encoder reranking
3. ML reranking with entity, authority, and recency signals

Perplexity has a strong freshness bias — a one-week-old blog post on an authoritative domain gets cited more readily than a two-year-old evergreen post. It also pulls 40% more citations from high-authority websites compared to mid-tier blogs.

Warning: Cloudflare published evidence in August 2025 that Perplexity uses undeclared crawlers with generic Chrome user-agent strings to circumvent robots.txt blocks. The New York Times has sued Perplexity over this behavior.

Google AI Overviews

Google AI Overviews are decoupling from traditional rankings. In July 2025, 76% of cited URLs ranked in the organic top 10. By February 2026, only 38% came from the top 10, with 31.2% from positions 11-100 and 31% from beyond the top 100.

Seven core ranking factors identified by Wellows:

• Semantic completeness (r=0.87 correlation)
• Multi-modal content (+156% selection rate)
• E-E-A-T authority (96% of citations come from authoritative sources)
• Entity Knowledge Graph density (15+ entities = 4.8x boost)
• Structured data markup (+73% selection rate)
• Content freshness
• Page experience signals

YouTube accounts for 18.2% of all citations sourced from outside the top 100 — a significant signal for video content strategy.

JSON-LD Structured Data for AI Citations

Structured data in JSON-LD format is the most impactful technical change you can make. Google explicitly recommends JSON-LD, and every AI engine tested prefers it because it is cleanly separated from HTML and easier to parse programmatically.

The Numbers

• FAQPage schema makes content 3.2x more likely to appear in AI Overviews (Frase.io study)
• Fully-populated Product + Review schema achieves a 61.7% citation rate vs 41.6% for generic schema (Growth Marshal, n=730)
• Structured data markup provides a +73% selection rate for Google AI Overviews (Wellows)
• Only 12.4% of websites currently implement structured data (Surfeo) — early mover advantage is real

The Generic Schema Penalty

This is the most counterintuitive finding: partially-filled, generic schema is worse than no schema at all.

A February 2026 peer-reviewed study by Growth Marshal (n=730 citations across ChatGPT and Gemini) found:

The penalty is most severe for lower-authority domains (DR 60 or below): attribute-rich schema achieves 54.2% citation rate vs 31.8% for generic. The takeaway: either fully populate every relevant attribute or skip schema entirely.

Priority Schema Stack for Technical Blogs

1. BlogPosting (every post)

{
  "@context": "https://schema.org",
  "@type": "BlogPosting",
  "headline": "AEO and GEO: The Complete Guide to Getting Traffic from AI Search in 2026",
  "description": "Learn how to get your site cited by ChatGPT, Perplexity, and Google AI Overviews.",
  "author": {
    "@type": "Person",
    "name": "Dylan Boudro",
    "url": "https://dylanboudro.com",
    "sameAs": ["https://github.com/starmorph", "https://linkedin.com/in/dylanboudro"]
  },
  "publisher": {
    "@type": "Organization",
    "name": "Starmorph",
    "url": "https://starmorph.com",
    "logo": {
      "@type": "ImageObject",
      "url": "https://blog.starmorph.com/static/images/logo.png"
    }
  },
  "datePublished": "2026-04-22",
  "dateModified": "2026-04-22",
  "image": "https://blog.starmorph.com/static/images/aeo-geo-optimization-guide.png",
  "mainEntityOfPage": "https://blog.starmorph.com/blog/aeo-geo-optimization-guide"
}

2. FAQPage (posts with Q&A content) — Nest within posts that naturally answer multiple questions. Each question/answer pair must be fully populated.

3. Organization (site-wide) — Name, URL, logo, and sameAs links to all verified profiles (GitHub, LinkedIn, Twitter).

4. BreadcrumbList — Provides navigational context that helps AI parsers understand site structure.

5. HowTo (tutorial posts) — Name, step names, step text, and totalTime for procedural content.

Configuring robots.txt for AI Bots

Most websites either have no AI bot directives or block all AI crawlers by default. This is leaving traffic on the table. GPTBot grew 305% year-over-year, and ChatGPT-User surged 2,825% in requests according to Cloudflare's 2025 crawler report. AI bots now originate 4.2% of all HTML requests on the web.

The AI Bot Landscape

OpenAI operates three bots:

Anthropic operates three bots:

Anthropic is unique in that all three bots honor robots.txt, including user-initiated requests. Blocking Claude-SearchBot "may reduce your site's visibility and accuracy in user search results."

Other notable AI bots:

Recommended robots.txt for Maximum AI Visibility

# Search engines
User-agent: Googlebot
Allow: /

# AI search bots - ALLOW for citation traffic
User-agent: GPTBot
Allow: /

User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: Claude-User
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Google-Extended
Allow: /

User-agent: Applebot-Extended
Allow: /

# Default
User-agent: *
Allow: /

Sitemap: https://yoursite.com/sitemap.xml

Note: 79% of top news sites block AI training bots. If you want AI citations but not training, you can allow search bots (OAI-SearchBot, ChatGPT-User, Claude-SearchBot) while blocking training bots (GPTBot, ClaudeBot).

Implementing llms.txt

The llms.txt specification was proposed by Jeremy Howard, co-founder of Answer.AI, in September 2024. It is a Markdown file served at /llms.txt that provides a structured index of your site content optimized for AI consumption.

Over 844,000 websites have adopted it, including Anthropic, Cloudflare, Stripe, Vercel, and Supabase.

Honest Assessment

No major LLM provider has officially confirmed that their crawlers read or prioritize llms.txt files. Otterly.ai's three-month study across 60,000+ AI bot hits found llms.txt was used in only 0.1% of AI visits. However, implementation takes minutes and costs nothing — it is a free option on future AI discoverability.

For a deeper dive on implementation, including Next.js build-time generation and the llms-full.txt companion file, see my complete llms.txt guide.

Format

# Your Site Name

> A one-to-three sentence summary of your site and who it serves.

## Main Content

- [Page Title](https://yoursite.com/page): Short description of this resource
- [Another Page](https://yoursite.com/other): What this page covers

## Documentation

- [API Reference](https://yoursite.com/docs/api): Complete API documentation
- [Getting Started](https://yoursite.com/docs/quickstart): Setup guide

## Optional

- [About](https://yoursite.com/about): Background information
- [Contact](https://yoursite.com/contact): How to reach us

The ## Optional section uses a reserved section name — URLs listed there can be skipped when context windows are limited.

Companion Files

Analytics show that llms-full.txt receives roughly 2x more traffic than the index version, suggesting AI tools prefer complete content in a single request when available.

Content Structure That Gets Cited

Content structure is the single highest-leverage change you can make for AI visibility. The Georgia Tech paper quantified this: the right structural patterns deliver 30-40% improvement in citation probability.

Six Structural Patterns That Increase Citations

1. Answer-first paragraphs. The first 40-60 words of every section should directly and completely answer the section heading. AI systems with real-time retrieval evaluate relevance primarily on opening content.

2. Question-based H2 headers. Format headings as natural questions users actually ask: "What is RAG?" instead of "RAG Overview." Question-based headers are 40% more likely to be cited.

3. Tables over prose. Tables earn approximately 2.5x more AI citations than equivalent information presented as paragraphs. An 81% extraction rate for tables vs 23% for the same data in prose.

4. Statistics every 150-200 words. Claims backed by specific data points have a 40% higher citation rate than qualitative statements. Include specific numbers, percentages, and benchmark data.

5. Short paragraphs (40-60 words). Each paragraph should contain exactly one idea that makes sense if extracted in isolation. This is the optimal extraction length for AI systems.

6. Modular sections. Each H2 section should function as a standalone answer if pulled out of context by an AI system. Test each section: does this make sense without the rest of the article?

Content Format Citation Rates

The Definition Pattern

AI platforms prioritize concise definitions. For any key term, use this pattern:

Term is/refers to [definition]. [One sentence of context]. [One concrete example].

Lead sections with these definitions — they are high-priority extraction targets.

Meta Titles and Descriptions

Pages with high semantic alignment in meta descriptions receive 4.7 AI citations vs 4.1 for low-alignment pages. The shift is from clickbait to factual, entity-rich summaries:

Before (CTR-optimized): "You Won't Believe How Fast LLMs Run on Apple Silicon!"

After (AI-citation-optimized): "Running LLMs on Apple Silicon: MLX, llama.cpp, and Ollama Performance Benchmarks for M-Series Chips"

Include key entities — product names, technical terms, framework names — that AI systems use for semantic matching.

Semantic HTML for AI Crawlers

AI crawlers fetch a page, parse the DOM and Accessibility Tree, and segment content into blocks for indexing and summarization. Clean semantic HTML reduces computational load and increases extraction accuracy.

HTML Elements and How AI Interprets Them

Implementation Checklist

• Blog posts wrapped in <article> inside <main>
• Sidebar and related content in <aside>
• All dates using <time datetime="2026-04-22">
• Images using <figure> and <figcaption>
• Strict heading hierarchy — no skipped levels (H1, H2, H3, never H1, H3)
• Blockquotes with cite attributes for external quotes

Sitemap Optimization

Bing's official guidance (July 2025) states that accurate <lastmod> values are critical for AI-powered search discovery. AI engines have a strong recency bias.

• Ensure <lastmod> reflects actual content modification dates, not build timestamps
• Include only canonical, indexable URLs
• Add <image:image> extensions for posts with key images
• Reference the sitemap in robots.txt
• Submit to both Google Search Console and Bing Webmaster Tools (which feeds Copilot)

Off-Site Brand Mentions: The New Backlinks

Both Lily Ray and Kevin Indig emphasize that third-party mentions are now the primary driver of AI citations. Mike King, two-time AI Search Marketer of the Year, calls this "relevance engineering" — optimizing content passages and brand mentions across the web to directly feed AI systems.

Where AI Platforms Find Sources

Actionable Channels

Reddit — Share genuinely helpful content in relevant subreddits. Reddit is the top community source for both Perplexity and Google AI Overviews.

YouTube — Create companion video content. 18.2% of Google AI Overview citations from outside the top 100 organic results come from YouTube.

Hacker News — Submit high-quality technical deep dives. HN links carry strong authority signals.

LinkedIn — Publish thought leadership articles. LinkedIn is cited in 1.3% of Google AI Overview results.

Dev.to / Hashnode — Cross-post with canonical URLs pointing back to your site.

GitHub — Maintain active repositories referenced in your posts. Code repos are authoritative signals.

The key insight: LLMs do not just crawl your site. They aggregate brand mentions across the entire web. Being mentioned positively on Reddit, in "best of" lists, on review sites, and in authoritative publications directly increases your citation probability across all AI platforms.

Measuring AI Traffic

You cannot optimize what you cannot measure. The challenge: 70.6% of AI-adjacent visits arrive without referrer headers and show up as "Direct" in analytics platforms.

Known AI Referrer Domains

Track these referrer domains in your analytics platform:

Analytics Implementation

In Google Analytics 4, create a custom channel group called "AI Traffic" using regex matching on these referrer domains. In PostHog, create a custom cohort using the $referrer property.

The Dark Traffic Problem

The 70.6% dark traffic figure means your analytics dramatically undercount AI referrals. Many AI platforms, especially mobile apps, do not pass referrer headers. Consider this when evaluating AI traffic impact — the real number is likely 3-5x what your analytics show.

Specialized Monitoring Tools

New Metrics Beyond Clicks

Given zero-click behavior in AI search, experts recommend tracking:

• Brand mention frequency in AI responses (share of voice)
• Citation rate per query category
• Assisted conversions from AI-influenced touchpoints
• Brand sentiment in AI-generated summaries

What the Experts Say

Lily Ray (VP of SEO, Amsive Digital)

Ray argues that GEO is not fundamentally new: "Many vendors claiming to offer cutting-edge optimization strategies are simply repackaging core SEO approaches using a different name." Her data shows that every URL surfaced in an LLM response is pulled from a live search index, meaning strong Google performance is foundational.

Her survey found that 38% of respondents received 0-0.5% traffic from ChatGPT, and 70% received under 2%. She emphasizes that off-site visibility — Reddit, Quora, G2, LinkedIn — now heavily influences AI citations.

Rand Fishkin (SparkToro)

Fishkin's clickstream analysis shows Google answers nearly two-thirds of queries without a click (zero-click searches), and this accelerated after Google's AI Overviews rollout. AI tools have nearly tripled their usage share but could take 6-10 years to rival traditional search at current growth rates. His recommendation: invest in "zero-click marketing" strategies where brand visibility itself has value, even without generating a click.

Mike King (iPullRank, 2x AI Search Marketer of the Year)

King coined "relevance engineering" — optimizing content passages, data feeds, and knowledge bases to directly feed AI systems. His position: LLMs pull citations from third-party sites heavily, so building brand mentions on other sites is critical. He launched SEO Week in April 2025 with 40+ speakers focused on AI search strategies.

Kevin Indig (Growth Memo)

Indig published an audit of 1.2 million ChatGPT responses. His research on the "decoupling of clicks from impact" changed how marketers evaluate GEO success. His 2026 predictions include the end of AI dashboards, the rise of agentic SEO, and a web divided between bots and verified humans. He emphasizes that thought leadership and original research on authoritative third-party sites are the best ways to improve brand mentions in AI responses.

Implementation Roadmap

Phase 1: Quick Wins (1 day)

These changes take minutes and require no content rewrites:

• Update robots.txt to explicitly allow AI search bots
• Verify sitemap.xml has accurate <lastmod> dates
• Create and deploy llms.txt and llms-full.txt
• Submit sitemap to Bing Webmaster Tools (feeds Copilot)

Phase 2: Technical Foundation (1 week)

• Add fully-populated BlogPosting JSON-LD to every post
• Add FAQPage schema to posts with Q&A content
• Add Organization schema site-wide with sameAs links
• Audit semantic HTML (article, main, section, time, figure)
• Set up AI traffic cohort in PostHog or GA4

Phase 3: Content Optimization (2-4 weeks)

• Rewrite meta titles and descriptions for semantic alignment
• Restructure top 10 highest-traffic posts with question-based H2s
• Add tables, statistics, and inline citations to existing content
• Convert comparison paragraphs to structured tables
• Add answer-first paragraphs to every section

Phase 4: Ongoing

• Publish new content following GEO structural patterns
• Build off-site brand mentions on Reddit, HN, YouTube, LinkedIn
• Cross-post to Dev.to and Hashnode with canonical URLs
• Monitor AI referral traffic monthly and adjust
• Track brand mention frequency in AI responses

Quick Reference

GEO Optimization Methods Ranked (Georgia Tech Paper)

AI Bot User-Agents to Allow

Schema Priority for Technical Blogs

Content Structure Rules

Conclusion

AEO and GEO are not a replacement for SEO — they are the next layer on top of it. The data is clear: AI referral traffic is small but growing at 500%+ per year, converts at 4-5x the rate of traditional organic, and the platforms serving it have hundreds of millions of users.

The implementation is straightforward. Start with the zero-effort wins (robots.txt, sitemap, llms.txt), build the technical foundation (JSON-LD, semantic HTML, measurement), then systematically restructure your content for AI extractability. The Georgia Tech research gives you a clear playbook: cite sources, add statistics, include expert quotes, use tables, and write answer-first paragraphs.

The sites that get cited today will compound their advantage as AI search grows. Every citation builds authority, every mention reinforces your brand in the training data, and every referral visitor converts at 4-5x the rate of a search click. The window for early-mover advantage is still open — but it is closing as 94% of CMOs increase their AEO investment this year.