Inside the Study "Generative Engine Optimization: How to Dominate AI Search"
This article dissects the comprehensive research paper from the University of Toronto (arxiv.org/abs/2509.08919)—translating dense academic findings into actionable insights anyone can understand.
Our approach is simple: We'll show you exactly what the researchers discovered using their technical language, then explain what it actually means in plain English, and finally give you concrete recommendations you can implement immediately.
Why This Research Paper Matters
The Research Team:
This study was conducted by Mahe Chen, Xiaoxuan Wang, Kaiwen Chen, and Nick Koudas at the University of Toronto. Their work represents one of the most comprehensive analyses of AI search engines to date, examining thousands of queries across multiple languages, industries, and AI platforms.
What the researchers stated in their abstract:
"The rapid adoption of generative AI-powered search engines like ChatGPT, Perplexity, and Gemini is fundamentally reshaping information retrieval, moving from traditional ranked lists to synthesized, citation-backed answers. This shift challenges established Search Engine Optimization (SEO) practices and necessitates a new paradigm, which we term Generative Engine Optimization (GEO)."
What makes this study different:
Scale: Analyzed 1,000+ consumer queries across 10 verticals
Breadth: Compared Google against ChatGPT, Claude, Perplexity, and Gemini
Depth: Tested across 5 languages (Chinese, Japanese, German, French, Spanish) and multiple query variations
Rigor: Used controlled experiments with quantifiable metrics including Jaccard overlap indices, domain classification systems, and statistical aggregation
Research methodology:
The team employed a sophisticated pipeline that included automated query generation via GPT-4o, parallel execution across multiple AI engines with web search enabled, systematic domain extraction and classification into Brand/Earned/Social categories, and comprehensive overlap analysis using normalized similarity measures.
This isn't opinion or speculation—it's peer-reviewed, data-driven research showing exactly how AI search engines behave differently from traditional search.
Citation: Chen, M., Wang, X., Chen, K., & Koudas, N. (2025). Generative Engine Optimization: How to Dominate AI Search. arXiv:2509.08919 [cs.IR]. Retrieved from https://arxiv.org/abs/2509.08919
Let's break down their findings
Let's break down their findings...Think of this as your bridge between cutting-edge AI search research and practical generative engine optimization strategies that actually work in 2025.
The Big Picture: Why This Research Matters
What the researchers said:
"The rapid adoption of generative AI-powered search engines like ChatGPT, Perplexity, and Gemini is fundamentally reshaping information retrieval, moving from traditional ranked lists to synthesized, citation-backed answers. This shift challenges established Search Engine Optimization (SEO) practices and necessitates a new paradigm, which we term Generative Engine Optimization (GEO)."
What this actually means:
Remember when getting your website to rank on Google meant everything? That playbook is becoming obsolete. AI search tools like ChatGPT and Perplexity don't show users a list of ten blue links—they give direct answers and cite a handful of sources. If your brand isn't one of those sources, you're invisible.
This isn't a small shift. It's like the difference between being listed in the Yellow Pages versus having the phone operator personally recommend your business when someone calls for help.
Our recommendation:
Don't abandon traditional SEO services—they still matter. But you need to add GEO to your strategy immediately. Start by auditing which AI search engines currently mention your brand when users ask questions in your industry.
Finding #1: The Earned Media Earthquake
What the researchers discovered:
"Our key findings reveal that AI Search exhibit a systematic and overwhelming bias towards Earned media (third-party, authoritative sources) over Brand-owned and Social content, a stark contrast to Google's more balanced mix."
In their automotive vertical study:
Google showed: 39.5% Brand, 15.4% Social, 45.1% Earned
AI search showed: 18.1% Brand, 0% Social, 81.9% Earned
What this actually means:
Your company blog and social media presence aren't cutting it anymore. AI search engines almost completely ignore Twitter threads, Instagram posts, and even your own website content. Instead, they're obsessed with what independent journalists, review sites, and industry experts say about you.
Think of it this way: Google is like a librarian who shows you where different books are located. AI search is like a professor who only cites peer-reviewed sources and ignores self-published material.
Our recommendation:
Shift at least 40% of your content budget from owned content (your blog, social media) to earned media campaigns. This means:
Hire a PR firm or dedicated media relations person who can get you featured in industry publications
Build relationships with journalists and reviewers in your space
Create newsworthy stories that give third parties a reason to write about you
Target high-authority review sites specific to your industry (like TechRadar for tech, or Car and Driver for automotive)
Learn more about building topical authority for AI search systems.
Finding #2: Each AI Search Engine Is Different (Really Different)
What the researchers discovered:
"AI Search services differ significantly from each other in their domain diversity, freshness, cross-language stability, and sensitivity to phrasing."
Specific findings:
Claude: Shows high cross-language stability, often reusing the same authoritative English sources across multiple languages
ChatGPT: Completely swaps its source ecosystem when you change languages—95.1% Earned media for niche brands with zero Social sources
Perplexity: More inclusive with 23.8% Social content (especially YouTube) and more Brand sources (34.6% in automotive)
Gemini: Most brand-leaning at 25.1% Brand sources, falls between extremes
What this actually means:
Imagine you're trying to get your restaurant recommended. One food critic only reads Michelin guides (ChatGPT). Another watches YouTube cooking channels and reads Yelp (Perplexity). A third trusts the same five international food magazines regardless of which city they're in (Claude). The fourth has a slight preference for checking out restaurant websites directly (Gemini).
You need a completely different strategy for each.
Our recommendation:
Stop thinking "AI search" as one thing. You need engine-specific tactics:
For ChatGPT & Claude (the strictest):
Prioritize getting featured in top-tier, authoritative publications
Focus on timeless, expert-level content
Don't bother with social media campaigns for AI visibility
For Perplexity:
Create high-quality YouTube content
Ensure accurate product listings on major retailer sites
Engage in community discussions that get referenced
For Gemini:
Maintain strong, well-structured content on your own domain
Use detailed schema markup
Balance earned media with owned content
Finding #3: The "Big Brand Bias" Is Real
What the researchers discovered:
In their cola brand study testing unbranded queries like "best cola brands":
ChatGPT: 56.3% major brands (Coca-Cola, Pepsi), 12.3% niche brands, 31.4% other
Perplexity: 67.9% major brands, only 5.8% niche brands
Coca-Cola received 107 mentions in Perplexity results, while niche craft sodas like Jones Soda barely registered.
What this actually means:
If you're not already a household name, AI search engines are actively working against you. They default to market leaders even when users don't specifically ask for them.
It's like asking a friend "what's a good soda?" and they only suggest Coke and Pepsi, even though you were hoping to discover something new.
Our recommendation:
If you're a smaller brand or startup, you need to fight twice as hard:
Dominate a specific niche: Instead of being "a great project management tool," become "the best project management tool for remote creative teams"
Over-invest in niche authority: Get featured in specialized publications, not just mainstream ones. A mention in a respected niche blog can carry more weight than a passing reference in Forbes
Create comparison content: Explicitly position yourself against major competitors with detailed comparison tables that AI can scan
Leverage Perplexity's openness: Since Perplexity includes more diverse sources including YouTube, video content is your secret weapon
For ecommerce brands, this is especially critical for product discovery.
Finding #4: Language Matters More Than You Think
What the researchers discovered:
"GPT shows the lowest overlap: domain sets across languages are consistently near-zero, i.e., it taps different site ecosystems by language."
Cross-language domain overlap findings:
Claude: High stability, reuses English authority sources across languages
ChatGPT: Near-zero overlap—completely different sources for each language
Perplexity & Gemini: Low to moderate overlap
Website language distribution under non-English prompts:
Japanese queries: 75%+ target language citations
Chinese queries: English still dominated (especially on Gemini)
Spanish queries: Mixed, with English slightly exceeding Spanish
What this actually means:
Simply translating your website into French doesn't make you visible in French AI search results. ChatGPT will pull from an entirely different set of French-language sources that probably don't include your translated content.
It's like having a store with signs in Spanish—that doesn't mean Spanish-speaking customers will find you if you're located in an English-speaking neighborhood.
Our recommendation:
For global brands, you need a multi-language earned media strategy:
For ChatGPT dominance: Build relationships with authoritative publishers in each target language. Hire local PR firms in major markets.
For Claude visibility: Double down on getting featured in top-tier English-language publications with global reach—this authority will transfer across languages.
For Perplexity/Gemini: Hybrid approach—maintain both English authority and local-language presence.
Don't just translate, localize authority: Getting mentioned in Le Monde (France) is worth more than translating your blog posts into French.
Check out our guide on getting cited in AI search systems for international markets.
Finding #5: Content Structure Trumps Content Volume
What the researchers discovered:
"Engineer content for machine scannability and justification... Create detailed comparison tables (vs. competitors or previous models). Use clear, bulleted pros and cons lists. State your value proposition explicitly (e.g., 'longest battery life,' 'most durable build,' 'best value for money')."
What this actually means:
AI search engines don't read like humans. They scan for specific, extractable facts that justify recommendations. A 2,000-word blog post about "why our product is great" is worthless. A comparison table showing "our battery lasts 18 hours vs. competitor A's 12 hours" is gold.
Think of AI as a research assistant who needs to justify their recommendation to their boss. They need specific, quotable facts, not marketing fluff.
Our recommendation:
Audit your website content and restructure it for AI scannability:
Must-haves on every product/service page:
Comparison tables with specific metrics
Bulleted pros and cons lists
Explicit value propositions in bold ("Longest warranty in its class: 5 years")
Schema markup for all specifications
Clear pricing with no "contact us" barriers
Content to create:
"X vs. Y" comparison pages for every major competitor
"Best X for Y" guides (e.g., "Best project management software for teams under 50")
FAQ sections with specific, technical answers
Specification sheets that AI can easily parse
See our comprehensive GEO framework for detailed implementation steps.
Finding #6: Freshness Varies By Engine and Vertical
What the researchers discovered:
In consumer electronics:
Claude: Mean article age 117 days, median 62 days, 92.5% coverage
Similar patterns for ChatGPT
In automotive:
Claude: Mean article age 331 days, median 148 days, only 61% coverage
Substantially older content, indicating reliance on static ranking pages
Perplexity consistently returned fresher content across both verticals.
What this actually means:
Some AI engines are citing articles from nearly a year ago, especially in certain industries. If all the recent articles about your industry ignore your brand, you're stuck being invisible based on old coverage.
It's like being judged by your high school yearbook photo when you're now 30 years old.
Our recommendation:
Short-term:
Create "newsjacking" opportunities—tie your brand to current industry trends
Pitch fresh angles to journalists who covered your space 6-12 months ago
Update and republish older articles that mentioned you with new data
Long-term:
Maintain a consistent PR cadence—you need regular mentions, not just a big launch splash
Build relationships with reporters who cover your vertical so you're their go-to source
Create annual reports or studies that generate new coverage cycles
For rapidly evolving industries, consider AI SEO agency partnerships that can maintain consistent earned media presence.
Finding #7: Local Search Is Fragmented Chaos
What the researchers discovered:
Local business category overlaps between AI search and Google:
Home Cleaning: 20.6% overlap
Dentists: 11.9% overlap
Auto Repair: 2.5% overlap
IT Support: 0.1% overlap
Cross-engine domain diversity in local services showed:
Only a handful of domains (like homestars.com) appeared across all engines
Gemini and Perplexity explored broader ranges
Claude and ChatGPT used narrower, more conservative source sets
What this actually means:
If you're a local business, traditional Google Local SEO and AI search optimization are completely different games. Getting reviews on Google doesn't automatically make you visible in ChatGPT.
It's like being listed in the local Chamber of Commerce directory but not in the phone book—different systems, different audiences.
Our recommendation:
For local businesses, you need a multi-platform review and citation strategy:
Directory presence: Get listed on major review aggregators that AI engines trust (Yelp, Angie's List, HomeAdvisor, etc.)
Industry-specific platforms: Every industry has authoritative review sites (Zocdoc for doctors, Avvo for lawyers, Houzz for contractors)
Local media: Get featured in local newspapers and magazines—these carry weight with AI engines
Google Business Profile: Still essential, but not sufficient
Structured data: Use LocalBusiness schema markup on your website
The fragmentation means you can't just "win Google" and call it done.
Finding #8: Schema Markup Is Your Secret Weapon
What the researchers discovered (implied throughout):
"Implement schema markup (Schema.org) with extreme rigor for all product specifications, prices, reviews, and availability to become an 'API-able' brand that AI agents can easily parse."
What this actually means:
Schema markup is code you add to your website that explicitly tells machines what information means. Without it, AI has to guess what "$299" refers to. With it, AI knows precisely that "$299" is the product price, not a dimension or model number.
Think of schema as adding labels to everything in your house before a move. Without labels, movers guess. With labels, they know exactly where everything goes.
Our recommendation:
Implement comprehensive schema markup across your entire site:
Priority schema types:
Product schema (name, description, image, price, availability)
Review/Rating schema (aggregate ratings, individual reviews)
Organization schema (your company info, social profiles)
FAQ schema (structured Q&A)
Article schema (for blog content)
LocalBusiness schema (for local companies)
BreadcrumbList schema (site structure)
Tools to use:
Google's Structured Data Markup Helper
Schema.org documentation
Schema validation tools
JSON-LD format (preferred by AI systems)
This is technical work that pays massive dividends. Consider this non-negotiable infrastructure.
The Meta-Strategy: Treating Your Website as an API
What the researchers concluded:
"The core principles of technical SEO—such as having a well-structured, crawlable site—are foundational, as AI agents require clean, machine-readable data to function. However, the findings reveal that a new strategy, which can be termed Generative Engine Optimization (GEO), is required."
What this actually means:
Your website isn't just a pretty brochure anymore. It's a database that AI agents query. If your site is hard for machines to read, you're invisible.
Imagine you're a librarian. A well-organized library with clear labels and catalog numbers is easy to work with. A library with books randomly stacked and handwritten notes is a nightmare. AI search engines will simply ignore the nightmare library.
Our recommendation:
Audit your entire web presence through the lens of "machine readability":
Technical infrastructure:
Clean, semantic HTML structure
Comprehensive schema markup (see above)
Fast page load speeds
Mobile-responsive design
Accessible to crawlers (no unnecessary JavaScript barriers)
Content structure:
Clear headings hierarchy (H1, H2, H3)
Data tables where appropriate (not images of tables)
Specifications in structured formats
Explicit statements (not implied meanings)
Information architecture:
Logical URL structure
Internal linking that makes relationships clear
Sitemap that's regularly updated
Breadcrumb navigation
This isn't about making your site "pretty"—it's about making it functionally parseable by machines.
Your 90-Day GEO Action Plan
Based on this research, here's what you should do right now:
Month 1: Audit & Baseline
Check current AI visibility: Ask ChatGPT, Perplexity, Claude, and Gemini questions in your industry. Do they mention you?
Audit current schema: Use validation tools to see what structured data you're currently providing
Map your earned media: Identify which third-party sites have mentioned you in the past 12 months
Analyze competitors: See which sources AI engines cite when they recommend competitors
Month 2: Quick Wins
Implement basic schema markup: At minimum, add Product and Organization schema
Create comparison content: Build at least 3 detailed comparison pages
Optimize existing content: Add explicit value propositions and structured data to top pages
Pitch 5 earned media opportunities: Target publications that AI engines already cite
Month 3: Scale & Systemize
Launch ongoing PR campaign: Build relationships with key journalists and publications
Create content framework: Develop templates for machine-scannable content
Set up monitoring: Track mentions across AI engines (consider tools or agencies specializing in this)
Expand schema coverage: Roll out comprehensive markup across entire site
Why This Matters More Than You Think
The researchers conclude with this warning:
"The complexity and competitiveness of the generative search environment have rendered ad-hoc SEO tactics obsolete. The only path to sustainable dominance is through a principled, disciplined, and continuous GEO methodology. This is not a one-time project but an essential managed service—an ongoing arms race where victory belongs to those with the best intelligence, the most impactful content, the strongest authority, and the fastest reaction time."
In plain English:
This isn't a "set it and forget it" project. AI search is evolving rapidly, and your competitors are reading this same research. The window for early-mover advantage is closing.
Companies that treat GEO as a continuous discipline—like they treat traditional SEO, PR, or content marketing—will dominate their industries in AI search. Those that treat it as a one-time optimization project will be invisible.
Important Limitations to Keep in Mind
The researchers were transparent about their study's limitations:
Temporal snapshot: This data is from August 2025. AI search engines are evolving rapidly. What's true today may shift in months.
Black box problem: We can see what AI engines output, but not exactly why they make their choices.
Classification subjectivity: The "Brand vs. Earned vs. Social" categories are useful frameworks, but not absolute truths.
What this means for you: Don't treat these findings as permanent laws. Instead, use them as current best practices while maintaining the flexibility to adapt as the landscape evolves.
Final Thoughts: The New Search Reality
The transition from traditional search to AI search isn't coming—it's here. The question isn't whether to adapt, but how quickly you can move.
The good news: Unlike traditional SEO where Google holds a monopoly, the AI search landscape is fragmented across multiple engines with different strengths. This creates multiple paths to visibility.
The bad news: That same fragmentation means you can't just "win SEO" through one strategy. You need engine-specific tactics, language-specific approaches, and continuous adaptation.
The opportunity: Most companies are still treating this as a future problem. Early movers who implement comprehensive GEO strategies now will build advantages that become harder to overcome as the space matures.
At The 66th, we specialize in helping companies navigate this exact transition. Whether you're just getting started or ready to implement an enterprise-level GEO strategy, we're here to help translate cutting-edge research into practical results.
Want to learn more? Explore our case studies or contact our team to discuss your specific situation.
This analysis is based on "Generative Engine Optimization: How to Dominate AI Search" by Chen et al. (2025), University of Toronto. The research paper provides extensive quantitative data across thousands of queries and multiple AI search engines. All findings and statistics cited are drawn directly from their published research.