How to Dominate AI Search: What the Largest Comparative Study Reveals

Based on the paper "Generative Engine Optimization: How to Dominate AI Search" by Chen, Wang, Chen & Koudas (University of Toronto). Published in September 2025.

Paper: arXiv:2509.08919

The Question Everyone Is Asking

ChatGPT, Perplexity, Gemini — these AI search engines are progressively replacing Google for more and more users. But how do these systems choose their sources? Are they similar to Google, or do they work in radically different ways?

This paper provides the first large-scale empirical answers. Researchers at the University of Toronto conducted controlled experiments across multiple verticals, languages, and query reformulations to compare AI search with traditional web search.

The Protocol: A Massive Comparison

The researchers systematically compared:

• Google (traditional search)
• Multiple AI search engines (ChatGPT, Perplexity, Gemini)

Across multiple dimensions:

• What types of sources are favored?
• How diverse are the domains?
• How fresh is the content?
• How stable are results across languages?
• How sensitive are results to query reformulation?

Finding #1: The Massive Bias Toward Earned Media

This is the paper's most striking result. The researchers classify sources into three categories:

Google presents a relatively balanced mix across these three types.

AI search engines show an overwhelming and systematic bias toward earned media — authoritative third-party sources. Brand-owned content and social content are largely ignored.

Concretely: if you're a brand relying on your own website to be visible in AI search, you have a problem. AI engines prefer what others say about you rather than what you say about yourself.

Finding #2: Every AI Engine Is Different

Contrary to what one might think, AI search engines don't all behave the same way:

• Domain diversity — some engines cite many different sources, others concentrate on a few domains
• Freshness — some favor recent content, others don't
• Cross-language stability — the same query in English and French can yield very different sources
• Reformulation sensitivity — slightly rephrasing a question can completely change the cited sources

This means a GEO strategy that works on ChatGPT won't necessarily work on Perplexity or Gemini. Optimization must be engine-specific.

Finding #3: The Big Brand Bias

AI search engines tend to favor well-known big brands. Niche players and small businesses are systematically underrepresented in generated responses.

This is an interesting paradox when compared to the foundational GEO paper, which showed that GEO could help smaller sites. The reality is more nuanced: GEO can help smaller players gain visibility once they're in the pool of considered documents, but the initial bias of AI engines favors big brands for getting into that pool.

The 4 Strategic Pillars of GEO According to This Study

Based on their empirical results, the researchers formulate a 4-point strategic agenda:

1. Content Engineering for Machines

Your content must be scannable and justifiable by a machine. This means:

• Clear, verifiable claims
• Structured data
• Cited sources that the LLM can verify

2. Dominate Earned Media

Since AI engines massively favor third-party sources, the most effective strategy is to get others to talk about you. Press articles, mentions in authoritative publications, expert reviews — this is what AI engines value.

3. Engine-Specific and Language-Specific Strategies

There is no universal strategy. Each AI engine has its own biases and preferences. Likewise, behavior varies by query language. Optimization must be tailored accordingly.

4. Overcoming the Big Brand Bias

For niche players, the key is to build AI-perceived authority through credibility signals: citations, data, mentions in authoritative sources.

What This Changes Compared to Classic GEO

The foundational GEO paper focused on how to modify your own content (adding stats, citations, etc.). This study adds a crucial dimension: what matters most is what others say about you.

This is a paradigm shift. GEO is not just an on-page optimization exercise — it's also a public relations and authority-building strategy.

Practical Implications

• For brands: invest in PR and earned media as much as in optimizing your website
• For content creators: position yourself as an authoritative source in your niche
• For SEO professionals: GEO requires different strategies depending on the targeted AI engine
• For small businesses: focus on building authority through third-party mentions

Paper: Chen, M., Wang, X., Chen, K., & Koudas, N. (2025). Generative Engine Optimization: How to Dominate AI Search. arXiv:2509.08919