Why AI Search Ignores Your Website (And Cites Your Competitor Instead)

AI engines prefer certain source types - regarless of traditional SEO signals

When you ask an AI search engine for a recommendation, it does not crawl the web in real time and rank what it finds. It synthesizes a shortlist from sources it already considers authoritative, and it provides explicit reasoning for each one. The researchers hypothesized that this synthesis process is systematically biased, that AI engines prefer certain source types over others regardless of traditional SEO signals, and that those preferences vary meaningfully by platform.

If true, this would mean traditional SEO success is no longer sufficient for AI search visibility. A brand could rank number one on Google and still be invisible in the environments where buyers increasingly conduct research.

‍

How they conducted this experiment

The team ran large-scale controlled experiments across multiple product categories, languages, and query variations, comparing how AI search engines select and cite information versus how Google ranks it.

The four AI engines tested were ChatGPT, Claude, Gemini, and Perplexity. Content was classified into three types: Brand-owned (official brand pages and product sites), Earned Media (third-party professional reviews, industry publications, expert roundups), and Social (forums, user-generated content, video platforms).

Jaccard Similarity, a mathematical measure of overlap between two sets, was used to compare which domains appeared in Google results versus AI citations for identical queries.

‍

The results

AI search has a structural preference for earned media

The clearest finding in the data was the content type distribution. In the automotive category:

Social content was eliminated almost entirely. Earned media nearly doubled in share. This was not a marginal shift. It was a structural one, and it held across categories.

The reason is not arbitrary. AI systems are designed to provide neutral, trustworthy answers rather than promotional content. Independent reviews, professional publications, and expert roundups carry the kind of editorial authority that AI systems use to justify their recommendations. Brand-owned content, even when factually accurate, carries inherent promotional bias that these systems discount.

Google rankings and AI citations barely overlap

Using Jaccard Similarity, the researchers found that even at the top 10 positions, AI search citations share less than 50% overlap with Google rankings in most product categories. Ranking highly on Google is not a reliable predictor of being cited by AI.

This is the finding that changes the strategic calculus most sharply. Brands that have spent years optimizing for Google may have built almost no presence in the source ecosystems that AI engines actually draw from.

Each AI engine has a distinct source selection strategy

These are not minor variations. They represent different content philosophies. A strategy built for one engine may actively underperform on another.

‍

The four pillars of GEO

The researchers distilled their findings into four strategic requirements for visibility in AI search.

Machine scannability

AI systems parse websites like code, not like humans. They need structured data to extract information accurately. Schema.org markup for products, reviews, FAQs, and how-to content makes the difference between information that AI can use and information that remains invisible to it. Prices, specifications, ratings, and features all need explicit semantic markup.

Justification optimization

AI engines do not just recommend. They explain why. Content needs to explicitly support the reasoning an AI would use: clear value propositions, pros and cons lists, comparison tables organized by use case, and feature breakdowns tied to specific applications. The research found that content optimized for justification saw up to 40% improvement in AI response visibility.

AI-perceived authority through earned media

AI systems privilege sources they already trust. Building that trust requires third-party endorsements: professional reviews in established publications, expert roundups, industry citations, and Wikipedia references. The citation network matters as much as the content itself. Backlinks from domains that AI systems already cite frequently are more valuable than backlinks from domains that do not appear in AI responses at all.

Lifecycle content coverage

AI search operates across the full customer journey. If a brand lacks post-purchase support content, AI may recommend a competitor during troubleshooting queries even if the user already owns the brand's product. Comprehensive coverage across awareness, consideration, decision, and loyalty phases is required to maintain visibility at every stage.

Big brand bias is real but not insurmountable

For unbranded queries, AI systems default to market leaders. When someone asks for the best option in a category without specifying preferences, established names appear by default. Smaller or newer brands can overcome this through extreme category specificity: owning a narrow enough category that earned media authority outweighs brand recognition. The research shows this is achievable with sustained focus, but it requires patience.

Zero-click outcomes are accelerating

When AI synthesizes a complete answer, users often do not click through to any source. Visibility and traffic become decoupled. A brand can be cited in every AI response for its target queries and see no corresponding increase in site visits. This does not make citation irrelevant. It makes brand recall and influence the primary measure of success, with traffic as a secondary signal rather than the primary one.

‍

Wrapping things up

The researchers confirmed what the citation data implies: the rules of search visibility have changed structurally, not incrementally. Ranking on Google and being cited by AI are increasingly separate outcomes, driven by separate signals, requiring separate strategies.

The practical implications are specific. Earned media is not a nice-to-have for GEO. It is the primary input. Structured data is not a technical checkbox. It is the layer that makes all other content usable. And the measurement framework needs to change entirely: citation frequency, justification quality, and cross-platform consistency are the metrics that reflect actual AI search performance.

What the research does not recommend is abandoning traditional SEO. Seventy-six percent of AI citations come from pages already ranking in Google's top ten. The foundation is still relevant. What needs to be built on top of it is different.

The transition from SEO to GEO is not a future consideration. For brands in categories where AI search adoption is already high, it is the current operating reality. The question is not whether to adapt, but how far behind it is safe to start.