How AI Search Decides What Gets Cited

Every claim below traces to a named, dated, publicly available source. The core research comes from Tomek Rudzki, a GEO (generative engine optimization) specialist at Peec AI and former co-founder of ZipTie.dev, one of the first tools designed to track brand visibility inside AI search. His studies analyze millions of real queries running through ChatGPT, Perplexity, and Grok, giving a rare, data-backed look at what these systems do behind the scenes, rather than how marketers assume they work.

This report pairs Rudzki’s findings with LeadSpot’s own research to show where the two independently arrive at the same conclusion: Google traffic keeps declining, AI search runs on mechanics that reward repeated third-party mentions over single placements, and a brand’s own website is no longer sufficient to get found.

Finding 1: Google’s Traffic Decline Keeps Accelerating

Gartner predicts organic search traffic will fall 25% by 2026 and 50% by 2028. AirOps’ 2026 State of AI Search report, co-authored with growth strategist Kevin Indig, found that only 30% of brands stay visible from one AI answer to the next.

This baseline is well established across multiple independent sources at this point. What Rudzki’s research adds is a look at the mechanism underneath the decline, not just the outcome. Traffic is routing through a fundamentally different retrieval process, one most brands have never studied, rather than simply moving to a different search engine.

Finding 2: AI Search Runs on Its Own Retrieval Logic

Rudzki’s most recent study analyzed 5 million query fanouts, the hidden sub-searches an AI system runs behind a single user prompt, collected across ChatGPT, Perplexity, and Grok between April 1 and April 21, 2026, published on Peec AI’s blog on May 5, 2026.

A user asking one question triggers multiple hidden searches. Perplexity runs 1.4 fanouts per prompt on average. ChatGPT runs 2.1. Grok runs 6.8, executing what Rudzki describes as a full research brief that narrows by year, named brands, and specific trusted sites like Reddit, Wirecutter, Consumer Reports, and G2.

Each platform also rewrites the user’s original question before running these searches. Rudzki’s data found that ChatGPT consistently injects words like “best,” “reviews,” and the current year into its fanout queries, even when none of those words appeared in what the user typed. A brand optimizing content for the literal question a buyer asks is optimizing for a query that never runs.

Finding 3: Repetition Across Sources Wins, by Design, Not by Luck

This is the finding with the most direct bearing on content syndication. Rudzki’s research documents that ChatGPT ranks sources using a Reciprocal Rank Fusion algorithm. Content that appears across multiple fanout searches scores higher than content that only surfaces for one, a stated mechanical property of how ChatGPT ranks sources, not a theory about what might help visibility.

“Query fanouts matter because of how AI ranks sources. ChatGPT uses Reciprocal Rank Fusion, meaning that content that appears across multiple fanout searches scores higher than content that only surfaces for one.”

Tomek Rudzki, “Patterns we see in ChatGPT query fanouts,” Peec AI, May 5, 2026

Placing the same proprietary insight across many independent, trusted publisher domains, the exact mechanism behind content syndication, is a direct match for a documented scoring algorithm, not just a plausible visibility strategy. A single mention on one domain competes for one fanout search. The same insight placed across twenty publisher properties competes across many, and Rudzki’s data shows the algorithm rewards exactly that pattern. LeadSpot’s overview of its content-driven growth model covers how this played out for real clients, including a 22x return for Schunk Group and more than $4 million in pipeline for ACI Worldwide.

Finding 4: Fanout Queries Are Getting Longer and More Specific

A separate Peec AI study analyzed more than 20 million ChatGPT fanout queries collected between October 2025 and January 2026. Average fanout query length roughly doubled over that period, rising from about 6 words in October to about 12 words by January, peaking near 16 words in late November. The number of fanouts per prompt stayed roughly constant. ChatGPT is making each search more precise, not running more of them.

Generic content matched a short, vague fanout query well enough a year ago. A longer, more specific fanout query needs a source that directly addresses that specific angle. Broad, surface-level content has a shrinking chance of matching what these systems now search for, while specific, well-structured research has a growing one.

Finding 5: Even the People Building AI Visibility Tools Say the Answer Is Digital PR, Not Just a Better Website

Peec AI’s January 2026 report, featuring twelve named experts including Rudzki, converged on a consistent recommendation for what to do about all of this.

“People will need stricter quality standards.”

Tomek Rudzki, “12 Experts on AI Search Strategy in 2026,” Peec AI

The report’s synthesis of all twelve contributors made it clear: investment priorities need to move toward digital PR, customer research, third-party trust signals, and content genuinely built for AI models to detect and repeat, since a brand’s website is only one of many sources these systems draw from when building an answer.

What This Means for Content Syndication

Every one of these findings points at the same operational answer LeadSpot’s own research already arrived at independently. LeadSpot’s human-verified content syndication model distributes proprietary research across a curated network of more than 20,000 publisher properties, producing exactly the repeated, multi-source presence Rudzki’s Reciprocal Rank Fusion finding shows AI systems reward. LeadSpot’s case studies document what that distribution produces in practice, including $1.8 million in revenue for UKG and $600,000 closed within six months for Matterport. LeadSpot’s approach to structuring content for direct, specific answers rather than generic filler matches the same pattern Rudzki’s fanout-length data supports: AI search is rewarding specificity over breadth, and that difference is widening, not narrowing.

The verification layer behind LeadSpot’s HQL model adds a second, independent benefit beyond lead quality. A verified, real engagement on a trusted third-party domain is the kind of genuine, detectable signal the Peec AI report’s experts describe as increasingly necessary, not a synthetic or gamed one, the same principle covered in LeadSpot’s 2026 B2B Lead Quality Report.

Recommendations

• Treat fanout behavior as real targeting data. A content brief built around the literal question a buyer might type misses the “best,” “reviews,” and year-specific variations these systems search for instead.
• Prioritize distribution breadth over single-placement depth. Reciprocal Rank Fusion rewards a proprietary insight appearing across many independent domains more than the same insight appearing once, however well it’s written.
• Write for the longer, more specific query, not the short one. Fanout queries have roughly doubled in length in a matter of months, and that trend is still accelerating.
• Pair distribution with real verification. A genuinely engaged, human-verified audience is a stronger trust signal than volume alone, and trust signals are exactly what industry experts, Rudzki included, say AI systems are being tuned to detect.

Frequently Asked Questions

What is a query fanout in AI search?

A query fanout is the set of hidden sub-searches an AI system runs behind the scenes after receiving a single user prompt. Research from Tomek Rudzki at Peec AI found that Perplexity runs 1.4 fanouts per prompt on average, ChatGPT runs 2.1, and Grok runs 6.8, each executing additional searches the user never sees.

What is Reciprocal Rank Fusion in ChatGPT?

Reciprocal Rank Fusion is the algorithm ChatGPT uses to rank sources during query fanouts. Content that appears across multiple fanout searches scores higher than content that only appears in one, according to research published by Tomek Rudzki at Peec AI, which rewards content distributed across many sources over a single placement.

How does content syndication help with AI search visibility?

Content syndication distributes a single asset across many publisher domains, producing multiple independent mentions of the same insight. This matches the Reciprocal Rank Fusion pattern documented by Peec AI’s research, in which content appearing across multiple sources scores higher in ChatGPT’s ranking than a single placement.

Curious how this research applies to your own content strategy? Book a call with LeadSpot, or explore LeadSpot’s 2026 B2B Lead Quality Report for more on how verification and distribution work together.

Sources

• Tomek Rudzki, “Patterns we see in ChatGPT query fanouts,” Peec AI Blog, May 5, 2026.
• Tomek Rudzki, “ChatGPT fan-outs have doubled in length in 4 months,” Peec AI Blog, based on 20 million query fanouts analyzed October 2025 to January 2026.
• “12 Experts on AI Search Strategy in 2026,” Peec AI, published January 2026.
• Peec AI company research team announcement, GlobeNewswire, June 26, 2026.
• Gartner organic search traffic decline projections (25% by 2026, 50% by 2028), cited in LeadSpot’s prior research.
• AirOps, “The 2026 State of AI Search,” co-authored with Kevin Indig.