Copy the formatted transcript to paste into ChatGPT or Claude for analysis
Ranking number one, page one on Google
will not serve you well in AI search.
[music] The waiting of frequency is
higher than the waiting of ranking. What
these companies are doing, they're
writing prompts in this way to force AI
overviews, but then the underlying data
is absolutely incorrect. You have
written about how 32.5%
of the AI citations are list.
>> Mystical work phenomenally well like
especially if you're focusing on bottom
of funnel uh search terms. Mystical are
like your 100% [music] best friend.
People run their sites behind
AI-generated overview
James Barry, CEO of LLM Refs, discusses the fundamental shift from traditional SEO to AI search optimization in this interview with Navneet Kaushal. Barry emphasizes that ranking #1 in Google no longer guarantees AI visibility, as frequency of mentions across multiple sources now matters more than single top rankings. He explains that AI search operates through two mechanisms: training data inclusion and real-time search grounding, where LLMs break queries into fanout sub-queries and analyze hundreds of results. Barry advocates for keyword-based tracking over prompt-based approaches, noting that his bootstrapped platform focuses on share of voice and citation metrics rather than arbitrary visibility scores. Key technical recommendations include ensuring crawler accessibility (particularly for Cloudflare users), implementing server-side rendering, using FAQ schema markup, and creating listicle and comparison content. He reports that 32.5% of AI citations come from listicles, which work exceptionally well for bottom-of-funnel searches. Barry disagrees with both the notion that AI search and SEO are identical and that they're entirely different, positioning them as related but distinct disciplines requiring parallel investment.
In AI search, the weighting of frequency (mentions across multiple sources) is higher than the weighting of ranking position. Having citations at positions 2, 3, 10, and 15 is more valuable than ranking #1 alone without other mentions.
AI search operates as a two-part system: getting into foundational training data and optimizing for real-time search grounding, where LLMs break prompts into 1-5 fanout queries that each return ~100 results for the model to evaluate.
Listicles account for 32.5% of AI citations and work phenomenally well for bottom-of-funnel search terms because they're easy for AI to scan with clear H2 structures indicating numbered options.
The most common technical barrier to AI visibility is Cloudflare's default crawler blocking, which site owners often don't realize is preventing AI engines from accessing their content, even when users report finding them via ChatGPT through third-party sources.