How Google Maps Keyword Fragmentation to User Needs in AI Search

Google’s Liz Reid, speaking on the Bloomberg Odd Lots podcast, detailed how AI Mode and AI Overviews are enabling more detailed, need-based queries. This shift, she explained, creates new challenges for Google and marks a fundamental change in search behavior that directly reshapes how SEO must be approached.

Keyword fragmentation is at the core of this evolution. Reid noted that users have always wanted to express longer, more natural queries but were forced to condense them into short phrases like “best restaurants in New York.” The real need might have been far more specific: a vegan-friendly restaurant with availability for a party of five. For nearly three decades in SEO, keyword research has been the bedrock of digital marketing. You pick a target keyword, then build content optimized around it. The hidden problem with short keyword phrases has always been their latent meanings. Google historically used click data to decipher ambiguous queries like “restaurants in New York.” While some SEOs believed clicks directly influenced rankings, another critical use was understanding user intent. Google ranked the most popular interpretation of a keyword first, meaning that even a well-linked page could fail to rank if its content aligned with a less common meaning.

Reid explained that users of AI-based search now articulate their actual problems or information needs in longer queries. This change strikes at the heart of what organic search struggled with, and the implications for SEO are profound.

“We have seen with AI Overviews meaningfully longer queries,” Reid said. “We see more natural language queries… It can also be like you were searching for restaurants. People would just be like, ‘restaurants New York.’ And you’re like, what do you want me to do with that query? … Part of why people would do that is they had a much more complex need. I want a restaurant in this location for five people. It can’t be too pricey. I have a vegan member. I also have kids. In the old world of keyword-ese, that information would be spread throughout the web. And now with AI Overviews and AI Mode, you can start to actually… tell the computer your real need.”

Several key takeaways emerge from this. First, a complex question asked in AI Search may not be fully answered by a single web page. Second, these complex queries are often one-off and rarely repeated, potentially lowering the value of optimizing for those exact phrases. The time spent crafting content for such specific phrases might be better used elsewhere. Third, because a site will likely share AI Overview space with another site, it becomes crucial to optimize other factors such as distinctive brand icons, relevant images, and even videos to claim as much AIO real estate as possible.

Perhaps the bigger insight is that it’s not simply about long-tail keywords. Google’s AI breaks down lengthy natural language queries into smaller, highly specific keyword phrases that reflect a portion of the information need. This process, called query fan-out, fires those smaller queries off to classic search. Google’s AI then picks from the top three results for each sub-query and synthesizes an answer. So SEOs should not exclusively chase long-tail queries. Because query fan-out relies on classic search, relevance still depends on the specific keywords that web pages are already optimized for.

Addressing real needs is another critical point Reid touched on. The process of breaking a complex natural language query into smaller parts becomes a quality issue. Because people aren’t searching with the same keyword phrases, Google cannot cache similar queries as efficiently as it can with organic search.

“I think it means you have to do a harder job on quality,” Reid explained. “You have to take this question, there’s many parts, and figure out how you break it apart. You have to think about things like latency, because if everyone uses the same keyword and it’s not personalized, you can cache it all. If queries get much more diverse, it has consequences. But I think we just see that it’s very empowering people. It takes some of the work out of searching.”

On the surface, “addressing the user’s real need” might sound like an empty slogan. But it is a practical framework every SEO should use when auditing web pages. Instead of limiting scope to keywords, headings, or technical issues, ask how a page fills a specific need. Someone recently asked me to review their site struggling with indexing, suspecting a technical glitch. My response was that while everyone hopes for a technical fix, the real problem often becomes clear when you ask, “What need is this page filling?” and “How is this page not just different, but different and better?”