Google has been doing what it always does, and that is to constantly iterate to try and retain the best product it can.<\/p>\n
Large language models (LLMs) and generative AI chatbots are a new reality in SEO, and to keep up, Google is evolving its interface<\/a> to try and cross the divide between AI and search. Although, what we should all remember is that Google has already been integrating AI in its algorithms for years.<\/p>\n Continuing my IMHO series and speaking to experts to gain their valuable insights, I spoke with Ray Grieselhuber<\/a>, CEO of Demand Sphere and organizer of Found Conference. We explored AI search vs. traditional search, grounding data, the influence of schema, and what it all means for SEO.<\/p>\n \u201cThere is not really any such thing anymore as traditional search versus AI search. It\u2019s all AI search. Google pioneered AI search more than 10 years ago.\u201d<\/p>\n Scroll to the end of this article, if you want to watch the full interview.<\/p>\n Why Grounding Data Matters More Than The LLM Model<\/p>\n The conversation with Ray started with one of his recent posts on LinkedIn<\/a>:<\/p>\n \u201cIt\u2019s the grounding data that matters, far more than the model itself. The models will be trained to achieve certain results but, as always, the index\/datasets are the prize.\u201d<\/p>\n I asked him to expand on why grounding data is so important. Ray explained, \u201cUnless something radically changes in how LLMs work, we\u2019re not going to have infinite context windows. If you need up-to-date, grounded data, you need indexed data, and it has to come from somewhere.\u201d<\/p>\n Earlier this year, Ray and his team analyzed ChatGPT\u2019s citation patterns, comparing them to search results from both Google and Bing. Their research revealed that ChatGPT\u2019s results overlap with Google search results about 50% of the time, compared to only 15-20% overlap with Bing.<\/p>\n \u201cIt\u2019s been known that Bing has an historical relationship with OpenAI.\u201d Ray expanded, \u201cbut, they don\u2019t have Google\u2019s data, index size, or coverage. So eventually, you\u2019re going to source Google data one way or another.\u201d<\/p>\n He went on to say, \u201cThat\u2019s what I mean by the index being the prize. Google still has a massive data and index advantage.\u201d<\/p>\n Interestingly, when Ray first presented these findings at Brighton SEO in April, the response was mixed.\u00a0\u201cI had people who seemed appalled that OpenAI would be using Google results,\u201d Ray recalled.<\/p>\n Maybe the anger stems from the wishful idea that AI would render Google irrelevant, but Google\u2019s dataset still remains central to search.<\/p>\n It\u2019s All AI Search Now<\/p>\n Ray made another recent comment online<\/a> about how people search:<\/p>\n \u201cHumans are searchers, always have been, always will be. It\u2019s just a question of the experience, behavior, and the tools they use. Focus on search as a primitive and being found and you can ignore pointless debates about what to call it.\u201d<\/p>\n I asked him where he thinks that SEOs go wrong in their approach to the introduction of GEO\/LLM visibility, and Ray responded by saying that in the industry, we often have a dialectical tension.<\/p>\n \u201cWe have this weird tendency in our industry to talk about how something is either dead and dying. Or, this is the new thing and you have to just rush and forget everything that you learned up until now.\u201d<\/p>\n Ray thinks what we should really be focusing on is human behavior:<\/p>\n \u201cThese things don\u2019t make sense in the context of what\u2019s happening overall because I always go back to what is the core instinctual human behavior? If you\u2019re a marketer your job is to attract human attention through their search behavior and that\u2019s really what matters.\u201d<\/p>\n \u201cThe major question is what is the experience that\u2019s going to mediate that human behavior and their attention mechanisms versus what you have to offer, you know, as a marketer.<\/p>\n \u201cThere is not really any such thing anymore as traditional search versus AI search. It\u2019s all AI search<\/a>. Google pioneered AI search more than 10 years ago. They\u2019ve been doing it for the last 10 years and now for some reason everyone\u2019s just figuring out that now it\u2019s AI search.\u201d<\/p>\n Ray concluded, \u201cHuman behavior is the constant; experiences evolve.\u201d<\/p>\n Schema\u2019s Role In LLM Visibility<\/p>\n I turned the conversation to schema to clarify just how useful it is for LLM visibility<\/a> and if it has a direct impact on LLMs.<\/p>\n Ray\u2019s analysis reveals the truth is nuanced. LLMs don\u2019t directly process schema in their training data, but there is some limited influence of structured data through retrieval layers when LLMs use search results as grounding data.<\/p>\n