The minute ChatGPT launched to major acclaim in 2022, marketers had one question: “How do I get it to give the answers I want about my brand?” A natural question, given that for years influencing brand discoverability had been possible through SEO and SEM.
The answer at the time wasn’t easy or delightful to hear: you kinda can’t. That’s because the training material for an LLM was pre-baked into the model as it was learning. And that training material often came from typical web crawls. So if you already had good webcrawl best practices implemented on your site, it was likely going to pick up the basic signals you wanted to send.
But AI doesn’t stand still. Now ChatGPT, Perplexity, Gemini and other LLM chat platforms can access the internet to look things up, giving marketers another go at influencing the answer. In fact, Gartner is predicting that search engine volume will drop 25% by 2026 because of AI chatbots as well as other virtual agents.
Do you have questions about what this means for marketers? We certainly did. So we sat down with two of our experts on LLM models and data and system architectures, Principal Solution Architect Bobby Knapp and VP, Data Science & AI Alex Liss, and they broke it all down for us.
Huge: First off, what should we call this new kind of SEO — the kind where we’re focused on optimizing for LLMs?
Alex: I’ve started using Generative Engine Optimization or GEO because it feels the catchiest. I’ve also seen AIAO (AI Agent Optimization), but that’s just a mouthful and sounds a bit like Old MacDonald’s farm.
Bobby: I’ve seen a lot of names being floated out there including LLMO, SAIO, LEO, AEO and AIO. If my 15 years in technology have taught me anything, it’s that you can never have too many standards for the same thing. So I have proposed “Language Model Framework for Agentic Optimization” (LMFAO). Internal review is still pending, but I feel confident. ;)
Huge: OK so we have a name – Generative Engine Optimization. Why are people turning to generative engines like ChatGPT to search in the first place?
Bobby: Chat experiences built on large language models (LLMs) bring personalization to search but, more importantly, allow people to start their journey wherever they want. We’ve been trained for 30 years (all the way back to AOL!) to use keywords. But imagine if you could start your discovery from any point. Try searching Google with something like “Where should my family go on vacation for $5000?”—yet that’s a fairly normal search in a generative engine.
Huge: So LLMs are providing answers and starting to be asked to shop or accomplish more complex tasks? How could I, as a marketer, know if my site is LLM-friendly?
Bobby: I think the more important question is “How can I know if my brand is LLM-friendly?” In traditional search, only about 12% of queries end with a user clicking a result outside of the top 3. In this world, catering your site to search engines is the only way to break through the noise. GEO shifts the paradigm in two major ways: it selects what sources to use on behalf of the user; and it uses a much larger number of sources for each query.
Most generative search engines average in the range of 6-7 sources per search with each source being from a different domain. Even if your first-party content appeared in every relevant search, you only control about 15% of the narrative. More importantly, some studies have shown significant bias towards sourcing from content aggregators and user generated content with earned media being the single largest contributor.
Succeeding in GEO is about much more than optimizing content on your site. It’s about getting your brand out to as many channels as possible with as much positive attention as possible, particularly in the realms of earned media and UGC.
Alex: A major part of GEO is emphasizing user intent. This means site content will move from being indexed by functional descriptors of what it is, but contextual descriptors of what it does. A great example comes from the femcare brand Viv. Last summer, they pivoted their blog content to emphasize action-oriented language like “here is why Viv is your safest tampon choice” for women concerned about the presence of contaminants like lead and arsenic. This positioned Viv to take advantage of a trending topic going into generative AI search engines around sustainable tampons. Qualified visitors who came in through these AI searches converted at 4x higher than average.
As a general best practice, the open-source coalition Schema.org has an framework for enriching your site tags with LLM-optimized metadata that captures intent through action-oriented queries from LLM search. A richer schema for your site can deliver AI readability, capture the nuance of your content including different formats (images, videos, audio files), structure and narrative flow. There are also emerging standards like