What makes generative AI search different than regular SEO? Or is it different at all? Are these just all the same things? To understand the differences, and the differences really are subtle at this point, I think it requires going through the architecture of the systems. And when you look at the architecture of the systems at the most fundamental level, generative AI search doesn’t have an index, at least not in the same way search did before. There’s something kind of like an index, but it’s not at the center like it is in what we kind of understand as traditional SEO. And I think that’s interesting if there’s not this index at the center of everything. Because if AI search, generative AI search at least, doesn’t rely on a traditional index, then what are we actually optimizing for?
Because if you think about it, when you go through all the various SEO tactics and all the strategies that we have and that have been developed over the last several decades as SEO has formalized as a practice, it all comes down to one core idea. Get into the index and more specifically get into Google’s index. So we talk about how effectively can Google crawl pages? Well, that’s about crawling pages so that way Google can find those pages and bring them into the index. We talk about canonical tags to control duplications. That’s about controlling what pages are in the index, what pages are actually going to be considered by Google and potentially rank in search results. We have things like helpful content or just generally quality content guidelines.
And all of that’s about better indexing. Even just keyword optimization, the most probably basic SEO tactic there is. It’s about optimizing your keywords and the way in which you’re using the language on your page to describe things correctly so that way Google can index it correctly. Another example might be redirects. We’re using redirects to guide robots from one page that’s indexed to another page that we would rather have be indexed at a new URL. You can kind of go through a lot of other examples, but the main idea here is that everything we’re doing with SEO, or at least everything we have been doing with SEO, is about getting into Google’s index. And once our pages are in Google’s index, Google can then retrieve pages from the index that are related to a particular search query.
And then they can use what’s returned from the index to show 10 blue links, to show featured snippets, to show knowledge panels, carousels, right? all the other sort of non-generative search features on a search result page. And it’s obviously very sophisticated. It’s very complicated. There’s also a lot of AI in the mix to help build that index and to process all the pages and to determine what to pull out of that index in the first place. But my main point here is that at the core of everything in traditional non-generative, ranking-based search, it’s the index. And everything we’re thinking about is the index. And this is where generative AI search becomes something very different than that.
Generative AI search is powered by a large language model, an LLM. That’s what’s at the center of everything within generative AI search. It’s not an index. And that just fundamentally means generative AI search is going to work differently. Instead of an index, the LLM is making decisions about what is actually shown in the response to somebody’s prompt. Instead of pulling content out of an index and trying to retrieve pages that are listed in that index, the LLM is generating an answer on the fly, relying on lots of different sources. The primary source that they’re relying on is its own training data.
What was the large language model actually trained on? What does it know about the world based on all the information that was fed to it during the training process? That’s how it learned to model language and understand patterns in language. Other sources include a knowledge base. This is sometimes referred to as a RAG, or Retrieval Augmented Generation system. This is sometimes kind of thought of as an index and it is in a way that it’s a database containing some kind of information and it’s organized but it’s not at the center of everything. It’s an additional source that can optionally be brought in for the large language model to consider in order to respond to somebody’s prompt.
Large language models can also hit websites directly, right? It can do live website hits during figuring out how to respond to somebody’s prompt. There are also APIs and feeds and new agentic protocols that we can use to influence large language models. We have agent interactions more generally even outside of those protocols where they can go to a website and complete some kind of task on our behalf as part of that response.
The other big one that’s factoring into generative AI on some platforms more than others is the history of all the conversations. And that makes the responses in generative AI search a lot more personalized than maybe what we would have seen before. So the idea here is that the LLM is taking all of these sources in and the large language model is what’s figuring out, okay, based on everything that I’m looking at, based on all of these different inputs, how am I going to respond to this? So one way I I kind of like to think about this is that an LLM is basically an influencer, right? And our job is to influence the influencer. And that means optimiz optimization is going to shift from how do we get into Google’s index to how do we influence patterns? How do we influence relationships?
How do we make sure that we have really consistent information? So that way we can directly and sometimes indirectly reshape how the large language model is going to respond to a prompt about a product or service we’re offering or a prompt about our business. And you have to think about the website and online presence then as a whole not as isolated components that you have to optimize. It’s a different kind of mindset we have to use to approach things. And this difference in architecture I think also explains one of the biggest challenges with AI search which is why don’t we get traffic from AI search or at least why don’t we get as much traffic from AI search and it comes down to a large language model’s goal is to use all the sources available to deliver an answer that is it is not designed to send traffic to a website. It’s designed to deliver an answer to the person who entered the prompt. And that’s very different than what we see in non-generative search.
The index’s job is to give you a list of things that are really related to your search query. So that way you can click on those and go off and explore something else. It’s a very different kind of operation because of the very different way these things fundamentally work. You can think of it this way. If an AI response needs to generate a click, that means the large language model failed at its job by not delivering a complete answer. If a human has to complete a task that an AI agent started, the AI agent then failed at its job. And that’s really different than what we see in traditional ranking-based search.
So bringing all of this together, this means the goal of optimizing within AI search or generative AI search is about influencing the large language model so that our company’s website can be cited, summarized, mentioned, and ultimately just used by the large language model in that response. And that’s very different than how we’ve had to think about optimizing for traditional search. A lot of the tactics end up being the same. A lot of the actual things we’re going to do end up being the same, but they take on a slightly different view or lens of how we think about those tactics because of the different goals that these systems ultimately have because of their different architectures.
If you have questions about any of this, if you need help optimizing for both traditional and generative AI search, let me know. This is the kind of work that we do at Elementive to help our clients out and understand how to approach SEO, all sides of it, given all the different ways that it’s evolving. If you do want help or have questions, you can email me at matthewelementive.com. If you liked this video, please subscribe. Thank you for watching.