@tjrobertson52

One of the biggest differences between doing SEO for classic search engines versus AI search is what keywords you target. This isn't the only difference, but it's one I wanna talk about today. Traditionally, when you're doing SEO, you target the keywords that humans are most likely searching for in search engines.

with AI search, it's not actually the humans searching, it's the AI. The human enters a prompt, sure, but then the AI takes that prompt and does what's called a query fan at and performs a series of 3 to 6 very specific searches based on the prompt. So if you wanna get recommended an AI search, your brand needs to be showing up for those search terms.

This is a big part of why there's so much opportunity in AI search right now. People aren't targeting these terms and they're very different from the terms that humans search for. There's already a ton of data on the terms that humans search for, which is why those terms are super competitive.

Very few people are collecting data on the terms that AI search for. And so if you have this data, it's very valuable. I want to talk about the three best ways to acquire it.

Most common method I see SEO is using right now is also I think the least effective. Trying to guess what the query found out might be based on Google search results. They're looking at the questions that Google includes in their people also ask section after performing a search.

The problem with this, of course, is that AI search much differently than humans and this isn't a new strategy. People have been doing this for years so they don't mirror AI searches and they're already competitive. The next method is to scrape the query fanout terms directly.

If you enter a prompt into ChatGPT, ChatGPT will actually expose the query fanout terms and you can retrieve those terms in the inspector. I was actually just on the Edward Stern podcast and he has another episode that shows how to do this. The problem is this only works in ChatGPT and older versions of Gemini.

So you won't be able to see the query fan out in newer versions of Gemini or in Google's AI mode. And just last week, someone shared a tool with me that they built that tries to approximate the query fan act. You can enter a prompt into the tool and then pick a model and it will ask the model what it would search for if you entered that prompt.

So you're not seeing the actual queries, but it does seem pretty close. I'll put a link to that tool in the caption. To capture the actual queries, you'll need to do this manually or use a paid tool like profound or what we use peaks dot AI.

But again, this only works for ChatGPT and older Gemini models. But it's this third method that I think is most effective right now. Bypass the query fan out altogether and just go straight to the pages that are cited.

While we can't see the query fan out directly, we can reliably track which pages are cited for a given prompt. Again, using a tool like peak or profound, you can track a large amount of prompts day after day to get a representative sample of the kinds of prompts people are entering and what pages are cited. If you then just look at the titles of those pages, you get a really good idea not only of what AI is searching for, but also what pages it ends up siding.

As of right now, this is absolutely the best data you can use for keyword research in AI search.