AI Fanning Explained - what happens between user query and AI output and how to exploit it

Speaker A: So I've been obsessing over what happens between the user hitting send on chat and then the response that it gives back. What does the behind the scenes look like? This obsession started from me not being able to sleep. And after I was contemplating why gold prices are rising and how this is typically a signal of a world war, especially when it's central banks that are purchasing that gold stock. So immediately after that, I then started thinking about the underlying mechanism on how AI does research before it answers your query that you write, and how businesses can use this to their advantage. And that's what this episode is about. Today I'm basically just doing a solo episode explaining AI fanning, how it works, the mechanisms behind it, and really just everything I learned around it so that you can learn it too. As always, this episode is brought to you by Graph, the AI agent for marketing analytics. Graph is a data pipeline, data warehouse, and also a data visualization tool. Plug in all of your marketing data and get insights. Go to graph.com to start a free trial. Can't believe we got that domain. Still stoked. And so first, let's talk about AI search fanning. And I just want to explain what's going on here and how I'm conceptualizing it in my own mind. So first off, when a user does a query, the first thing that happens is basically the AI is enriching it. And the reason for this is a lot of the times the queries are just like pure lizard brain. It's like six words, which means nothing. Right? And so the AI basically has to ask itself, what are they actually asking in context? Is there a way that we can expand this? And so it basically gives itself that context. And then from there, what it then goes and does is it writes queries and goes and searches the Internet. Now it does all this in parallel, simultaneously. There's different figures that people are saying. I'm, um, of the belief it's around like 80 of these that are occurring. And what it's doing is it's going and doing these varied searches to aggregate information from the open web so that it can put it into a composite text. And how we know this is happening is that we can actually see this within the search console data of websites. Here's an example of what I mean by this. So, so what I've done is I've created a custom regex that's showing searches that have over 50 characters and you can see the data here. It's actually crazy, you know, search impressions of 3135 for this exact keyword phrase that's over 50 characters long. That just doesn't make sense. Right. Also zero clicks. So let's look over the aggregate data. Over the last six months, 96,000 impressions have occurred, but only 14 clicks. Again, none of this adds up. So what's happening here? AI is doing all of these queries. That's why there's no clicks occurring. And it's the exact same phrases that are being queried. It's seeing what's ranking on page one through three, it's scraping the URLs of those websites and then it's going directly to the blog page. So it looks like no click is occurring, but in reality it's taking into consideration that text into the context that it's about to use as the answer for whatever it is that it's going to show you. So let's go back to this visualization. We've now seen that AI is doing these queries, these searches on their own. We know what it's pulling in, it's creating this composite text. And then what you end up seeing as a user is the streaming write up. And this is the magic of the whole thing, is that everything that's basically happening in between this section here is all occurring behind the scenes relatively instantaneously. All you're feeling as the user is this and you're feeling this. And it's pretty magical from an experience. So now that we understand how this conceptually works, let's dive into how do you exploit this arbitrage. What is the strategy here? There is literally so many observability tools that are coming out currently. It's driving me absolutely insane seeing how many launch and how many VC dollars are basically going into this. We're not in an AI bubble, I swear. The only real way to make impact based off of this knowledge that we've just gone over together is to basically show up in the results for what it is scraping in that AI, uh, fanning process. Now. Now, to actually accomplish this, it's pretty straightforward. All you have to do is find what AI is referencing currently and get yourself included there. So how do you do that? You basically want to look at keywords that are related to your product and go use AI chats to see the text or see the URLs that's being referenced. You then take those URLs, you want to reach out to those blog post owners and you want to ask to be included. Listicles are the best. Get added to listicles and then it turns into surface area for your brand. So every company right now is trying to understand how do I basically get suggested through this AI search process. This is how you do it. It's by creating more surface area for the brand so that the AI incorporates that into it. But you don't need to just make more content. You need to make content that's ranking on page one through three. And the fastest way to do that is to find what's ranking currently. So that's an explanation of AI fanning. There are people that are way smarter about this than I am, but this is how I'm visualizing it and I wanted to share that with you. If you are one of those people that's smarter, I want to talk to you. I want to learn from you. Also, just comment below with your knowledge so that other people can learn from you as well. And if you're on Apple Podcasts, subscribe. If you're on YouTube, subscribe. And if you're on Spotify, subscribe. Those subscriptions are how we keep the lights on. I live to serve you. We'll talk to you soon. Cheers.