TRANSCRIPT

Hi everyone. I wanted to talk briefly about query fan outs. You may have heard this term, what is it? And how can we get ahead of this? So ultimately, there’s a relationship between AI search software, LLM software such as ChatGPT, Perplexity, Claude, etc. And of course Google’s own products, Google AI mode, a relationship between the software in the AI space and search engines. So for example, ChatGPT may have an ability to search Bing or Google, and of course Google AI mode searches its own Google Index. Because of this relationship between these two entities, ultimately we’re talking about a part of the process that involves search. What a query fan out.

Specifically, this is the process in which the LLM tool looks at the intent of your search prompt and maybe some additional signals as well, and figures out potential search queries to search the respective search index to get additional information around that topic from the web and ingest the content back from multiple sites. So getting ahead of this means can we predict what those potential queries could be and can we ensure that our website has content that speaks to those different topics, which will in turn maximize the chance of being cited in the final response? So that’s the concept. I wanted to show you a couple of cool things we can do to get a sense of how this can look. So we’ll start by jumping into ChatGPT and I’ll just run a quick search query here.

Best software to send money overseas for commercial purposes. I run that search in ChatGPT. Now, ChatGPT has gone ahead and identified a couple of different search queries and searched the web and returned some results before finally splitting back the details here. And what’s cool is that we have a, an ability to extract the real world query fan out terms that were actually spun up in the search. And we can do that by looking at a plugin which I’ll pull open over here and here you can see in this case two query fan outs that were searched on our behalf. And you can see it’s not an exact match to the specific term that I used. It adds additional modifiers, International payments platform, cross border, B2B payments.

Ultimately, ChatGPT is trying to understand the core of the intent of my prompt and then spinning up searches across the web that expand beyond the explicit language I used. So in this case it’s two searches and that would have powered the results that came back. So that’s real world data that we can actually extract from the data that’s sent back between my browser and ChatGPT servers and therefore we can really see what that looks like in the real world. We can’t do the same for other applications such as Google AI Mode, but we can use tools to guesstimate what they could be. So coming over here, this is a tool developed by Dayan, which uses Gemini’s model to estimate potential query fanouts.

It doesn’t guarantee that these are the ones that will be used by Google AI mode when you’re running a real world search, but at least it gives you a sense of what it could potentially be. So again, I’m going to put the same query in here. I’m going to click Generate fan out queries, wait a few seconds and eventually it comes back. Here we go with various expanded queries. In this case it’s a much longer list. Now again, like I said before, it’s not guaranteed that these will be the specific search queries that in this case Google AI mode runs on our behalf, but it can give us a good indication of potential queries.

And this is a good tactic for us to think about because we can say, hey, let’s make sure that we as a business, let’s say we sell solutions for international payments. We want to make sure we have content that addresses all of these different queries. We should be the authority on all these topics, having content that addresses the intent of each respective query here and that will hopefully increase our chance of success in being ingested into the AI response and being cited as a source, which hopefully leads to further discovery of our brand. Just want to show you that little pro tip. I’ll leave links to the two tools as I’ve shown today, I’ll leave them linked in the description and looking forward to hearing your thoughts. Thanks everyone.