Is your SEO strategy ready for LLM grounding? Explore the distinction between training data and live web retrieval, and discover how to optimize your brand's visibility in AI search results.

Click on the whiteboard image above to open a high-resolution version!

Happy Friday, Moz fans. Today I want to talk about a concept that I think is going to be important for SEOs in, well, going forward actually. And that concept is the difference between in- and out-of-model responses.

So this is about LLM responses.  But it also very much applies to these sort of hybrid experiences that we see in Google search right now, such as AI Mode, AI Overviews, web guides. If you don't know what web guides are, I'll link a resource on that below. But I think that's going to be important as well.

Anyway, back on topic, in-model and out-of-model responses.

So there are basically two things that can happen (oversimplifying) when you make an LLM prompt. One of the first things a model is going to do is decide whether it needs to do any grounding searches.

So that means does it need to make any external queries, which normally means Google searches. Even with ChatGPT it normally means Google searches, at the moment, to decide whether it needs any fresh information, or whether it needs to check anything that it's not sure of.

So if it doesn't, then, for example, we might have a query, like "write me a poem," that doesn't refer to anything particularly factual or fresh or topical. So you would probably just get a response straight out of the LLM without it making any of these external grounding searches, and then you get a response.

So if you wanted to influence this response, the only thing you could do really would be to influence the underlying training data of the model.

So to give you some context, GPT-4.0 finished training in, I think, late 2022, and GPT-4.5 finished training in late, I think, maybe August 2024. So this is quite a long time span between model refreshes, and a lot of people are still using models which at this point are coming up to four years old or even more.

So your frame to sort of quickly influence this is nonexistent. And in the long term, you're basically trying to influence any of the content that goes into the training data for these models, which at this point is basically any written content whatsoever that ends up being ingested by a computer.

It's incredibly broad. And that obviously doesn't just mean your own site. That could mean external sites. But it can also, at this point, mean things like books. It's gotten incredibly, incredibly broad with this underlying training data.

I think what's more interesting for SEOs is out-of-model responses. So this is when the LLM, or the tool you're using, decides it is going to make some external searches.

So an example query would be, "What happened in the December 2025 core update?" So I tried this query in a few different tools. I'll share a link to this below, but in ChatGPT, you can actually, at the moment, see in Dev Tools in your browser what the grounding searches are.

And with some of these Google experiences, you can see what we assume are representative in terms of what it's doing with query fan-outs as well. So we have a rough idea of what some of the searches might be that are being made. It will be things like, in this case, did it exist? So the first thing would be confirming that there actually was a December 2025 core update, and I'm not just making it up.

Then what was the timeline? Then, who was impacted? Then what might the themes be? And it's obviously going to find various articles ranking in Google search for these kinds of queries. And often, these will not be the kind of queries that a human might type in, shall we say. They will be quite lengthy queries.

It will look at the first few results for each of those queries and maybe more, and it will start to formulate its answer from that external information.

Now, the interesting thing here is that the timeline could be quite quick in terms of you influencing this, right? It's basically just how quickly can the pages that you change get ingested by Google.

Obviously, that changes from site to site. In some sites, that's going to be minutes or hours, if you manage to influence a major news publication. In some sites, it might be a bit slower. But this is certainly much quicker than trying to influence underlying training data. 

In terms of how you do it, there are three things which I've put in what I would say right now, a sort of order of effectiveness. 

Barnacle SEO

I think the big one is barnacle SEO. So if you think about who's going to be ranking for these queries, it's not necessarily your own site. In fact, it's probably typically not mainly your own site that is ranking. So some of these external authoritative sites that you can influence, maybe with your own presence on them, think about your social profiles, think about Medium, think about LinkedIn, think about Wikipedia, think about YouTube. 

All of these things that you can influence, that you might historically have done to own a branded SERP or something like this, you would now be doing to try and own some of these non-branded terms and make sure that you control more than just your own results because Google or ChatGPT or whatever doesn't care whether it's looking at the data from your site or someone else's, right? 

So barnacle SEO is going to be a big one. 

Digital PR

Digital PR is also going to be very important in this space to try and influence some of those authoritative third-party sites that you don't directly control. 

Update your own site

And lastly, of course, your own site, because hopefully you do also rank for these queries. 

So you've got quite a sort of practical, and I would say familiar to SEOs, way to influence these responses. 

I think this is going to be an important distinction. So when you're thinking about whether it's realistic to try to influence a query, or even when you're thinking about what queries you might like to track in tools like Moz or STAT for your AI visibility tracking, I think you need to bear in mind which of these camps a query is going to be falling in.

That's it. That's it for me. I hope that was useful. Thank you.