LLMs are new, and generative engine optimization is a hot topic. Clients want to know how they can dominate their share of the market, and you want to deliver results. But how will you prove it? How do you know if it’s even worth investing in? 

While there’s no shortage of tools that track AI visibility, the reality is that visibility alone isn't the goal; meaningful, engaged users are. And the best place to start evaluating LLM traffic? Google Analytics 4 (GA4).

GA4 is a free website analytics tool that most websites already have installed. LLM traffic is also (mostly) clearly defined by the source in GA4, making it easy to identify and analyze the users originating from chats. This can help you prove if you’re actually hitting your AI search KPIs with the GEO tools and features you're using.

How to use Google Analytics 4 (GA4) in your GEO strategy 

GA4 is important for your GEO strategy because it can show you if you’re getting valuable traffic from LLM citation clicks. This information can help you decide how much to invest in your GEO efforts and which content best aligns with users from this channel. 

For the most part, Google is pretty good at marking LLM traffic for us, so finding it in GA4 is straightforward.

To see how much traffic your site is getting from LLMs, go to GA4, go to Reports, and find the Traffic Acquisition report.

Out of the box, Google uses pre-defined channel groupings that were created in 2020 (at the latest). This means all of those LLMs that we want to track? They’re falling under “Unassigned” or “Referral” traffic.

Instead of using this dimension, you’ll need to change the primary dimension by clicking on the down arrow and selecting “Session source / medium.” This will give you the name of the site the user was on prior to visiting your site (most of the time).

From here, some people will automatically search the table to view the LLMs they care about, such as ChatGPT, Claude, Perplexity, and so on. 

I think this is a massive disadvantage. 

By adding a filter right away, you risk missing out on a potential LLM you hadn’t considered before. Different industries are using different LLMs and different LLM-driven tools. Lacking insight into those specific models could be a major missed opportunity.

So, what should you do instead? At least once a month, review the traffic acquisition report with Session source/medium as the primary dimension and review all listed sources.

As you go through all of your sources, look up the referral sites and see if there are any sites that are actually LLMs (or tools using LLMs) that you’d like to show up on.

Understanding how LLMs are referencing your site

The strategies you should lean into the most right now are the things that are already working. In this case, that means we should focus on the LLMs currently sending us traffic and determine what content we already have that is gaining their attention. 

To do this, we will apply a filter to isolate the engines we care about (including any of the engines you discovered in the last section) using the report filter. The filter will look something like this: Session source / medium is exactly chatgpt.com / referral, chatgpt.com / none, perplexity.ai / referral.

Then, we’ll add in the landing page dimension by clicking on the little blue plus sign next to the primary dimension name.

This information will give you insight into the content LLMs (and users) currently value.

To analyze this report, look for patterns in your landing pages. Are there specific folders that get more traffic? Specific words that show up in URLs most often?

Fun fact: If you add a filter in the search bar for a word or folder, a percentage will populate under the totals for each metric, which will show you a comparison to the site report totals.

For instance, on this site, there were 275 LLM sessions. It seemed that a good portion of traffic was landing on blog posts. But just to be sure, I added “blog” to the search bar and saw that 63% of all LLM sessions started on a blog post.

Now, there’s a good chance this site is being accessed/cited for queries or content beyond the pages that appear here, but these are the pages that matter, as they are actively sending traffic to the site.

Discovering the value of LLMs for your site

Most folks will stop there, though. They find the pages that are receiving traffic and try to optimize other pages in the same way, but there’s one more VERY important question to ask yourself: Is this traffic meaningful?

Are these people doing anything of value on your site? If not, maybe we don’t need to focus on these channels right this second. Maybe, instead, we focus on scaling the traffic sources we are seeing meaningful traffic from, or maybe we need to find a way to steer the LLMs to some of our more useful content (think middle-to-bottom-of-funnel content).

In the same report from earlier, look at the key events (and session key event rate) metric (assuming it's set up). It will tell you which LLMs and which pieces of content are sending the most valuable traffic to your site, i.e., which traffic is converting. 

Note: If you do not have key events set up, or you’re not seeing any key events, look at any of the other meaningful events you have set up, or, if you don’t have any events set up, look at engagement metrics and identify which sources and landing pages led to the most engaged audience.

This is where the real information comes into view. Looking at this data, I can see that ChatGPT is sending the most qualified traffic to our site, and this blog post is a leading piece of content.

So, what do we do now? Analyze and optimize this page for conversion, and analyze similar pages for possible GEO opportunities.

Turn analytics into action

As someone who spends a lot of time with data, I want to offer you this: analytics comes down to two things: trends and outliers.

The first thing you should be looking for in these reports is trends. Are certain words showing up in your page paths more often than others? Or in your page titles? Are there certain topics that top pages have in common?

If you can’t find any trends, is there a clear outlier? A page or two that perform oddly well?

Once you’ve identified the content that is performing (whether it’s a group of pages or a single page), you’ll take it one step further.

What is different about this page or group of pages? This is when we might start looking at page structure and elements, topics covered, EEAT signals, metadata, content style, etc.

The goal here is to identify what we think the differentiator is.

Create and test a hypothesis

Now that you have an idea of what causes some of your content to outperform the rest of your content, it’s time to run a test.

To do this, you’ll need to select a few similar pieces of content to test this idea out on. For instance, if your content on training new hires is doing well, you could choose other pages that cover new-hire topics (e.g., onboarding, integration, planning, etc.). 

Then, you make the change that you think led to better results on those specific pages.

Note: Resist the urge to apply these changes or run these tests on every single page on your site. Each page and site section is different and has its own objective; treating them all the same is a recipe for disaster (not to mention it takes away your control group).

Analyze your results

Running a test is only actually running a test if you analyze the results and ideally, you run them against your baseline metrics. To determine whether or not your hypothesis was correct, you must...

1. Set a baseline: Note your current LLM sessions, key event rate, and top landing pages.
   

2. Make targeted changes: One or two percent of your pages at a time, not your whole site.
   

3. Wait: Give it at least 30 to 60 days.
   

4. Review the same GA4 reports: Are you seeing changes in traffic patterns? Key event rates?

   
   

Don't expect overnight results. And honestly? Don't expect massive traffic increases either. The idea is to maximize what we already know is cited and turn it into higher-value traffic, not to game the system.

Moving forward with your GA4-informed GEO strategy

You don’t need to guess what LLMs want to see from your site. Start by analyzing what’s already working for you.

Start small: Pick one high-performing page or group of pages, identify what makes it successful, test those elements on similar pages, and measure the results over 60 to 90 days. Repeat the process as you gather more data. You can also look into which AI bots are crawling your site.

The goal isn't to chase every LLM trend. It's about understanding which LLM traffic actually matters to your business, then optimizing strategically for more of it.