How to Track LLM Traffic in Google Analytics 4 (GA4)

Most marketing teams have no idea how much traffic they are already receiving from ChatGPT, Perplexity, Claude, and Gemini. Not because it is not there — it is. But because their GA4 setup is not built to see it. Sessions from AI platforms arrive labeled as direct, disappear into the general referral bucket, or sit unattributed in data that no one has a reason to look at closely. The result is that a growing acquisition channel stays invisible inside the tool that is supposed to tell you where your visitors are coming from.

This is a solvable problem, and the fix is less complicated than most people assume. By the end of this guide, you will have a GA4 configuration that isolates LLM traffic by platform, tracks it across all your standard reports, connects it to conversion events, and sends you alerts when something significant changes. The entire setup takes under an hour. The strategic clarity it creates lasts as long as you run it.

Why LLM Traffic Is Harder to Track Than Standard Referral Traffic

Before walking through the configuration, it is worth understanding the specific behavior that makes LLM traffic easy to miss in GA4 — because that understanding shapes every decision in the setup that follows.

When a user clicks a link from a standard website, the browser passes the referring URL to the destination site. GA4 captures it cleanly as a referral session from that domain. LLM platforms behave differently, and inconsistently across platforms and contexts.

ChatGPT passes a clean referrer from chat.openai.com or chatgpt.com when a user clicks a link directly in the interface. But when a user copies a URL from a ChatGPT response and pastes it into their browser, or accesses it through certain mobile or embedded contexts, the referrer is stripped entirely. GA4 receives the session with no source data and assigns it to direct. The same session that would have been cleanly labeled as coming from ChatGPT instead disappears into your direct traffic volume.

Claude is more aggressive about stripping referrer headers. In most browsing contexts, sessions that originate from a Claude response arrive in GA4 as direct traffic regardless of how the user navigated. This means that brands seeing meaningful Claude engagement — showing up in Claude responses, being recommended to Claude users — may have almost no Claude-attributed traffic in their GA4 data despite real visits occurring.

Perplexity tends to pass referrer data more consistently than ChatGPT or Claude, but its mobile app and certain embedded interfaces behave differently from the desktop web version. Google Gemini has its own referrer patterns, and Microsoft Copilot routes through bing.com subdomains and edgeservices.bing.com in ways that do not obviously signal AI origin without specific filtering.

The practical consequence is this: your direct traffic volume is almost certainly inflated with LLM referrals that have lost their source data. A regex-based configuration can recover the attributable portion. UTM parameters can recover some of what regex cannot catch. But a meaningful slice will remain permanently dark — which is why understanding the scale of total LLM traffic requires combining GA4 data with a dedicated AI visibility tracking tool like RankTim, which measures citation and mention frequency upstream of the click.

What You Will Build in This Guide

The complete LLM traffic tracking setup in GA4 consists of six components that work together:

A Custom Channel Group that persistently categorizes LLM sessions by platform across all reports. A Traffic Acquisition regex filter for fast, on-demand views in the standard reports. A GA4 Exploration with three tabs built specifically for LLM traffic analysis. A UTM parameter convention that recovers dark LLM traffic through proactive tagging. Conversion event configuration that connects LLM sessions to business outcomes. And anomaly alerts that notify you when LLM traffic changes significantly.

Each component serves a different purpose, and together they give you the complete picture that no single configuration element can provide alone.

Step 1 — Verify Your GA4 Data Stream Before Building Anything

Filters and channel groups built on a broken or incomplete data stream produce misleading output. Before touching any configuration, verify that the foundation is solid.

Open analytics.google.com and navigate to your GA4 property. Go to Admin → Data Streams and select your web data stream. Check the Last 48 Hours panel — it should show recent activity. If it is empty or showing errors, the tracking tag needs investigation before proceeding.

Within the same data stream settings, confirm that Enhanced Measurement is enabled. You want page views, scrolls, outbound clicks, site search, video engagement, and file downloads tracked automatically. These events matter when you compare behavioral quality between LLM traffic and organic search traffic — engagement signals are how you demonstrate that LLM visitors are higher-quality, not just present.

Navigate to Admin → Data Filters and check for any active filters that might exclude valid sessions. Internal IP exclusions are common and legitimate — but developer traffic filters or geographic filters can accidentally suppress LLM referral sessions you need. Review active filters and confirm that none of them would remove sessions from AI platform domains.

Finally, open the Realtime report and confirm that sessions are flowing. If you can see active users and their sources in real time, your data stream is healthy and you can proceed.

Step 2 — Build the Custom Channel Group for Persistent LLM Categorization

The custom channel group is the most important element of this entire setup. Unlike a report filter that you apply and remove manually, a channel group persists across all standard reports and explorations automatically. Once built, every LLM session that comes into your property gets categorized into the right platform bucket without any additional configuration on your part.

Navigate to Admin → Data Display → Channel Groups and click Create new channel group. Name it "LLM Platform Traffic" — specific enough that any team member knows exactly what it represents.

You will create six channel definitions within this group. For each one, set the condition type to Session source and the match type to Matches regex.

Channel 1 — ChatGPT / OpenAI

Name: ChatGPT / OpenAI Regex: chat.openai.com|chatgpt.com|openai.com

Channel 2 — Perplexity

Name: Perplexity AI Regex: perplexity.ai

Channel 3 — Claude / Anthropic

Name: Claude / Anthropic Regex: claude.ai|anthropic.com

Channel 4 — Google Gemini

Name: Google Gemini Regex: gemini.google.com|bard.google.com

Channel 5 — Microsoft Copilot

Name: Microsoft Copilot Regex: copilot.microsoft.com|edgeservices.bing.com|bing.com/chat

Channel 6 — Other LLM Platforms

Name: Other LLM Platforms Regex: writesonic.com|bnngpt.com|nimble.com|you.com|phind.com|cohere.com|mistral.ai|poe.com|pi.ai|meta.ai

Save the channel group. From this point forward, every report in your GA4 property where you switch the dimension to this custom channel group will automatically break out LLM sessions by platform. This replaces the need to apply a filter every time you want to see AI traffic — the categorization is always there, waiting for you.

Step 3 — Add the Quick-Access Regex Filter to the Traffic Acquisition Report

The custom channel group handles persistent categorization, but you also need a one-click view in the standard Traffic Acquisition report for daily monitoring and stakeholder sharing. This is the filter you will reach for when you want a fast answer to "how much LLM traffic did we get this week?"

Navigate to Reports → Acquisition → Traffic acquisition. Click the filter icon at the top of the report — it looks like a funnel — and add a new condition:

• Dimension: Session source / medium
• Match type: Matches regex
• Value:

.*openai.*|.*chatgpt.*|.*copilot.*|.*gemini.*|.*perplexity.*|.*bard.*|.*claude.*|.*anthropic.*|.*edgeservices.*|.*writesonic.*|.*bnngpt.*|.*nimble.*|.*poe.com.*|.*phind.com.*|.*you.com.*|.*mistral.*|.*meta.ai.*|.*pi.ai.*

Click Apply, then use the Save option in the top right to save this as a custom report. Name it "LLM Traffic — All Platforms". This report will now appear in your saved reports list and can be accessed in one click without rebuilding the filter each time.

Step 4 — Build a Dedicated LLM Traffic Exploration

The saved report from Step 3 is useful for quick checks. The Exploration you will build here is where the real analysis happens — where you answer the questions that actually inform content and GEO strategy decisions.

Go to Explore in the left navigation panel. Click Blank to create a new exploration from scratch. Name it "LLM Traffic Deep Analysis — 2026."

In the Variables panel on the left side, import the following dimensions: Session source, Session medium, Landing page + query string, Country, Device category, New / returning, and Browser. Import the following metrics: Sessions, Engaged sessions, Engagement rate, Average session duration, Conversions (all relevant conversion events), Event count, and Bounce rate.

Now create three tabs within this exploration, each focused on a different analytical question.

Tab 1 — Platform Comparison

Set the visualization to Table. Rows: Session source. Values: Sessions, Engaged sessions, Engagement rate, Average session duration, Conversions. Apply your LLM regex as a segment filter to restrict data to AI platform sessions only. This tab answers: which LLM platforms are sending the most traffic, and how does user quality compare across platforms?

Tab 2 — Content That Gets Cited

Set the visualization to Table. Rows: Landing page + query string. Values: Sessions, Engagement rate, Conversions. Apply the same LLM segment filter. Sort by Sessions descending. This tab answers: which specific pages on your site are LLM platforms citing most frequently? These are your highest-performing GEO assets — the pages worth refreshing, expanding, and protecting.

Tab 3 — LLM vs. Organic Comparison

Set the visualization to Table. Create two segments: one for LLM traffic using your regex, one for Organic Search. Rows: Session source. Values: Sessions, Engagement rate, Average session duration, Conversions, Bounce rate. This tab answers: how does the behavioral quality of LLM-referred visitors compare to standard organic search visitors on the same metrics? This is the analysis that demonstrates the strategic value of AI visibility investment in terms your stakeholders can understand.

Save the exploration. Return to it weekly and use it as the primary analytical tool for your LLM traffic program.

Step 5 — Implement UTM Parameters to Recover Dark LLM Traffic

Even with a perfect GA4 configuration, a meaningful portion of LLM traffic will arrive as direct because the platforms stripped the referrer header before passing the session. UTM parameters are the only reliable way to recover this traffic for attribution.

The principle is straightforward: add UTM-tagged URLs to the content you publish, optimize for AI citation, and distribute in contexts where LLM platforms are likely to pick it up. When a user clicks a link from an AI response that carries your UTM parameters, GA4 captures the source regardless of whether the referrer header was passed.

Use this naming convention consistently across all LLM attribution UTMs:

utm_medium: Always use ai_referral for all LLM platform traffic. This is the most important parameter — it ensures that even when the specific source is unknown, you can still segment all AI referral traffic cleanly on the medium dimension.

utm_source: Use the platform name in lowercase with no spaces: chatgpt, perplexity, claude, gemini, copilot. When the specific platform is unknown or you are tagging for general AI distribution, use ai_referral as the source value as well.

utm_campaign: Use descriptive campaign identifiers that reflect the content type or initiative: geo_comparison_content, product_explainer_series, thought_leadership_q2_2026. This allows you to analyze which content campaigns are generating LLM-referred traffic, not just which platforms are sending it.

A fully constructed UTM URL for a comparison page looks like this:

https://yoursite.com/blog/comparison-page/?utm_source=chatgpt&utm_medium=ai_referral&utm_campaign=geo_comparison_2026

Where to apply UTM tags: add them to the canonical URLs you use in your internal linking for pages you actively optimize for LLM citation. Include them in any press releases, guest contributions, or directory listings where your content URLs appear. Add them to the URLs in your structured markup (FAQ schema answer links, for example) where relevant.

The combination of regex filtering capturing clean referrers and UTM parameters capturing the dark portion gives you the most complete LLM attribution picture currently possible in GA4. Neither approach alone is sufficient — together they cover most of the attributable traffic.

Step 6 — Configure Conversion Events to Measure LLM Traffic Outcomes

Session counts and engagement metrics tell you that LLM traffic is arriving and that users are interacting. Conversion events tell you whether that traffic is delivering business value. This step connects your LLM tracking to the outcomes your organization actually cares about.

Navigate to Admin → Events in your GA4 property. Review the events list and identify which events correspond to meaningful business actions for your site. Common conversion events to configure include: form_submission (contact forms, demo requests, quote requests), sign_up (newsletter subscriptions, account creation), purchase (for e-commerce), file_download (lead magnets, whitepapers), and phone_call_click (for service businesses).

Click the toggle next to each relevant event to mark it as a conversion. GA4 will then track conversion rate for these events across all traffic sources, including the LLM channels you have set up.

With conversions configured, open your LLM Traffic Deep Analysis exploration from Step 4. Add your conversion events to the metrics in each tab. Now you can see not just how many sessions each LLM platform sends, but how many of those sessions complete meaningful actions. The platform-level conversion rate comparison — does ChatGPT traffic convert better than Perplexity traffic on your specific conversion goals — is where LLM traffic analysis starts to directly inform strategic prioritization.

Set a calendar reminder to export this data monthly into a shared tracking spreadsheet. The month-over-month trend of LLM traffic volume and LLM conversion rate is one of the clearest indicators available of whether your GEO strategy is working.

Step 7 — Create Anomaly Alerts for LLM Traffic Changes

Your GA4 configuration is now built to see LLM traffic clearly. The final element is making sure that significant changes in that traffic do not go unnoticed between your regular review sessions.

AI model updates, new competitor content being indexed by LLM platforms, changes in how platforms retrieve and cite sources, and the publication of content that suddenly starts getting widely recommended can all produce sharp movements in your LLM traffic volume. Without alerts, you might not notice a 40% week-over-week change until your next scheduled report review — by which point you have lost the opportunity to investigate while the signal is fresh.

In GA4, navigate to the Insights section from the main navigation (represented by a lightbulb icon) or access Admin → Custom Insights. Create two custom insights:

Alert 1 — LLM Traffic Volume Change Condition: Weekly sessions from session source matching your LLM regex increases or decreases by more than 30% compared to the prior week. Notification: Email to the relevant team members.

Alert 2 — LLM Conversion Rate Drop Condition: Conversion rate from LLM traffic sources (medium = ai_referral or source matches LLM regex) drops by more than 20% compared to the prior four-week average. Notification: Email immediately.

These thresholds are starting points. Once you have two to three months of baseline LLM traffic data, adjust the percentage thresholds to reflect your actual week-over-week variance. A site with highly variable LLM traffic might need a higher threshold to avoid alert fatigue. A site with stable LLM traffic patterns might benefit from a tighter threshold.

Interpreting Your LLM Traffic Data: A Practical Reading Guide

Once the configuration is in place and data has been accumulating for four to six weeks, here is how to read what you are seeing.

High engagement rate from LLM platforms (above 65%) paired with longer average session duration than organic search confirms the pre-qualification effect. Users who arrive from LLM recommendations have already received context about your brand before they clicked — they know what you do and why they are visiting. High engagement rate and long session duration are the behavioral signature of that pre-qualification.

Conversion rate significantly higher than organic search is the metric that makes the business case for GEO investment. If LLM-referred visitors convert at two to three times the rate of organic search visitors — which is consistently observed for brands with strong AI visibility in commercial categories — that premium justifies disproportionate investment in content and external coverage that drives AI citation.

Most LLM traffic landing on comparison pages or detailed explainers rather than homepage or product pages tells you which content format AI platforms prefer to cite in your category. This is directly actionable: if comparison content drives the most LLM traffic, create more comparison content and expand existing comparison pages.

One platform dominating your LLM traffic while others contribute little signals a platform-specific gap. If 85% of your LLM traffic comes from Perplexity and almost nothing from ChatGPT, your content may be structured in a way that Perplexity's retrieval mechanism responds well to but ChatGPT's does not. This gap points toward a content structure or external citation deficit worth investigating.

Very low overall LLM traffic volume despite having the tracking set up correctly usually means one of two things: either your content is not being cited by LLM platforms for the queries your audience asks, or a large proportion of LLM referrals are arriving as direct traffic because of referrer stripping. In the first case, AI visibility optimization is the priority. In the second case, UTM parameter coverage needs to be expanded.

Connecting GA4 LLM Data to Your AI Visibility Strategy

GA4 LLM tracking shows you the downstream outcome: sessions, engagement, conversions, platform breakdown. It is an excellent measurement of what happened after a user clicked through from an LLM response. What it cannot tell you is what is happening upstream — which specific prompts your brand appears in, how you are positioned relative to competitors in those responses, which external sources are driving competitor citations, and how your overall AI visibility is trending.

That upstream layer is what RankTim is built for. Where GA4 answers "how much LLM traffic did we receive and what did it do," RankTim answers "why is that traffic the volume it is, and what would it take to grow it." The two tools are complementary — GA4 for outcome measurement, RankTim for visibility measurement and optimization targeting.

If your GA4 data shows that LLM traffic is low, RankTim will tell you which prompts you are missing from and which competitors are appearing there instead. If your GA4 data shows strong LLM traffic from Perplexity but almost nothing from ChatGPT, RankTim will show you whether that gap reflects a citation rate difference or a click-through rate difference — a meaningful distinction that points toward different solutions.

Maintaining Your LLM Tracking Configuration Over Time

The AI platform landscape changes faster than almost any other channel in digital marketing. New platforms emerge, existing platforms update their referrer behavior, and the sources that LLM platforms prefer to cite in your category shift as models update and new content is indexed. A configuration that was complete and accurate in Q1 may be missing new platforms or failing to categorize reclassified referrers by Q3.

Build a quarterly review of your LLM tracking configuration into your standard operations calendar. In each quarterly review, run your Traffic Acquisition report with no filters applied and sort by sessions. Look through the top 100 source entries for any domains that look like LLM platforms but are not captured by your current regex or channel group definitions. If you find any, add them immediately.

Check whether any platforms you are already tracking have introduced new domains or subdomains. LLM platforms occasionally change their interface URLs or introduce new app domains that pass referrer data under a different hostname than your current regex expects.

Review your UTM parameter coverage across your most important pages. Any high-value page that has been published since your last review should have UTM-tagged internal links if it is a potential LLM citation target.

Update your conversion event configuration if your conversion funnel has changed. New landing pages, new signup flows, or new product pages that have become conversion entry points should have the relevant conversion events configured and marked in GA4.

The quarterly maintenance routine takes thirty to forty-five minutes when done systematically. The compounding value of accurate, up-to-date LLM tracking data makes that time investment one of the most efficient operations in your analytics program.

LLM Traffic Tracking FAQs

Why does most of my LLM traffic show up as direct in GA4?

Several LLM platforms — particularly Claude and mobile interfaces for ChatGPT — strip the HTTP referrer header before passing the request to the destination site. GA4 receives the session with no source information and assigns it to direct by default. This is a browser and platform behavior that cannot be fixed with GA4 configuration alone. UTM parameters added to your content URLs are the only reliable way to attribute these sessions correctly, because UTM data is carried in the URL itself rather than in the referrer header.

Does GA4 automatically track traffic from ChatGPT without any setup?

GA4 will capture and attribute sessions that arrive with a clean referrer header from chat.openai.com or chatgpt.com as referral traffic from those domains. Without a custom channel group or report filter, however, those sessions are mixed into your general referral bucket alongside every other non-search referrer. The setup in this guide makes LLM sessions visible as a distinct, platform-broken-out channel rather than an invisible subset of referral traffic.

How often should I check my LLM traffic data?

Weekly is the right cadence for most brands. LLM traffic can shift meaningfully within days of a model update, a new piece of competitor content being indexed, or a change in how a platform cites sources for a particular query type. Anomaly alerts (Step 7) handle the in-between periods by notifying you of significant changes as they happen. The weekly review session is for trend analysis, not just anomaly response.

Should I create a separate GA4 property just for LLM traffic?

No. Separating LLM traffic into its own property fragments your data and eliminates the ability to compare LLM traffic against organic search, direct, and paid channels within the same reporting context. The custom channel group approach gives you complete isolation of LLM traffic when you need to focus on it, and seamless integration into your overall traffic picture when you need the full view.

My LLM traffic is very low after setting up the filters — is something wrong?

Check two things. First, verify the data stream is healthy by looking at the Realtime report for live activity. Second, check whether your site has content that LLM platforms are likely to cite — a site with no content optimized for LLM citation will have very low LLM traffic regardless of how well the GA4 tracking is configured. If your tracking is confirmed to be working but traffic is genuinely low, that is a content and AI visibility problem, not a tracking problem. RankTim can show you which queries in your category your brand is missing from and which competitors are appearing there instead.

How do I know if the regex filter is capturing all LLM platforms?

Run your Traffic Acquisition report with no filters applied, sort by Sessions descending, and scan through the source entries looking for any LLM platform domains that your current regex is not matching. Do this check quarterly — new platforms and new domain variants appear regularly. If you find uncaptured sources, update both your custom channel group definitions and your Traffic Acquisition regex filter to include the new patterns.