How to Track AI Referral Traffic GA4 Using Custom Channels

Google Analytics recently added an 'AI Assistant' channel category, but standard reports still often categorize generative AI visitors as generic Direct traffic, hiding your AI referral traffic GA4.

Data suggests the direct traffic AI platforms send looks identical to a user manually typing your URL into their browser. To capture these hidden visits, you must configure a new Custom Channel Group and use specific regular expressions (regex) to identify AI platform domains like ChatGPT and Perplexity.

We've found that to track ChatGPT traffic analytics, relying on default analytics metrics leaves gaps in your data. Google Analytics doesn't automatically isolate these visits, so manual segmentation is required.

If you've recently stared at a sudden, unexplained spike in Direct traffic while leadership demands answers, you know how frustrating this visibility gap feels. Analytics platforms misclassify website visits as 'Direct' traffic. Up to 60% of traffic labeled this way is organic, and all dark social and chat application referrals get misattributed because these platforms strip necessary referral data.

We'll walk through a 4-step framework for configuring custom channel groups, building explorations, and measuring AI traffic conversions.

Why segmenting AI search matters for your ROI

Proving the financial value of organic content is getting harder as search behaviors shift. Right now, B2B marketers cite attributing a return on investment to their content marketing as a primary measurement challenge. When AI chat engines intercept the user journey, they disrupt standard tracking models, but they also bring highly qualified visitors.

Traffic from AI platforms tends to convert at a significantly higher rate than other channels. When we ignore these visitors or let them blend into generic buckets, we lose the ability to justify the budget spent optimizing for generative engines.

Beyond missing attribution, unsegmented AI traffic distorts your wider reporting. Consider what happens when analyzing landing page performance and finding a high volume of '(not set)' sessions cluttering the data. Users click links from AI chats, read the content, and leave the tab open until the GA4 session times out. Timeout behavior creates messy, unassigned data that skews conversion rate calculations across the board. Segmenting this traffic cleans up your baseline metrics while proving to leadership how valuable AI-driven visits are.

Step 1: Compile advanced regex configurations for emerging LLMs

Accurate AI tracking requires catching every variation of their referring URLs. Building individual exact-match filters for every possible subdomain becomes an unsustainable mess. If you try to manually map every regional variant or mobile app URL, you'll miss data. Instead, we use regular expressions (regex) to capture the core domain patterns across all major AI engines.

In most cases, a unified regex for generative AI search catches regional subdomains and mobile app variants automatically.

Core regex patterns for top AI platforms

You need strings that account for both root domains and dynamic subdomains. These patterns catch the primary generative engines driving outbound clicks:

• ChatGPT: .*chatgpt\.com.*|.*openai\.com.*
• Gemini: .*gemini\.google\.com.*|.*bard\.google\.com.*
• Perplexity: .*perplexity\.ai.*
• Claude: .*claude\.ai.*

We usually combine these into a single string to create one unified channel rule. The piped format looks like this: .*chatgpt\.com.*|.*openai\.com.*|.*gemini\.google\.com.*|.*bard\.google\.com.*|.*perplexity\.ai.*|.*claude\.ai.*

Identifying new URL variants

AI platforms launch new features and spin up new subdomains constantly. A tool might shift its chat interface to a new URL structure overnight, instantly breaking rigid exact-match rules. You can protect your tracking by using regex to look for the foundational string anywhere in the referring URL.

To maintain this list, make it a habit to check the "Referrer" dimension in your standard traffic acquisition reports quarterly. Look for unclassified domains containing "ai", "chat", or "bot". When a new platform gains traction, test the exact URL format before adding its pattern to your master regex string. Keep the syntax clean. Overcomplicating the expression with unnecessary wildcards can accidentally pull in irrelevant referrers.

Step 2: Configure a custom channel group in Google Analytics 4

With your regex strings compiled, apply them inside Google Analytics 4. You can't modify the default channel grouping rules, so you must create a custom group to isolate this traffic.

A dedicated custom channel group prevents these AI visits from disappearing into generic buckets.

Building the custom channel

Navigate to the Admin panel in GA4, select Data Display, and click on Channel Groups. Create a new custom channel group and name it something clear, like "AI Referrals".

Within this group, add a new channel. Set the condition to match "Session source" using the "matches regex" operator. Paste the combined regex string from the previous step into the value field. The platform will now flag any session originating from those specific domains.

Managing sequential rule order

This part is tricky. Reportedly, GA4 processes custom channel groupings sequentially from top to bottom. A common frustration occurs when setting up this group and realizing the AI data still isn't populating correctly. The culprit is almost always rule order.

If your generic "Referral" channel sits higher in the list than your new "AI Referrals" channel, the system flags the Perplexity or ChatGPT visit as a standard referral and stops processing. It never reaches your custom rule. To fix this, drag your AI channel to the top of the priority list. The system checks for AI referrers first before dropping the remaining traffic into generic buckets. Apply the changes, and the new data segment will begin categorizing future sessions.

Step 3: Build custom GA4 explorations for AI traffic

Standard reports give a high-level overview, but they rarely provide enough depth to understand user behavior. Analyze the full journey of your AI visitors by building a custom Exploration in Google Analytics 4.

Constructing the free-form report

Open the Explore tab and select a Blank free-form report. Set your date range to cover at least the last 90 days to capture meaningful trends. In the Variables panel, import "Session default channel group" (or your newly created custom channel group) and "Landing page + query string" as your primary dimensions.

Next, import your core metrics. We recommend starting with Sessions, Engaged sessions, and Session conversion rate. Drop the channel group into the Rows section and your metrics into the Values section. You'll see a clear, side-by-side comparison of how your AI traffic stacks up against Organic Search, Direct, and Social.

Filtering noise and benchmarking performance

Raw traffic counts can be misleading. We've generally found that AI chat engines trigger automated link previews or bot crawls that inflate session numbers without delivering real human visitors. To filter out this noise, focus heavily on the "Engaged sessions" metric rather than raw sessions. An engaged session typically requires the user to stay on the page for a minimum duration, trigger a conversion event, or view multiple pages.

Once you filter the data, benchmark the AI channel group against your standard organic search traffic. Look closely at the session conversion rate. Because AI chat engines synthesize answers and provide specific source links contextually, visitors who click through usually arrive with higher intent. Comparing these two segments side-by-side validates whether your strategy for earning AI citations is driving qualified visitors or just vanity impressions.

Step 4: Analyze AI traffic quality and conversion rates

Traffic volume only matters if it drives business outcomes. After isolating the AI segment, tie those specific sessions to your primary conversion events.

Cross-referencing landing page performance

To find out which content formats perform best in generative engines, add "Landing page" as a secondary dimension to your AI traffic report. The new dimension reveals exactly which pages attract clicks from platforms like ChatGPT. Look for patterns in the winning pages. Highly technical guides, proprietary data reports, and deep-dive comparisons dominate these referrals. Identify the exact pages that successfully bridge the gap between an AI summary and a website visit, and replicate that structure in future content briefs.

Proving downstream business value

Once you connect the landing pages to specific conversion events—like a software demo request or a newsletter signup—the real value of this traffic becomes clear. Visitors from top AI chat platforms often convert at higher rates, outperforming the average of standard organic search.

To make this data actionable for leadership, export the metrics into Looker Studio. A dedicated dashboard that visualizes quarter-over-quarter AI traffic growth alongside conversion numbers changes the conversation entirely. When a strategist pulls this specific report to justify the content marketing budget, the data proves that optimizing for generative search yields highly profitable, downstream conversions. The Looker Studio dashboard replaces visibility guesswork with concrete proof of return on investment.

Strategic next steps for AI traffic reporting

Custom channel groups replace a messy analytics problem with a clear visibility advantage. The transition from reactive guesswork to proactive tracking gives you the baseline data needed to measure real answer engine optimization efforts.

Since AI platforms change their domain structures, treat your regex configuration as a living document. We recommend scheduling a brief quarterly review of your referral reports to spot newly launched platforms or changing subdomain structures. Update your channel rules accordingly. Open your calendar right now and set a recurring 90-day reminder to check your Referrer logs and update your rules.