How to Track AI Referral Traffic in GA4
Configure GA4 to isolate AI referral traffic from ChatGPT, Claude, Perplexity, and others. Step-by-step regex setup plus data limits explained.
Your GA4 dashboard is likely undercounting AI referral traffic by a significant margin. When an answer engine like ChatGPT or Claude sends a buyer to your site, that visit lands in Direct or Referral. This is a channel that's reshaping how your buyers research products as we speak, and you don't want it to get lumped in with other earned traffic.
This walkthrough configures GA4 to isolate that traffic into its own channel, so you can measure engagement and conversions against organic search instead of losing the signal.
Why GA4 misclassifies AI traffic by default
GA4 puts most AI referral traffic in the wrong bucket because of how it reads the HTTP referrer header. When mobile apps and browser policies strip the referrer, the session defaults to (direct) / (none) and disappears into the same pile as bookmarks and typed URLs. Link attributes can suppress it too.
Three mechanics drive the misclassification. Mobile apps open external links inside WebView (iOS) or Chrome Custom Tabs (Android), and neither propagates the host app's referrer. Link attributes like rel=noreferrer and browser Referrer-Policy headers suppress the header outright. GA4's Default Channel Group, which users can't edit, only recognizes a subset of AI platforms in the first place.
Vendor studies put Direct misclassification for AI-referred traffic around 65–82%, with one Loamly study of 446,405 visits finding 70.6% misclassified as Direct in GA4. An Attrifast analysis of 38 SaaS and ecommerce sites put ChatGPT sessions in Direct/(none) 65 to 82% of the time, with a median of 71%. These figures come from vendor-published site cohorts rather than population-representative research, but they converge on the same conclusion: the majority of AI referral traffic is invisible without manual configuration.
AI crawlers vs. AI referral traffic
Focus GA4 on human click-throughs. Track crawlers and user-triggered fetchers separately in server logs. Autonomous bots like GPTBot and ClaudeBot fetch pages to train or index models. They never fire GA4's JavaScript tags, so they never appear in your reports regardless of how you configure channels. Trying to capture them in GA4 is a category error.
Human referral traffic works differently, since a person reads an AI answer, clicks a citation link, and lands on your site through a browser that may or may not pass a referrer. That session fires your GA4 tag. That's the traffic this configuration isolates.
One nuance for server-log work later. User-triggered fetchers like ChatGPT-User, Perplexity-User, and Claude-User carry bot user-agent strings and fire only when a user asks a question. Count them separately from referral sessions in engagement reporting.
What GA4 now auto-categorizes
Google confirmed a native "AI Assistants" channel in GA4's Default Channel Group on May 13, 2026, with broad availability around June 7, 2026. Google's documentation defines it as the channel by which users arrive "from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok" and states plainly that it "excludes Google's AI Overviews and AI Mode." When a session matches, GA4 sets medium to ai-assistant and campaign to (ai-assistant).
As of mid-2026, the named platforms in the channel definition are ChatGPT, Gemini, Deepseek, Copilot, and Grok. Google's launch announcement also named Claude, but Claude never made the operative definition, so it may not trigger the native channel on its own. The custom group below catches it either way. Google expands this list over time, so treat these names as a snapshot, not a permanent set. Three gaps keep the custom group necessary:
- Perplexity is absent. It continues to land in the Referral channel, despite having the best referrer pass-through rate of any major platform.
- Google AI Overviews and AI Mode are excluded. Both merge into
google / organicand cannot be separated from standard search clicks. - GA4 does not apply the channel retroactively. GA4 will not reclassify historical data before May 2026.
Verify the current named list against Default Channel docs before you rely on it. Google names example sources but does not publish a complete enumerated list of matched domains, and it reviews requests for new sources "at least once per year and often more frequently." Because the default group can't be edited and misses Perplexity outright, a custom channel group remains necessary.
What you'll need before starting
You need Editor access at the property level to build a custom channel group. Google's channel group documentation is explicit: "You must be an Editor or above on the Analytics account at the property level to create and edit channel groups." Explorations have a lower bar. Under GA4 role permissions, Viewer role and above can create private explorations. New explorations are private by default, and sharing one requires at least the Analyst role under exploration sharing.
Assemble the referrer domains you'll track before you touch the interface. These are the verified domains across the six major platforms:
- ChatGPT:
chatgpt.com,chat.openai.com,openai.com - Claude:
claude.ai,anthropic.com - Perplexity:
perplexity.ai - Gemini:
gemini.google.com,bard.google.com - Copilot:
copilot.microsoft.com,bing.com/chat - Grok:
grok.com,x.ai,grok.x.com
How to track AI traffic in GA4 step by step
The process runs in two phases: build a non-destructive segment in Explorations to see the data first, then codify it into a custom channel group that appears in your standard reports. Explorations let you validate the regex against real sessions without altering how GA4 categorizes anything. The channel group makes the classification permanent and reportable.
Step 1: Build an AI traffic segment in Explorations
Start in Explorations because it changes nothing. Open a blank free-form Exploration, create a new session segment, and set the condition to Session source matching your AI regex. GA4 shows you which sessions qualify without touching your channel definitions, so you can confirm the pattern catches real traffic before committing to it.
Use this first look to estimate how much AI referral traffic your property already captures through a passed referrer. Whatever shows here is the visible minority. The Direct-classified majority won't appear, which is the point of understanding the data limits before you build reporting on top of them.
Step 2: Create a custom channel group
Go to Admin, then Data display, then Channel groups, and create a new group. Add a channel rule and name it something like AI Assistants. Set the condition field to Session source and choose the "matches regex" operator, which is where your assembled pattern goes.
Rule ordering is the failure point that catches most people. GA4 evaluates channels top to bottom, first-match-wins: "Traffic is included in the first channel whose definition it matches given the current order of channels in the group." Google's configuration guidance is direct: "reorder your channel list so that 'AI Assistants' appears above 'Referrals'." If a broad Referral rule sits above your AI rule, every AI session matches Referral first and your AI channel reports zero. Drag your AI channel above Referral, then click Apply. When a freshly built group still shows AI traffic under Referral, rule order is the usual cause, not the regex.
Step 3: Apply the AI referrer regex
Match this pattern against the Session source field. It covers all six platforms in a single rule:
^.*(chatgpt\.com|chat\.openai\.com|openai\.com|claude\.ai|anthropic\.com|perplexity\.ai|gemini\.google\.com|bard\.google\.com|copilot\.microsoft\.com|bing\.com/chat|grok\.com|x\.ai|grok\.x\.com).*Two syntax requirements govern GA4's RE2 regex. Escape literal dots as \. so the pattern doesn't treat them as wildcards. Separate domain alternatives with the pipe character | as an OR operator.
Extend the pattern as new engines emerge. When a platform starts sending measurable traffic, add its domain inside the parentheses with another pipe. The structure holds. You're only appending alternatives.
Step 4: View AI traffic in the Traffic acquisition report
Open Reports, then Acquisition, then Traffic acquisition, and switch the primary dimension to your custom channel group. The AI Assistants channel now appears as its own row alongside Organic Search, Direct, and Referral.
Prioritize engagement and conversion quality over raw session counts. Engagement rate tells you whether AI-referred visitors read the page. Session key event rate shows how often those sessions trigger a meaningful action. Conversions tie the channel to pipeline. Session volume is the least useful number here because referrer stripping suppresses it. The quality metrics are what hold up in a board conversation.
Compare AI traffic performance vs. organic search
Put AI Assistants and Organic Search side by side on engagement and conversion quality. Raw traffic comparisons mislead because GA4 structurally undercounts AI volume, but rate-based metrics are per-session and unaffected by the missing Direct-classified traffic. If AI-referred sessions convert at a higher key event rate than organic search, the channel is punching above its visible weight.
A buyer who arrives from an AI answer has already had your brand described and contextualized by the engine. That's a different entry point than a keyword click, and you usually see that difference in engagement metrics. Track the trend over quarters, not the absolute numbers in any single month, because the measurable share is growing fast. Total monthly AI-referred sessions grew 9.9x from November 2024 to May 2026 in one Previsible analysis, and ChatGPT accounted for 92.4% of trackable LLM referral traffic as of May 2026.
Understand dark AI traffic and data limits
No GA4 configuration recovers the majority of AI traffic, and you need to say that out loud before anyone builds a forecast on these numbers. Referrer stripping in mobile app environments is the reason. When ChatGPT's iOS app opens a link in WKWebView, the referrer header doesn't travel with it, and the session lands as (direct) / (none) no matter how precise your regex is.
Use pass-through rates to quantify the gap. Mobile app referrer behavior varies sharply by platform, and the differences explain why some engines are nearly invisible in GA4:
| Platform | Mobile app referrer pass-through | GA4 appearance |
|---|---|---|
| ChatGPT (iOS) | ~8% | (direct)/(none) |
| ChatGPT (Android) | ~11% | (direct)/(none) |
| Claude (iOS) | ~10% | (direct)/(none) |
| Claude (Android) | ~12% | (direct)/(none) |
| Gemini (iOS) | ~9% | (direct)/(none) |
| Perplexity (iOS/Android) | ~70% | perplexity.ai / referral |
Independent researchers have not verified these practitioner-measured figures, but the pattern is consistent. Perplexity is the outlier: roughly 70% of its mobile app sessions pass https://www.perplexity.ai/ as the referrer, the best mobile attribution behavior among major AI platforms. Set expectations accordingly. Your GA4 AI channel shows the measurable minimum, and Perplexity will be overrepresented in it relative to its actual share of AI traffic.
Google AI Overviews as a special case
AI Overviews is the hardest AI source to isolate because it passes no distinguishing referrer at all. When a user clicks a citation inside an AI Overview, the HTTP Referer is https://www.google.com/, identical to a standard organic search click. As one attribution spec puts it: "Clicks from a Google AI Overview citation present an HTTP Referer identical to a normal organic SERP click. There is currently no public referrer field that distinguishes them."
Google appends no UTM parameters either, so GA4 attributes these clicks as google / organic with no way to separate them from traditional search. Google Search Console's AI Overviews filter is currently the only official measurement tool for this traffic. If AI Overviews matter to your category, use GSC for that check.
Visualize AI traffic in Looker Studio
Connect your GA4 property to Looker Studio and set your custom channel group as a dimension to build a repeatable AI traffic dashboard. Once the channel group exists in GA4, Looker Studio reads it directly, so you can chart AI Assistants against Organic Search on engagement and conversion quality without rebuilding the segment each time.
The value is standing reporting instead of ad hoc pulls. A dashboard that refreshes on schedule turns a monthly manual export into a board-ready view your senior operator can maintain, and it makes the quarter-over-quarter trend in AI referral quality legible to a CEO who doesn't live in GA4.
Supplement GA4 with other tools
GA4 only captures the referrer-passing minority, so pair it with tools that measure AI visibility at the source. Four approaches close different parts of the gap.
- Bing Webmaster Tools AI Performance report: Microsoft's AI Performance report, launched in public preview February 10, 2026, shows how your content is cited across Copilot and Bing's AI summaries. It tracks citations even when users don't click, the AI equivalent of GSC impression data, and it doesn't depend on referrer headers. Microsoft defines a citation narrowly: its AI answer visibly references or shows your content. The metric does not represent traffic, clicks, or user engagement.
- Server log analysis: Filter log lines for known bot tokens like
GPTBot,ClaudeBot,PerplexityBot,OAI-SearchBot, and the user-triggered fetchers, then validate source IPs against vendor-published JSON files or Forward-Confirmed Reverse DNS. This is the only reliable method to estimate true AI mobile traffic volume, since referrer stripping defeats GA4 entirely. Watch for spoofing. Reported rates range from roughly 5-8% in SimilarWeb log analysis and 5.7% in HUMAN/Satori data to 30-60% in some weeks in ScaleGrowth analysis, which is why IP validation is the required verification layer. - UTM tagging for shared AI links: When you control a link that gets shared into AI contexts, tag it with UTM parameters so the session attributes correctly even if the referrer is stripped. UTM tagging works for links you control. Organic citations remain outside your control.
- Specialist LLM analytics tools: Specialist LLM analytics tools measure what GA4 has no concept of: prompt-level share of voice and brand mentions in AI answers. They also track answer sentiment. Otterly pricing starts at $29/month for 15 prompts. Semrush's standalone AI Visibility Toolkit is $99/month for one domain and 25 prompts and integrates with GA4 and GSC. Profound is enterprise-only with no self-serve tier, based on a Profound AI review. Many of these tools define "visibility" proprietarily, so treat cross-vendor comparisons with appropriate skepticism.
GA4 tells you what converted after the click. These tools tell you whether you're being cited before the click, which is the earlier and more strategic question.
Stitching GA4, server logs, Bing Webmaster Tools, and a specialist prompt tracker together is a lot of reconciliation to answer one question: are AI engines sending you buyers, and are they citing you before the click. GrowthOS runs citation tracking and content production as one loop, measuring AI visibility across Presence, Reputation, Perception, and Influence while GA4 keeps counting what converts after the click. If you're tired of reconciling five dashboards by hand, book a demo to see the consolidated view. Engagements start from $6,000/mo.