How to track AI referral traffic in Google Analytics

If you’ve recently spotted chatgpt.com or perplexity.ai appearing in your GA4 referral report, you’re seeing the trackable fraction of your AI traffic. Not all of it. Depending on how users arrived from those platforms, a substantial portion of those same visits may have landed in your Direct channel with no attribution at all.

This matters commercially. AI-referred traffic tends to convert at higher rates than organic search because users who click through from an AI response have already had their question partially answered. Misattributing that traffic to Direct skews your channel performance data and makes it harder to understand what’s actually driving results.

Most guides on this topic stop at the GA4 setup. This one goes further: it covers the three tracking methods worth implementing, explains why a significant share of AI traffic is invisible regardless of your configuration, and shows how to use the data you do have to make better decisions about content and AI visibility.

Why GA4 doesn’t show AI traffic by default

GA4 has no built-in AI channel. When a user clicks a link from ChatGPT, Perplexity, or Gemini, GA4 categorises the visit as either Referral (if the platform passes a referrer header) or Direct (if it doesn’t). Without manual configuration, AI traffic has no distinct identity in any standard acquisition report.

A referrer header is a piece of data a browser sends to a website identifying the previous page. If it’s present, GA4 logs the visit as Referral and shows the source domain. If it’s absent, GA4 has nothing to work with and defaults to Direct.

Google added a custom channel group regex example for AI assistants to its GA4 documentation in July 2025. That was the first time the platform officially recognised AI tools as distinct traffic sources. But it’s a user-configurable option, not a default. Out of the box, GA4 still has no AI category, and every standard acquisition report treats AI visits as generic referrals or direct traffic.

That’s what the three methods below are designed to fix.

The three methods for tracking AI traffic in GA4

There are three ways to surface AI referral traffic in GA4: a quick filter in the standard Traffic Acquisition report for a one-off check; a custom Explore report for trend analysis and multi-source filtering; and a custom channel group that makes AI traffic a permanent fixture in your reporting. The third method is the one worth implementing for ongoing use.

Method 1: Quick check using the Traffic Acquisition report

This takes under two minutes and works well when you want a fast snapshot of a specific platform.

• Go to Reports, then Acquisition, then Traffic Acquisition.
• In the search bar above the data table, filter by Session source. Enter chatgpt.com, perplexity.ai, or gemini.google.com individually to see traffic from each platform.
• Add Landing page as a secondary dimension to see which pages each AI platform is citing.

This method only shows one source at a time and doesn’t persist between sessions. It’s useful for a quick sense-check, but for anything ongoing, use method 2 or 3.

Method 2: Custom Explore report for trend analysis

The Explore section in GA4 lets you build a saved report that filters for multiple AI sources at once using a regex pattern. A regex (regular expression) is a text-matching formula that lets GA4 identify sessions from multiple domains using a single rule.

• Go to Explore and create a blank exploration.
• Add Session source/medium as a dimension. Also add Landing page and query string.
• Add Sessions, Engaged sessions, Engagement rate, and Key events as metrics.
• Apply a filter on Session source using Matches regex (partial match), then paste the pattern below.
• Set the visualisation to a line chart. Use a date range of at least 90 days to see meaningful patterns.

Use this regex pattern, which covers the major AI platforms as of mid-2026:

chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|
copilot\.microsoft\.com|deepseek\.com|grok\.com|meta\.ai|you\.com

Review and update this pattern quarterly. New platforms launch regularly and established ones sometimes change their referral domains.

Method 3: Custom channel group (recommended for ongoing reporting)

A custom channel group is a user-defined traffic classification in GA4 that groups sessions by source rules you specify. Once set up, it makes AI traffic appear as its own channel in every standard acquisition report, alongside Organic Search, Paid Search, Direct, and Referral.

• Go to Admin, then Data display, then Channel groups.
• Click Copy to create new on the Default channel group. Name it clearly, for example: Default + AI Traffic.
• Click Add new channel. Name it AI Traffic.
• Set the condition: Source matches regex (partial match). Paste the same regex pattern from method 2.
• Drag the AI Traffic channel above Referral in the channel order. This step is consistently skipped in other guides and it matters. GA4 assigns each session to the first matching channel in the list. If AI Traffic sits below Referral, AI visits get categorised as generic referrals instead.
• Save and set as your primary channel group.
• Allow 24 to 48 hours for the data to populate.

One practical benefit: custom channel groups apply retroactively to historical data. Once you’ve saved the group, GA4 will show AI traffic in your acquisition reports going back to when the property was created.

The dark traffic problem: why GA4 only shows part of the picture

GA4 only captures AI referral traffic when the referring platform passes a referrer header. A significant portion of AI-driven visits arrive without one, landing in Direct instead. This isn’t a configuration error you can fix. It’s a structural feature of how mobile apps, copy-paste behaviour, and some AI platforms handle outbound links.

There are three main reasons referrer headers don’t reach GA4.

The first is mobile apps. ChatGPT’s iOS and Android apps open outbound links in environments that strip referrer data before the request reaches your server. Even when a user taps a link directly from the app, the visit arrives with no referrer header attached.

The second is copy-paste behaviour. Many users copy a URL from an AI response and open it separately in a browser tab. When a URL is pasted rather than clicked, no referrer is passed. GA4 sees a new session with no source information and classifies it as Direct.

The third is in-app browsers and sandboxed environments. Some AI platforms open links in embedded browsers that suppress referral data for privacy or technical reasons, producing the same result.

The scale of the gap is difficult to pin down precisely, but research points to it being substantial. Analysis of over 446,000 sessions across sites using dedicated AI traffic detection found that approximately 70% of AI referral traffic arrived without a referrer header and was categorised as Direct rather than Referral (Clickport, April 2026). Both figures come from commercial tools with an interest in framing the gap as large, so treat them as directional rather than precise. The underlying cause is well-documented regardless of the exact percentage.

One partial improvement arrived in June 2025, when ChatGPT began appending utm_source=chatgpt.com to outbound links in its search citations and More sources section. This makes some ChatGPT traffic trackable even without a referrer header, and you may now see utm_source=chatgpt.com appearing in your GA4 sessions alongside the referrer-bearing traffic. The limitation is that this covers only part of ChatGPT’s output. Conversational inline links, free-tier responses, and mobile app traffic still don’t carry UTMs.

There’s one further blind spot worth naming clearly: clicks from Google AI Overviews pass through as standard google / organic traffic in GA4. There’s no referrer signal that distinguishes an AI Overview click from a traditional organic result click. If you’re looking to measure Google AI Overview performance specifically, GA4 alone won’t get you there.

For UK and EU sites, consent rejection compounds the problem further. On sites with compliant cookie banners, users who decline tracking are excluded from GA4 data entirely. The combined effect of referrer stripping and consent rejection means that on some sites, a meaningful proportion of AI visits are invisible to analytics regardless of configuration.

How to diagnose dark AI traffic using landing page patterns

If a portion of your AI traffic is landing as Direct, you can still detect its presence by examining the behaviour of your Direct sessions. AI-driven Direct visits tend to share a recognisable pattern: they land deep in the site, engage at above-average rates, and cluster on pages that feature in AI-cited content.

This isn’t a measurement method. It’s a diagnostic. It produces signals, not precise attribution. But it’s more useful than accepting Direct traffic as a black box.

Here’s how to run it:

1. Go to Reports, then Engagement, then Landing page. Filter the report to show only Direct sessions.
2. Look for deep landing pages receiving Direct traffic: specific guides, service pages, case studies, or product pages. Brand pages and homepages attracting Direct sessions are more likely to be genuine direct visits. Deep, specific pages receiving Direct traffic with high engagement rates are more likely to be AI-influenced.
3. Cross-reference with your custom AI channel. If the same page appears in both your AI Traffic channel and your Direct channel with similar engagement patterns, the Direct sessions on that page are probably the dark traffic version of the same referral source.
4. Compare your Direct traffic growth rate against branded search query volume in Google Search Console. If Direct is growing faster than branded queries, the difference is likely AI-influenced visits that GA4 can’t attribute. Branded search volume is a reasonable proxy for genuine direct intent.

None of this is definitive. But it gives you a basis for estimation, and it prevents you from treating a growing Direct channel as evidence of brand awareness when the real driver may be AI citation.

What to do with AI traffic data once you have it

Tracking AI referral traffic is a starting point, not a destination. The value is in what the data tells you: which pages AI platforms are citing, which content formats are generating referrals, and how AI as a channel is moving relative to organic search.

Start with the pages receiving AI referrals. If a specific guide or service page is consistently generating sessions from Perplexity or ChatGPT, that’s a signal its structure is working for AI citation. Look at what those pages have in common, whether that’s a clear answer-first format, structured data, or a depth of coverage on a specific topic, and apply the same approach to other pages.

Compare key event rates between your AI Traffic channel and your Organic Search channel. In most datasets, AI-referred sessions show higher engagement and conversion rates than organic search sessions, because users who click through from an AI response have already had context built around the recommendation. If you see that pattern in your own data, it strengthens the case for prioritising content that earns AI citation.

For trend analysis over 12 or more months, connect your GA4 property to Looker Studio. Looker Studio is Google’s free data visualisation tool that removes the date restrictions present in GA4 Explore reports and lets you build shareable dashboards for client or stakeholder reporting. It won’t extend your data further back than your GA4 property goes, but it makes long-term AI traffic trend analysis considerably easier to present.

There is a ceiling to what GA4 can tell you about AI visibility. It measures referral sessions where a click took place and the referrer survived attribution. It tells you nothing about how often your brand is cited without a click, where you sit relative to competitors in AI-generated responses, or what AI platforms actually say about your business. For that layer of measurement, you need AI visibility benchmarking that tracks citation frequency and share of voice across the major platforms directly. That’s the part of the picture GA4 won’t show you, however well the tracking is configured.

Conclusion

GA4 can show you a real and trackable portion of your AI referral traffic. The custom channel group setup takes around 20 minutes, applies retroactively, and gives you a dedicated AI channel in every standard acquisition report going forward. That’s worth doing.

What it won’t do is show you everything. A meaningful share of AI-driven visits arrive without referrer data and land as Direct, regardless of how well your GA4 is configured. The landing page diagnostic gives you a way to infer their presence. But measuring the full picture of AI visibility, citation frequency, brand share of voice, competitor positioning, requires a different layer of measurement entirely.

If you want to understand not just the clicks AI sends to your site but how your brand performs across AI platforms more broadly, our GEO solutions and AI visibility benchmarking team can help. Learn about GEO solutions.

Frequently asked questions about tracking AI visibility

Why is my ChatGPT traffic showing as Direct in GA4?

ChatGPT’s mobile app and free-tier responses don’t pass referrer headers, so GA4 has no information about where the visit originated and defaults to Direct. Since June 2025, ChatGPT has appended utm_source=chatgpt.com to links in its search citations and More sources section, which helps with attribution for some visits. Conversational inline links and mobile app traffic still don’t carry UTMs, so a portion of ChatGPT traffic will remain in Direct regardless of your GA4 configuration.

Does GA4 track Google AI Overviews traffic separately from organic?

No. Clicks from Google AI Overviews pass through as standard google / organic traffic in GA4. There’s no referrer signal that distinguishes an AI Overview click from a traditional organic result. If you want to estimate AI Overview impact, compare landing page performance for your target pages before and after AI Overviews began appearing for those queries, but treat this as inference rather than direct measurement.

How often should I update my AI traffic regex in GA4?

Review it at a minimum every quarter. New AI platforms launch regularly, and established ones sometimes change their referral domains. Bing Copilot changed its referrer pattern more than once in 2025. Check your Referral report monthly for unfamiliar domains containing “ai”, “chat”, or “assistant” in the domain name, and add any new sources to your regex promptly.

Is AI referral traffic worth tracking if the volumes are small?

Yes, for two reasons. First, AI-referred sessions typically show above-average engagement and conversion rates, meaning each visit carries more commercial weight than volume alone suggests. Second, the data you collect now forms a baseline for measuring how AI visibility changes over time. Starting later means no historical comparison when the channel becomes more material.

Can I track AI referral traffic without building everything in GA4?

Looker Studio connected to your GA4 property replicates the reporting without GA4’s date restrictions in Explore and makes dashboards easier to share with stakeholders. Server-side log files capture all inbound requests including AI crawler activity, which can show you the scale of AI engagement on your site even where session-level attribution isn’t possible. For the visibility that referral tracking can’t reach, dedicated AI benchmarking tracks citation frequency, brand positioning, and share of voice across the major AI platforms directly.