AI Traffic in Google Analytics 4: How to Separate It from Referral Traffic

Traffic from ChatGPT, Perplexity, Gemini and other AI assistants is becoming an important part of website acquisition. Yet identifying these visitors in Google Analytics 4 (GA4) is not always straightforward. Some sessions appear under the dedicated AI Assistant channel, others remain classified as Referral, and certain visits have no identifiable source at all. Understanding these differences is essential for measuring how AI-generated recommendations contribute to website visits, enquiries and sales. With the right GA4 reports and a carefully configured custom channel group, marketers can distinguish identifiable AI traffic from conventional referrals without losing sight of the limitations of attribution.

How Google Analytics 4 Identifies Traffic from AI Assistants

AI referral traffic occurs when someone follows a link to a website from an AI-generated response. For example, a potential customer might ask ChatGPT to recommend accounting software, receive several suggestions and visit one of the recommended websites. If the browser passes information about the referring source, GA4 can recognise where that session originated. Common AI sources include chatgpt.com, perplexity.ai, claude.ai, gemini.google.com and copilot.microsoft.com. These visits differ from traditional referrals because the recommendation comes from an AI conversation rather than a conventional website article, directory or partner link.

On 13 May 2026, Google introduced a dedicated AI Assistant channel in GA4’s default channel grouping. This was a significant change for acquisition reporting. When Google Analytics recognises an eligible AI referrer, it automatically assigns the medium ai-assistant, uses (ai-assistant) as the campaign name and places the traffic in the AI Assistant channel. Google’s documentation identifies services such as ChatGPT, Gemini, Claude, DeepSeek, Copilot and Grok among the relevant assistants. This means that marketers working with current GA4 data no longer need to assume that every identifiable AI visit is hidden inside Referral.

However, the automatic classification does not capture every AI-related visit. Recognition depends on whether GA4 receives usable source information and whether that source matches Google’s classification rules. Traffic from an assistant that GA4 does not recognise may still appear under Referral, while visits without a detectable referring source can be attributed to Direct. The distinction matters when comparing acquisition channels: a low figure in AI Assistant does not necessarily mean that AI services contribute little traffic. It represents the sessions identified by the available tracking and classification methods, not every visit influenced by AI-generated information.

Where AI Traffic Appears in GA4 Reports

The first place to check is Reports, then Acquisition, followed by Traffic acquisition. Select Session default channel group as the main dimension and look for AI Assistant alongside Organic Search, Referral, Direct and other acquisition channels. This report shows session-based performance, making it suitable for examining visits attributed to AI services. If AI Assistant is missing, check the reporting period and whether the property has recorded qualifying traffic. A missing row does not prove that the website has never received visitors from AI-generated answers.

For a more detailed view, change the main dimension to Session source / medium. This identifies individual traffic sources rather than grouping every visit into a broad acquisition category. Recognised AI traffic may display an assistant’s domain with the medium ai-assistant, while other assistant domains can appear with referral as their medium. Inspecting both is important because filtering only for referral would exclude sessions that Google already classifies correctly. Source-level reporting also helps distinguish genuine assistant referrals from links originating on unrelated websites with similar names.

Another useful comparison is between Traffic acquisition and User acquisition. The former focuses on the sources responsible for sessions, while the latter describes how users were initially acquired. Someone might first visit through ChatGPT, return later through Google Search and subsequently purchase through an email campaign. These visits can appear differently depending on the report and attribution scope. For routine AI traffic monitoring, session-based dimensions provide a clearer account of where individual visits originated. First-user information remains useful when examining whether AI assistants introduce new audiences to the business.

How to Separate AI Referrals from Regular Referral Traffic

The most practical starting point is to review which AI sources are already visible in the property’s existing traffic data. In Traffic acquisition, select Session source / medium and inspect entries associated with known assistant domains. ChatGPT may generate traffic associated with chatgpt.com or chat.openai.com, while Perplexity can appear through perplexity.ai. Other relevant domains include claude.ai, gemini.google.com, copilot.microsoft.com, grok.com and deepseek.com. The exact sources recorded will depend on user journeys, referrer availability and GA4’s recognition rules. Avoid assuming that every listed service must appear in every account.

For a quick analysis, open Explore and create a Free form exploration. Add Session source / medium as a dimension and Sessions, Engaged sessions and Key events as metrics. Apply a filter that matches known AI domains, then review the resulting sources and their performance. This approach helps identify assistant referrals without modifying the property’s reporting configuration. It is especially useful for an initial audit or for checking whether traffic from less familiar AI services is currently mixed with conventional referrals. Keep a version of the exploration without the filter to compare the results against total website acquisition.

For ongoing reporting, a custom channel group is more convenient. Google Analytics allows users with the appropriate Editor permissions to create rule-based channel groups that organise traffic according to source, medium and other supported dimensions. Unlike changes to the tracking code, this configuration does not require website development work. Custom channel groups can also be applied retrospectively to available data, making them useful when analysing historical AI referrals. They do not create missing attribution information, however: a custom rule can regroup recorded sources but cannot identify the original assistant when that information was never collected.

Creating a Custom AI Traffic Channel in GA4

Open Admin, locate Data display and select Channel groups. Choose Create new channel group, enter a descriptive name such as Acquisition Channels with AI, and add a channel called AI Traffic. Set a rule using Source and the matches regex operator. A focused expression for common assistant domains is (chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|grok\.com|deepseek\.com). This expression is intended to identify source values containing those domain names. Review the actual values in your GA4 property and adjust the expression if additional assistant sources appear.

Next, add another condition group to the same AI Traffic channel using Medium exactly matches ai-assistant. Treat the source-based rule and the medium-based rule as alternatives, rather than requiring both conditions to be true. This allows the custom channel to include traffic already recognised by GA4 and additional identifiable AI referrals. Move the new channel above Referral and the existing AI Assistant channel, ensuring it takes precedence when a session satisfies multiple definitions. GA4 evaluates channels according to their order and assigns matching traffic to the first eligible channel. Save the group once the conditions and ordering are correct.

Return to Traffic acquisition and select the newly created custom channel group as the reporting dimension. You should now see AI Traffic as a separate category whenever matching sessions are available. Compare its session count with the default AI Assistant channel and examine which sources account for any difference. The new group is an alternative reporting view, so the original default channel definitions remain available. Check that recognised assistant sessions have not been duplicated across categories and that conventional referral sources remain outside the AI group. Revisit the source rules periodically, as assistant domains and Google’s automatic classification can change.

How to Measure AI Traffic Performance and Avoid Attribution Errors

Separating AI traffic is useful only when the resulting figures support better marketing decisions. Begin with sessions and users to understand traffic volume, then examine engagement rate, average engagement time per session and key events. For an online retailer, key events might include completed purchases or other actions configured as important business outcomes. For a service provider, they could include enquiry submissions or appointment requests. Comparing these figures with Organic Search and conventional Referral traffic reveals whether visitors arriving from AI recommendations behave differently after reaching the website.

Landing pages provide another valuable perspective. In an exploration, combine Landing page + query string with session-based source information to identify which content attracts visits from AI assistants. A detailed product comparison might receive relatively few visits but generate qualified enquiries, while a general educational article may attract a larger audience with limited immediate commercial activity. Review these pages alongside their purpose, engagement and key-event performance rather than judging them by session volume alone. Where sample sizes are small, compare longer periods and avoid drawing strong conclusions from just a handful of visits.

It is equally important to recognise what GA4 cannot measure. A visitor who reads an AI-generated answer without clicking a link creates no website session. A person who copies a URL from an assistant and opens it separately may appear under Direct because referral information is absent. Privacy settings, mobile applications and intermediate redirects can also affect source identification. Consequently, AI Assistant and custom AI Traffic figures should be treated as measurable, attributed visits rather than a complete record of AI influence. Google’s AI Overviews and AI Mode introduce a further distinction: traffic from these Google Search experiences is generally reported within Organic Search, not the standalone AI Assistant channel.

Building Reliable AI Traffic Reports for Marketing Decisions

A practical monthly report should show AI-attributed sessions, engaged sessions, engagement rate, key events and, where relevant, purchase revenue. Include the main assistant sources and the landing pages receiving their visits. Use the same reporting period and comparison metrics for Organic Search and conventional Referral traffic. If the custom AI Traffic group contains both Google’s recognised AI sessions and additional assistant referrals, state that definition clearly in the report. This prevents colleagues from comparing figures built with different classification rules or mistaking a reporting change for genuine traffic growth.

Consider a website recording 12,000 monthly sessions from Organic Search, 1,400 from conventional referrals and 320 from identifiable AI sources. If 160 of those AI sessions are already captured by the default AI Assistant channel, the custom group could reveal another 160 sessions previously distributed across other recorded categories. These numbers are illustrative rather than industry benchmarks. The important point is that the wider definition changes the visibility of acquisition sources, not the total number of sessions collected. If the 320 AI sessions produce 24 key events, that provides a useful starting point for comparing business outcomes with other channels.

Maintain the report by checking assistant source names, reviewing unexpected increases in Direct traffic and confirming that key-event tracking remains consistent. Google Search Console can provide complementary information about search performance, although it does not separately identify every visit influenced by AI-generated search features. Avoid estimating exact AI traffic from unexplained Direct sessions or presenting assistant citations as equivalent to website visits. The most dependable approach is to combine GA4’s native AI Assistant classification with a maintained custom channel group, compare meaningful outcomes and acknowledge the visits that cannot be attributed reliably. This produces a clearer basis for evaluating AI-driven discovery throughout 2026.