How to Track AI Traffic in GA4: Segmenting ChatGPT, Perplexity, and Gemini Referrals

The limitation is that GA4 measures visits to your site, not visibility inside an AI platform. It cannot tell you how often your brand was mentioned, whether your page was cited in an answer, which prompt triggered the recommendation, or how many people saw an AI result but never clicked. A visible referral is a strong signal because the session arrived with usable source information; an AI-influenced visit that arrives without that information is a weak or hidden signal.

That distinction matters when interpreting AI traffic analytics. Some sessions influenced by AI assistants may appear as direct traffic when no usable referrer is passed, as organic traffic when the user moves through a search engine first, or as unassigned when GA4 cannot place the session into a standard channel cleanly. Treat GA4 as a way to segment observable AI referrals, not as a complete measurement of AI search demand or answer-engine visibility.

Where ChatGPT, Perplexity, and Gemini traffic can appear in GA4

Start your search in the GA4 traffic acquisition report, because AI referrals are usually easiest to spot at the session level. If you need analytics platform background first, start there before working with acquisition dimensions. Go to Reports → Acquisition → Traffic acquisition and change the primary dimension to Session source/medium. This shows where the session began and how GA4 classified the visit, which is the most useful view when someone clicks from an AI assistant into your site.

• Session source/medium identifies the origin and medium for the current visit, such as a referring domain paired with referral. Use this when you want to isolate visits that arrived from ChatGPT, Perplexity, Gemini, or another assistant during the selected date range.
• First user source/medium appears in the User acquisition report and describes how the user was first acquired. This is useful for understanding whether AI was the first known entry point, but it is weaker for measuring repeat visitors whose original acquisition happened through another channel.
• Session default channel group groups traffic into GA4’s channel buckets, such as Referral, Organic Search, Direct, or Unassigned. Use it for a high-level channel view, but do not rely on it alone because AI-related visits can be split across more than one bucket.
• Landing page shows the first page viewed in the session. Pair it with session-level acquisition dimensions to see which articles, product pages, comparison pages, or home pages AI assistants are actually sending people to.

When a usable referrer is passed, these visits often look like GA4 referral traffic. When the referrer is stripped by an app, browser, redirect, privacy setting, or intermediate search experience, the same kind of AI-influenced visit may be recorded as Direct, Organic Search, or Unassigned. The practical takeaway: build your AI segment from observable session-level fields, then use landing pages and engagement metrics to interpret quality rather than assuming one clean “AI” channel exists.

Common source patterns for ChatGPT, Perplexity, Gemini, and other AI assistants

Use the source list as a seed set, not a permanent taxonomy. In GA4, look for these values in Session source, Session source/medium, and, when available in an Exploration, Page referrer. The source tells you the domain GA4 associated with the session; the referrer can sometimes expose the fuller referring URL. That difference matters because a clean source like chatgpt.com / referral is easier to segment than a broader ecosystem source such as google.com or bing.com.

• ChatGPT: Start with chatgpt.com. You may also see OpenAI-related domains in some properties, so keep an eye out for source values containing openai. A practical starting regex to track ChatGPT traffic in GA4 is (^|.)(chatgpt.com|openai.com)$, then adjust it to match the exact source values in your property.
• Perplexity: Look for perplexity.ai as the cleanest signal. If your data shows subdomains, use a pattern that allows them, such as (^|.)perplexity.ai$. This is usually more separable than Google- or Bing-linked AI traffic because the referring domain is distinct, which helps with Perplexity traffic patterns.
• Gemini: Look for gemini.google.com or other source/referrer values containing gemini. To track Gemini traffic in GA4, use a cautious pattern such as gemini.google.com|gemini and review the matching rows before adding broader Google domains. Do not assume all google.com referral or organic traffic is Gemini-influenced.
• Microsoft Copilot and Bing AI experiences: Check for copilot.microsoft.com, source values containing copilot, and AI-influenced sessions that may appear under bing.com. The tradeoff is precision: copilot.microsoft.com is a stronger signal, while bing.com can blend ordinary search, Copilot interactions, and other Microsoft surfaces.
• Claude, You.com, and other answer engines: Add patterns such as claude.ai, you.com, and any emerging assistant domains you see in your own GA4 rows. Keep these in a separate “other AI assistants” group if volume is low, so your ChatGPT, Perplexity, and Gemini comparisons stay readable.

A starter combined regex might look like chatgpt.com|openai.com|perplexity.ai|gemini.google.com|copilot.microsoft.com|claude.ai|you.com. Use it as an editable filter, not a finished measurement rule. Clean AI referrals have recognizable assistant domains; weak or ambiguous signals sit inside larger platforms, appear as Direct, or arrive through Google and Microsoft sources that cannot be confidently separated without additional context.

Create a reusable GA4 comparison for AI referral traffic

Turn the seed list into a report comparison so you can view likely AI sessions beside your default traffic instead of rebuilding filters row by row. In GA4, a comparison is a report-level subset: it does not change the underlying data, but it lets you isolate one group of sessions while keeping the full report context visible.

1. Open Reports in GA4.
2. Go to Acquisition → Traffic acquisition.
3. Click Add comparison.
4. Choose an Include condition.
5. Select Session source or Session source/medium as the dimension.
6. Choose a regex-style match operator, then enter a pattern such as chatgpt|openai|perplexity|gemini|copilot|claude|you.com|phind|komo.
7. Apply the comparison and scan the report for sessions, engagement rate, average engagement time, and key events from the matched sources.

Use Session source when you want the broadest clean match against the referring domain or source label, such as rows containing chatgpt, perplexity, or claude. Use Session source/medium when you want to keep the channel context attached, such as separating chatgpt.com / referral from any unusual medium values that may appear in your property. The practical tradeoff is simplicity versus specificity: source-only filters are easier to maintain, while source/medium filters are better when you need to exclude non-referral rows or audit how GA4 classified the sessions.

Page referrer is useful when it is available in an Exploration or report configuration because it can capture a fuller referring URL rather than only the source label. Use it when the source value is too broad or when you are investigating whether a specific assistant URL pattern is present. Do not rely on it as your only reusable filter, though, because referrer values can be absent, shortened, redirected, or mixed with ordinary web referrals.

Before treating the comparison as your working AI traffic segment, review the matching rows. A pattern like gemini may be useful for Gemini referrals, but broad words can also match unrelated campaigns, page paths, or source names depending on your data. A stronger GA4 custom segment starts narrow, proves that the matched rows are genuinely AI-related, and then expands as new sources appear.

Use GA4 Explorations to inspect landing pages, engagement, and key events

For deeper inspection, move the segment into GA4 explorations, where you can control the table layout instead of relying on the default acquisition report columns.

1. Open Explore → Free form to create a free-form exploration.
2. In Dimensions, import Session source/medium, Landing page + query string, Page path, Device category, Country, and Session default channel group.
3. In Metrics, import Sessions, Engaged sessions, Engagement rate, Average engagement time, Key events or Conversions, Total revenue where relevant, and Event count.
4. Set Rows to Session source/medium, then add Landing page + query string beneath it so each AI source is broken down by entry page.
5. Add a filter where Session source or Session source/medium matches a regex such as chatgpt|openai|perplexity|gemini, then refine the pattern based on the actual rows in your property.

Session source/medium should be the first row dimension because it keeps ChatGPT, Perplexity, and Gemini visible as separate traffic sources. Landing page + query string shows the first page of the session, including URL parameters, which helps identify whether AI visitors are arriving on articles, product pages, comparison pages, or support content. Page path is cleaner when query parameters create too many duplicate rows.

Use the metrics to judge quality, not just volume. Sessions tells you how often the source sent visits; Engaged sessions and Engagement rate show whether those visits met GA4’s engagement conditions; Average engagement time helps separate quick exits from deeper reading; Key events, Conversions, and Total revenue connect the traffic to business outcomes; Event count can reveal useful actions that are not marked as key events.

This side-by-side structure is more useful than one blended AI bucket because each assistant can send different intent. If one source produces a few sessions and no key events, treat the result as a signal to monitor rather than a performance conclusion; low-volume rows can swing sharply from one visit or one conversion.

How to interpret AI traffic once it is segmented

Treat each matched row as a lead-quality clue, not a verdict on the assistant that sent it. Start by sorting the Exploration by landing page, then ask whether the entry pages match the kind of intent you would want from an AI-assisted recommendation: pricing pages, comparison articles, product detail pages, category pages, demo pages, case studies, and practical how-to content are usually stronger signals than thin archive pages or unrelated blog posts.

Next, compare behavior and conversion outcomes against organic search, general referral, and direct traffic over the same date range. A strong AI segment does not need to be the largest segment; it should show signs that visitors are reading, clicking, researching, or converting at a rate that justifies attention. Good signals include above-average engagement rate, meaningful average engagement time, scroll depth events on long-form content, video or file-download events, internal search, product views, add-to-cart actions, form starts, and completed lead or purchase key events. Weak signals include short sessions, no follow-up page views, no content events, and a mismatch between the page visited and the likely question that brought the user there.

For content sites, use AI traffic analytics to identify which articles are being surfaced as answers or references. If an informational article receives AI referrals and those visitors continue to related service pages, newsletter forms, calculators, webinars, or comparison content, that is stronger than traffic that reads one page and leaves. For online stores, focus on whether AI-referred sessions reach product pages, category pages, cart steps, account creation, or purchases, and whether total revenue or average order value is directionally competitive with other referral sources.

Be conservative with small samples. One purchase can make a low-volume AI source look exceptional, and one poor day can make a useful source look weak. Look for repeated patterns across several weeks: the same assistant sending visits to the same cluster of pages, engagement remaining stable, and key events appearing more than once. That pattern is more actionable than a single spike.

Use UTMs when you control AI-generated or AI-tool links

For AI experiences you own or influence, make the link itself carry the attribution. Organic citations inside public assistants are usually outside your control, so you cannot append tags to every ChatGPT, Perplexity, or Gemini recommendation. But you can tag links inside a custom GPT, an internal AI assistant, a partner’s AI tool, a sales enablement bot, a documentation bot, or a campaign that sends users through an AI-guided experience before they reach your site.

Use consistent UTM parameters so controlled AI sessions do not collapse into referral, direct, or an ambiguous source. A practical convention is: ?utm_source=custom_gpt&utm_medium=ai_assistant&utm_campaign=product_recommendation. utm_source names the origin, such as custom_gpt, partner_ai_tool, or docs_bot. utm_medium groups the traffic type, such as ai_assistant or sales_bot. utm_campaign identifies the initiative, such as q3_demo_push, onboarding_help, or pricing_education.

This is especially useful when you track AI traffic in GA4 across both public referrals and owned AI experiences. Keep public assistant traffic in your regex-based AI referral segment, but report tagged AI-tool traffic through its UTM source and medium. Avoid using a referral exclusion as a cleanup shortcut unless you are intentionally removing a self-referral or payment-style handoff; excluding an AI tool domain can erase useful source information that would have helped you separate controlled AI traffic from organic assistant referrals.

Maintain your AI traffic segment and report trends responsibly

Set a monthly checkpoint for the regex itself: look for new assistant domains that are producing sessions, then decide whether each one deserves a narrow match. In Traffic acquisition and your saved Exploration, review the rows already caught by the segment plus nearby referral rows for new assistant-related values, and add only the patterns that clearly belong. A good update is specific, such as a recognizable assistant domain; a weak update is a broad word that could also match campaigns, page paths, or unrelated referrals.

• Keep a simple change log for regex edits, UTM naming changes, new key events, site migrations, and major content launches so trend shifts are not mistaken for demand shifts.
• Review landing pages and key events alongside sessions. Rising GA4 AI referrals matter more when the same pages repeatedly attract engaged visits, form starts, purchases, demo requests, or other qualified actions.
• If recurring reporting outgrows GA4’s interface, use Looker Studio for shareable dashboards or BigQuery for deeper historical tables and joins. Looker Studio is easier for stakeholders; BigQuery gives analysts more control.

The durable takeaway is conservative: AI referral traffic in Google Analytics can show the observable clicks that arrive with usable attribution, not the full universe of AI-influenced discovery. Treat the segment as a directional floor, then use trends, landing-page patterns, and conversion quality to decide what deserves action.

Turn AI referral data into a repeatable reporting habit

Make the report useful by assigning ownership, cadence, and a standard readout. The repeatable workflow is simple: use Traffic acquisition to spot likely assistant sources, save a comparison for quick channel review, use an Exploration when you need page-level detail, then judge the segment by landing pages, engagement, key events, and revenue rather than sessions alone.

• Use the comparison for recurring dashboards because it keeps the AI-referral subset visible inside familiar GA4 reports.
• Use the Exploration for diagnosis because it lets you combine Session source/medium, Landing page + query string, Page referrer, device, engagement metrics, and conversion metrics in one table.
• Use the regex list as a living filter. Keep recognizable patterns such as ChatGPT, Perplexity, Gemini, Copilot, Claude, or other assistant domains narrow enough to avoid accidental matches.

The strongest signal is not “AI traffic went up.” It is a repeated pattern: the same sources send visitors to relevant pages, those visitors stay engaged, and the sessions contribute to key events, purchases, leads, or other outcomes. For online store SEO, that may point to better category content or product page optimization; for publishers, it may show which explanatory pages AI tools tend to surface.

The practical way to track AI traffic in GA4 is to treat it as trend evidence, not a complete attribution model. Some influence will remain hidden in direct, organic search, in-app browsers, privacy-limited sessions, or AI search experiences that do not pass clean referrers. Even so, a disciplined segment helps you make better content, SEO, and reporting decisions, including how to improve visibility in AI search engines.