Setting Up GA4 to Capture AI Search Referrals: The Complete Tracking Guide

According to Semrush's 2026 Traffic Analytics report, direct traffic has grown as a share of total sessions by an average of 11 percentage points across B2B SaaS sites since January 2025 — and most of that growth is not from people typing URLs directly into browsers. It is AI-assisted discovery converting to unattributed sessions. GA4, in its default configuration, is incapable of distinguishing an AI-referred visit from a bookmarked return visit, making it structurally blind to one of the fastest-growing acquisition channels in B2B marketing.

This is not a niche problem. If your site is receiving any meaningful traffic from ChatGPT, Perplexity, Claude, or Gemini — and by Q1 2026, virtually every B2B site above 5,000 monthly sessions was — you are operating with a measurement gap that affects every downstream decision in your marketing stack. Your attribution model is wrong. Your channel mix reports are wrong. Your content ROI analysis is wrong. The AI dark funnel is real and growing, and GA4's defaults do nothing to illuminate it.

The good news is that this is a solvable configuration problem. The bad news is that solving it requires touching five separate areas of GA4 plus your UTM convention, your Search Console setup, and optionally your BigQuery export. This guide covers every one of them, in implementation order, with the exact settings and SQL queries required. It is written for marketing ops, analytics engineers, and growth teams who want an accurate picture of how AEO investments translate to measurable site behavior.

Why GA4 Misses AI Traffic by Default

Before touching any settings, it helps to understand the four mechanics by which GA4 loses AI-referred traffic to the Direct channel. Each has a different fix, and conflating them leads to configurations that address one problem while leaving the others open.

The referrer-stripping problem. Modern browsers follow the Referrer Policy specification, which by default strips the referrer header when a user navigates from an HTTPS page to a different HTTPS domain via a standard link. ChatGPT (chat.openai.com) and Claude (claude.ai) both operate over HTTPS and both serve links to external sites in ways that trigger the default referrer policy. When a user clicks a link in a ChatGPT answer and lands on your site, the browser may deliver zero referrer information at all — GA4 sees the session arrive with no origin and credits it to Direct. This problem affects a meaningful share of ChatGPT clicks and a smaller but non-zero share of Claude clicks.

The redirect-chain problem. Some AI assistants, including certain ChatGPT interface states and the Bing Copilot experience, route outbound clicks through a redirect intermediary before landing on the destination URL. This redirect breaks the referrer chain — the referrer recorded on your site is the redirect domain, not the AI assistant domain, and if your GA4 isn't configured to recognize that redirect domain as an AI source, the session is again mislabeled.

The referral exclusion problem. GA4's default referral exclusion list is designed to prevent your own domain and major payment processors from creating new sessions. But some GA4 implementations — particularly those migrated from Universal Analytics with legacy configurations — have AI assistant domains on the exclusion list, either added manually during initial setup or inherited from a template. When a domain is on the referral exclusion list, any session that arrives from that domain is treated as a continuation of the user's previous session (if one exists) or as a new Direct session (if not). The result is that Perplexity or Gemini referrals are silently converted to Direct with no log that the exclusion happened.

The session timeout problem. GA4 treats sessions that resume after a timeout (default: 30 minutes of inactivity) as new Direct sessions, regardless of how the original session was acquired. A user who reads a Perplexity answer, goes away for an hour, then returns to your site and converts will appear in GA4 as a Direct conversion. In B2B contexts with longer research cycles, this problem is compounded because users may discover your brand via AI in one session and return via branded search in a later session — the AI role in the conversion is never captured.

Understanding which of these four problems is affecting your specific property matters because the fixes are different. The referrer-stripping problem is addressed by GTM-based referrer capture. The redirect-chain problem is addressed by custom channel definitions and source mapping. The referral exclusion problem is addressed by auditing and correcting the exclusion list. The session timeout problem is addressed by BigQuery analysis and CRM correlation, not by GA4 configuration.

The AI Referrer Domain List You Need

The first practical step is building a comprehensive list of AI assistant referrer domains. This list forms the condition logic for your custom channel group, your referral exclusion audit, and your BigQuery queries. As of May 2026, the confirmed referrer domains for the major AI assistants are:

Note that the mobile apps for ChatGPT, Claude, and Perplexity generally strip referrer data entirely — clicks from in-app link taps arrive with no referrer, which means they cannot be attributed to AI search through referrer-based tracking alone. This is the core driver of the dark funnel problem: the fastest-growing usage context for AI assistants (mobile in-app) is also the most attribution-opaque.

Keep this list in a shared document and update it quarterly — new AI assistants enter the market regularly, and existing assistants change their link-handling behavior with product updates.

Custom Channel Grouping Setup

GA4's channel groupings determine how traffic sources are bucketed in standard reports, including the Traffic Acquisition report, the Landing Page report, and conversion attribution views. The default channel groupings do not include a category for AI Search. Without a custom grouping, AI referral sessions get split across Direct, Referral, and (rarely) Organic Search, with no way to aggregate them.

Step 1: Navigate to the channel group settings. In GA4, go to Admin > Data Settings > Channel Groups. You will see the default system channel group, which cannot be edited. Click "Create new channel group" to begin building your AI-aware version.

Step 2: Create the AI Search channel. Add a new channel named "AI Search." Set the condition type to "Session source" and configure it to match any of the referrer domains in your list. The condition should use "contains" logic to match both www and non-www versions: session source contains "perplexity.ai" OR session source contains "chat.openai.com" OR session source contains "chatgpt.com" OR session source contains "claude.ai" OR session source contains "gemini.google.com" OR session source contains "copilot.microsoft.com" OR session source contains "you.com" OR session source contains "phind.com" OR session source contains "meta.ai".

Step 3: Set channel priority correctly. Channel groups evaluate rules in order from top to bottom, applying the first matching rule. Your AI Search channel must appear above the "Direct" channel in the priority stack — otherwise, sessions with stripped referrers that would match Direct are never evaluated against the AI Search condition. In practice, you should order channels as: Paid Search, Paid Social, AI Search, Email, Organic Search, Organic Social, Referral, Direct, Unassigned.

Step 4: Configure the AI-Assisted Branded Search channel. Add a second new channel named "AI-Assisted Branded Search." This channel captures a key dark funnel behavior: the user discovers you via AI, then later executes a branded search. Set the condition to: session source matches "google.com" OR "bing.com" AND medium matches "organic" AND landing page matches your brand name or domain pattern. This will not catch all dark funnel conversions, but it gives you a directional signal that you can correlate against AI citation events to estimate indirect AI influence. For a deeper view on this methodology, see the AEO citation tracking playbook.

Step 5: Apply and validate. Apply the new channel group to your Traffic Acquisition report and compare the AI Search channel volume against the prior period's Direct traffic. In almost every case, teams who complete this setup see AI Search appear as a channel representing between 1% and 12% of total sessions, with traffic that was previously credited to Direct. The exact percentage depends on how frequently your brand is cited in AI assistant answers, which you can measure using share of model methodology.

Referral Source Configuration: Fixing the Exclusion List

The referral exclusion list is one of the most commonly misconfigured GA4 settings, and it is the silent killer of AI referral attribution. Navigate to Admin > Data Streams > [your stream] > Configure Tag Settings > List unwanted referrals. Review every entry.

For most GA4 properties, the list should contain only: - Your own domain and any subdomains - Payment processor domains (stripe.com, paypal.com, braintreegateway.com, etc.) - Single sign-on provider domains if you are using social login

The list should NOT contain: - Any AI assistant domain - Any search engine domain (google.com, bing.com should never be excluded) - Any social media domain (linkedin.com, twitter.com referrals are valuable signal)

If you find any AI assistant domains on your exclusion list, remove them immediately. Sessions that previously arrived via Perplexity and were silently converted to Direct will now appear as proper Referral sessions — but only for sessions going forward. Historical data that was mislabeled as Direct cannot be recovered in GA4's standard interface; it requires BigQuery analysis of the raw event stream.

The self-referral fix. One related configuration issue: if your site uses a login wall, checkout flow, or embeds that require navigating through a subdomain, you may need to add those subdomains to the cross-domain measurement configuration rather than the exclusion list. Teams who add app.yoursite.com to the exclusion list to prevent login redirects from creating new sessions are inadvertently causing GA4 to drop all referrer information for post-login sessions. The correct fix is to configure cross-domain measurement in the Google Tag configuration, not to add the subdomain to the exclusion list.

Custom Dimensions for AI Visibility Tracking

GA4's built-in traffic source dimensions (source, medium, campaign) give you enough to identify AI-referred sessions in aggregate. But for AEO teams, you need finer-grained dimensions that answer more specific questions: Which AI assistant sends the most engaged sessions? Which content gets cited by AI assistants? How does landing page behavior differ between AI-referred and search-referred sessions?

Four custom dimensions provide these answers.

AI Referrer Source (Event-scoped). Create an event-scoped custom dimension named "AI Referrer Source" mapped to a custom event parameter also called "ai_referrer_source." In Google Tag Manager, configure a trigger that fires on Page View events where the Referrer variable contains any AI assistant domain. When the trigger fires, send a custom event with the parameter set to the full referrer domain value. This gives you a breakout of ChatGPT vs Perplexity vs Claude traffic that is not available in the standard source/medium breakdown.

AI Landing Page Category (Event-scoped). Create a custom dimension named "AI Landing Page Category" that categorizes the landing page URL into content types: blog, documentation, comparison-page, product-page, homepage, case-study. Configure this in GTM using a lookup table variable that maps URL path patterns to categories. When you segment AI-referred sessions by this dimension, you will see which content categories drive the most AI referral visits — typically comparison pages and documentation, not blog posts, which aligns with the broader SaaS AEO citation pattern research.

AI Session Quality Score (User-scoped). Define a user-scoped custom dimension that scores AI-referred users by engagement depth: 1 for single-page sessions, 2 for multi-page sessions without conversion events, 3 for sessions with conversion micro-events (demo requested, pricing viewed, contact form initiated), 4 for sessions with macro-conversions. Populate this via a custom event that fires after each meaningful engagement action. The resulting distribution tells you how AI-referred sessions compare in quality to organic search sessions and paid sessions — this data point is often the most persuasive piece of evidence for increasing AEO investment in leadership discussions.

AI Citation Content Tag (Event-scoped). For content that you have actively optimized for AI citation (structured schema, FAQ markup, comparison tables), add a custom tag to the HTML of those pages that GTM can read and push as an event parameter. Name the dimension "AI Citation Optimized" with a boolean value. This lets you directly compare conversion behavior between sessions that landed on AEO-optimized pages versus unoptimized pages — the most direct measurement of whether schema and content structure work is delivering engagement value beyond just the citation.

Perplexity- and ChatGPT-Specific Referrer Quirks

Two assistants deserve special attention because their link-handling behavior is unusual enough to require specific configuration.

Perplexity.ai passes the Referrer header reliably when users click answer links, which makes it the easiest AI assistant to track. However, Perplexity also has a "Pro Search" mode where it fetches content on the user's behalf rather than linking to it, meaning a Perplexity citation may drive zero click traffic even when Perplexity is actively citing your content. The implication is that Perplexity traffic in GA4 systematically undercounts Perplexity citation frequency — the GA4 signal is directionally useful but should be supplemented with direct AEO citation measurement.

ChatGPT has the most complex attribution picture. The web interface (chat.openai.com) passes a referrer for clicked links, but the iOS and Android apps strip referrers entirely. The ChatGPT-4o model with Browse enabled fetches URLs directly during response synthesis, which may log your content in server logs but never generates a GA4 session. The ChatGPT Operator API, which third-party apps use to embed ChatGPT functionality, does not pass any referrer metadata. The practical implication is that ChatGPT traffic in GA4 represents only the subset of ChatGPT interactions that involved a web browser user clicking a direct link — probably 15–30% of all ChatGPT referral events depending on your audience demographics.

For both assistants, the right mental model is that GA4 referral data is a floor, not a ceiling. The true AI-influenced traffic is always higher than what GA4 reports. Building the dark funnel correlation analysis in BigQuery is the only way to estimate the actual ceiling.

UTM Conventions for AEO-Adjacent Traffic

Most AEO traffic arrives organically without UTM parameters — an AI assistant cites your page and a user clicks the link, with no opportunity for you to pre-tag the session. But several traffic vectors that are closely adjacent to AEO can and should be UTM-tagged to give GA4 the signal it needs.

Press releases and syndication. Any content you publish to PR Newswire, Business Wire, or GlobeNewswire will be indexed by AI crawlers and may drive AI-cited traffic. Tag all press release links back to your site with utm_source=pr-newswire (or the relevant wire service), utm_medium=press-release, and utm_campaign matching the announcement. This lets you track whether press release syndication is driving AI-cited traffic at a measurable rate.

Newsletter and email links. When you reference your own content in newsletters, tag the links with utm_source matching the newsletter platform, utm_medium=email. This prevents newsletter clicks from appearing as Direct traffic and polluting your AI Search channel signal.

Comparison and alternatives pages. For any comparison-page content that you actively distribute (shared in Slack communities, posted on LinkedIn, submitted to curated lists), tag the outbound links with utm_source=comparison-page and utm_medium=referral. When AI assistants cite these pages and users click through, the session will still arrive with the Perplexity or ChatGPT referrer, but the UTM data from the distribution campaign gives you a way to correlate comparison-page investment with subsequent AI citation rates.

AI tool integrations. If your product integrates with any AI assistant via a plugin, action, or tool definition, any traffic those integrations generate back to your marketing site or documentation should be tagged with utm_source=ai-plugin, utm_medium=integration, utm_campaign matching the specific tool. This is a small but growing traffic source in B2B, and without tagging it gets absorbed into Direct.

The full UTM convention for an AEO-oriented B2B marketing stack in 2026:

Search Console Correlation: The Signal You're Missing

Google Search Console (GSC) is an underutilized tool for AEO measurement because most teams think of it as a traditional SEO tool. But in 2026, GSC provides two signals that are directly relevant to AI search measurement.

Branded query volume as an AI dark funnel proxy. When users discover your brand via an AI assistant and then execute a branded Google search, that search appears in GSC as an impression and click on your brand name. If you track weekly branded query volume in GSC and weekly AI citation volume from a tool like Profound or Otterly, you will often see these two metrics correlate with a 2–4 week lag. Rising AI citation share in week 1 predicts rising branded search volume in weeks 3–5. This correlation is the closest available proxy for measuring AI dark funnel pipeline within tools that most teams already own.

Index coverage as an AI crawl proxy. Google and AI assistants largely crawl the same public pages, and GSC's Coverage report shows which pages Google has indexed. Pages that have index coverage errors (soft 404s, redirect chains, server errors) are likely to have the same issues for AI crawlers. Running a monthly GSC coverage audit is a cost-effective way to identify and fix pages that may be invisible to both Google and AI assistants.

To connect GSC data to GA4, use the native GSC Link in GA4's Admin > Product Links. Once linked, you can see organic search query data alongside GA4 session data in the Search Console reports section. The data is not available for custom channel analysis, but it gives you a baseline for identifying branded search lift that correlates with AI citation events.

The Playbook: Step-by-Step GA4 AEO Configuration

The complete configuration in recommended implementation order:

1. Audit the referral exclusion list. Before making any other changes, navigate to Admin > Data Streams > Configure Tag Settings > List unwanted referrals. Remove any AI assistant domains. Document the current state before you change it.

2. Create the AI Search custom channel group. Follow the steps above to create a custom channel group with the AI Search channel definition. Apply it to your Traffic Acquisition report and document the baseline AI Search session volume for the trailing 30 days.

3. Implement GTM referrer capture. In Google Tag Manager, create a new variable using the built-in HTTP Referrer variable type. Create a custom event trigger that fires on Page View when the referrer value matches your AI domain list regex. Send a custom GA4 event named "ai_referral_detected" with the referrer value as a parameter. This creates a raw event record for every AI-referred page view, even for sessions where the channel attribution ends up as Direct due to the referrer-stripping problem.

4. Configure custom dimensions. In GA4 Admin > Custom Definitions, create the four custom dimensions described above: AI Referrer Source, AI Landing Page Category, AI Session Quality Score, and AI Citation Optimized. Populate them via GTM events per the configurations above.

5. Link Search Console. If not already linked, connect your GSC property to GA4 via Admin > Product Links > Search Console Links. This enables branded search correlation analysis.

6. Enable BigQuery export. Navigate to Admin > BigQuery Linking and connect to a Google Cloud project. Enable the daily export for all events. Once active, the raw event stream — including page_referrer parameters — will be available for SQL analysis within 24 hours of each day's data.

7. Build the AI referral BigQuery query. Use the query template below to extract AI-referred sessions from your raw event data, including sessions where the referrer was present but GA4 attributed the session to Direct.

8. Set up Looker Studio dashboard. Connect Looker Studio to both your GA4 property and BigQuery export. Build a dashboard with four report pages: AI Search channel performance, AI referrer source breakdown, landing page category by AI source, and the branded search correlation chart. Share this dashboard with the CMO and growth team on a weekly cadence.

BigQuery Export for Advanced Analysis

The BigQuery export is where the full picture of AI-influenced traffic becomes visible. The standard GA4 interface is limited by session-attribution logic that cannot be overridden — once a session is labeled Direct, it stays Direct in every report. BigQuery gives you access to the raw page_referrer event parameter, which preserves the true referrer regardless of how GA4 labeled the session.

The core query for AI referral extraction:

```sql SELECT event_date, (SELECT value.string_value FROM UNNEST(event_params) WHERE key = 'page_referrer') AS page_referrer, (SELECT value.string_value FROM UNNEST(event_params) WHERE key = 'page_location') AS page_location, COUNT() AS sessions FROM `your_project.analytics_XXXXXXXXX.events_` WHERE _TABLE_SUFFIX BETWEEN '20260101' AND '20260525' AND event_name = 'session_start' AND ( (SELECT value.string_value FROM UNNEST(event_params) WHERE key = 'page_referrer') LIKE '%perplexity.ai%' OR (SELECT value.string_value FROM UNNEST(event_params) WHERE key = 'page_referrer') LIKE '%chat.openai.com%' OR (SELECT value.string_value FROM UNNEST(event_params) WHERE key = 'page_referrer') LIKE '%chatgpt.com%' OR (SELECT value.string_value FROM UNNEST(event_params) WHERE key = 'page_referrer') LIKE '%claude.ai%' OR (SELECT value.string_value FROM UNNEST(event_params) WHERE key = 'page_referrer') LIKE '%gemini.google.com%' OR (SELECT value.string_value FROM UNNEST(event_params) WHERE key = 'page_referrer') LIKE '%copilot.microsoft.com%' ) GROUP BY 1, 2, 3 ORDER BY 1 DESC, 4 DESC ```

This query returns every session that arrived with an AI assistant referrer in the raw event data, including sessions that GA4 attributed to Direct because the referrer was later stripped or overwritten. Comparing the session count from this query against the session count in your GA4 AI Search custom channel gives you the gap — the number of AI-referred sessions that GA4 is missing. In typical B2B SaaS environments, this gap is between 30% and 60% of true AI-referred traffic.

A secondary query calculates the conversion rate of AI-referred sessions by joining the session table to the conversion event table:

```sql WITH ai_sessions AS ( SELECT user_pseudo_id, event_date, (SELECT value.int_value FROM UNNEST(event_params) WHERE key = 'ga_session_id') AS session_id FROM `your_project.analytics_XXXXXXXXX.events_` WHERE _TABLE_SUFFIX BETWEEN '20260101' AND '20260525' AND event_name = 'session_start' AND (SELECT value.string_value FROM UNNEST(event_params) WHERE key = 'page_referrer') LIKE ANY ('%perplexity.ai%', '%chatgpt.com%', '%claude.ai%', '%gemini.google.com%') ), conversions AS ( SELECT user_pseudo_id, (SELECT value.int_value FROM UNNEST(event_params) WHERE key = 'ga_session_id') AS session_id FROM `your_project.analytics_XXXXXXXXX.events_` WHERE _TABLE_SUFFIX BETWEEN '20260101' AND '20260525' AND event_name IN ('demo_request', 'contact_form_submit', 'free_trial_signup') ) SELECT COUNT(DISTINCT a.session_id) AS ai_sessions, COUNT(DISTINCT c.session_id) AS ai_conversions, ROUND(COUNT(DISTINCT c.session_id) / COUNT(DISTINCT a.session_id) * 100, 2) AS conversion_rate_pct FROM ai_sessions a LEFT JOIN conversions c ON a.user_pseudo_id = c.user_pseudo_id AND a.session_id = c.session_id ```

In teams that have completed this analysis, AI-referred sessions from Perplexity and ChatGPT show conversion rates 1.4x to 2.8x higher than organic search sessions for bottom-of-funnel content. This data point — AI-referred sessions converting at a premium — is the single most persuasive piece of evidence for AEO investment in a CFO conversation. It transforms AEO from a brand-awareness play into a demonstrable conversion driver.

Team Reporting Templates

The measurement infrastructure above produces data that needs to be translated into reports for different audiences. Three reporting templates cover the most common stakeholder needs.

Weekly AI traffic report (Growth team). A one-page Looker Studio view showing: total AI Search channel sessions this week vs. last week vs. 4 weeks ago; AI referrer source breakdown (Perplexity vs ChatGPT vs Claude vs Other); top 10 landing pages by AI referral session volume; and AI session engagement rate vs. overall site engagement rate. This report should be auto-distributed to the growth team every Monday morning.

Monthly AEO impact report (Marketing leadership). A three-page deck showing: AI Search channel sessions and conversion rate trend (12-week rolling); BigQuery-adjusted AI referral session estimate including sessions misattributed to Direct; branded search volume trend from GSC and correlation to AI citation events from citation tracking tool; and content category breakdown of AI-referred sessions. This report feeds the CMO dashboard that belongs in board decks.

Quarterly AEO attribution analysis (CFO / finance). A single table and one chart showing: estimated AI-influenced pipeline value, calculated as AI-referred sessions × conversion rate × average deal size × attribution weight (typically 0.5–0.8 for last-AI-touch, adjust based on your sales cycle length); comparison to investment in AEO-related activities (tooling, content production, schema work); and payback period estimate. The methodology for this calculation is covered in detail in the AEO ROI framework for CFOs. The GA4 and BigQuery configuration described in this guide is the data foundation that makes this calculation credible rather than speculative.

What to Do When the Data Looks Wrong

Even after full configuration, you will encounter situations where the data appears inconsistent. Three common failure modes and their diagnoses:

AI Search sessions spike then disappear. This usually means one of your AI assistant domains was added back to the referral exclusion list by someone on the analytics team who did not understand why it was removed. Check the Admin > Audit Log to see recent changes.

BigQuery session counts are much higher than GA4. The most common cause is that your GTM trigger for the ai_referral_detected event is misconfigured and firing on page views rather than only on session starts. Check your trigger configuration and ensure it uses the "Session Start" trigger type rather than "Page View."

AI Search conversion rate drops suddenly. Usually indicates a landing page issue on a high-traffic AI-cited page. Run the BigQuery landing page breakdown query, identify the page with the most AI-referred sessions, and audit it for technical issues — slow load time, JavaScript rendering errors, or a schema validation failure that is sending mixed signals to AI crawlers.

Branded search volume does not correlate with AI citations. If you have citation tracking data showing AI citation improvements but no corresponding branded search lift, the most likely explanation is that the citations are generating awareness in a demographic that does not use Google as their search interface — typically in the developer or technical buyer segment, where users may go directly from an AI answer to a DM or LinkedIn search rather than a Google query. Supplement the GSC analysis with direct-traffic trend analysis as a second proxy.

Takeaway: GA4's default configuration treats AI search as invisible, and every B2B marketing team operating without the custom channel groupings, corrected referral exclusion list, GTM-based referrer capture, and BigQuery analysis described in this guide is making content, budget, and headcount decisions on fundamentally incomplete data. The configuration is not complex — most teams can implement the core changes in a week — but it requires deliberate attention across five areas of your analytics stack simultaneously. Teams that complete this setup consistently find that AI Search is already their third or fourth largest acquisition channel, that AI-referred sessions convert at a premium to organic search, and that their AEO investments have been producing measurable returns that were simply invisible in the default GA4 interface. Getting the measurement right is the prerequisite for every optimization decision that follows.