SEO
Join 500+ brands growing with Passionfruit!
The complete 2026 reference for analytics-owning marketers. Every AI referrer domain we’ve seen in the wild, the regex patterns that catch them, the channel group setup that makes AI traffic a first-class channel, plus Looker Studio templates and BigQuery queries for advanced teams. Solves the “direct traffic” dark channel problem.
Most marketing teams still cannot answer a basic question: how much traffic is AI actually sending us, and which pages does it cite? Every other major channel has a standard report in GA4. AI traffic does not, and Google has not committed a native solution timeline. The result is that ChatGPT, Perplexity, Gemini, Claude, and every other AI platform get lumped into “Referral” or misclassified as “Direct,” and the channel that matters most for the next 24 months is the one your team cannot measure.
The guide below fixes that. We expand on our earlier AI LLM chatbot traffic in GA4 walkthrough with the 2026 referrer landscape, deeper regex for 20+ AI platforms, solutions for the “dark traffic” problem where AI visits show up as Direct, a Looker Studio template, and BigQuery queries for teams with GA4 exports.
Why this matters now. According to Adobe Analytics data covering more than one trillion US retail site visits, AI-referred traffic grew 693.4% year over year during the 2025 holiday season, with AI conversions running 31% higher than non-AI traffic and revenue per visit up 254%. Pew Research’s 2025 study of 900 US users across 68,879 Google searches documented click-through rates dropping from 15% to 8% when AI summaries appear. And joint Forrester and 6sense 2025 research found 94% of B2B buyers now use LLMs during their purchase journey. The traffic is there. The attribution gap is what’s keeping most teams from seeing it. (For the full AEO system this attribution work plugs into, see our Claude for AEO playbook, and for the SEO-to-AEO bridge context, our SEO vs GEO vs AEO guide.)
Why GA4’s Default Reporting Fails for AI Traffic
Three structural issues prevent GA4 from showing you AI traffic accurately out of the box.
1. AI referrers are buried in the Referral channel
GA4 ships with a default channel grouping that lumps every non-search, non-social referral site into “Referral.” ChatGPT traffic, Perplexity traffic, and Claude traffic all appear next to traffic from Product Hunt, Reddit, and random blog posts. Without a custom channel group that elevates AI traffic to its own channel, you are asking your team to eyeball referrer data in a long table.
2. Many AI platforms strip the referrer header
Research published by analytics practitioners and confirmed in the wild: ChatGPT, Perplexity, and other platforms sometimes open outbound links with rel="noreferrer" or referrer-policy settings that strip the Referrer header entirely. GA4 sees the session arrive with no referrer and attributes it to Direct. The more valuable the AI traffic becomes, the more of it disappears into the Direct bucket without aggressive attribution workarounds.
3. Mobile app traffic loses the referrer entirely
ChatGPT’s mobile app, Perplexity’s mobile app, and Claude’s mobile app all route outbound clicks through the user’s default browser. The resulting session often arrives with no referrer data at all, classified as Direct. The OpenAI + Harvard NBER research paper from September 2025 documented 700-800 million weekly active ChatGPT users with 2 billion+ prompts daily, and a significant share of that usage is on mobile where the referrer attribution breaks.
4. Google AI Overviews traffic is invisible without Search Console
AI Overviews sessions arrive from google.com with standard organic attribution. In GA4 alone, you cannot separate “clicked from an AI Overview” from “clicked from a blue link.” The only way to isolate AI Overview traffic is cross-referencing Search Console query data with landing page patterns, which we cover below. For context on the traffic impact, see our analysis of whether AI search referrals are the new clicks.
The Complete 2026 AI Referrer Domain List
Below is the full list of AI platform domains we have seen sending real referral traffic in 2026. Bookmark this list. Update your regex whenever a new platform launches.
Primary platforms (highest volume)
chatgpt.com: OpenAI ChatGPT (primary domain since mid-2024)chat.openai.com: Legacy OpenAI domain, still seenopenai.com: Occasional referrals from OpenAI blog, documentationperplexity.aiandwww.perplexity.ai: Perplexitygemini.google.com: Google Geminibard.google.com: Legacy Bard, deprecated but still rare trafficclaude.ai: Anthropic Claudecopilot.microsoft.com: Microsoft Copilot (web)bing.com/chat: Legacy Copilot entry point
Secondary platforms (growing)
meta.aiandwww.meta.ai: Meta AIchat.mistral.ai: Mistral Le Chatgrok.comandx.com/i/grok: xAI Grok (X/Twitter)deepseek.comandchat.deepseek.com: DeepSeekyou.com: You.compoe.com: Poe (Quora)phind.com: Phind (dev-focused)tongyi.aliyun.com: Alibaba Tongyi Qianwenqwen.aliyun.com: Alibaba Qwenkagi.com/assistant: Kagi Assistant
Specialized / emerging
duckduckgo.com/aichat: DuckDuckGo AI Chathey.app: HeyPihuggingface.co/chat: HuggingChatpi.ai: Inflection Piyandex.com/alice: Alice
ChatGPT-specific subdomains to watch
As of late 2025, OpenAI launched Atlas (a browsing agent) and other ChatGPT sub-products. Watch for new subdomain patterns like atlas.openai.com, chatgpt.com/atlas, chatgpt.com/browse, and chatgpt.com/search. Add these to your regex when you first see them in your data.
Method 1: Custom Channel Group (The 2026 Regex)
A custom channel group in GA4 is the single highest-leverage setup change for AI attribution. The channel group elevates AI traffic to its own surface that appears alongside Organic Search, Direct, and Referral in every standard report.
Setup steps
In GA4, navigate to Admin, then under the Property column, select Channel Groups.
Click Create New Channel Group. Name it
AI Traffic Channels.Click Add new channel. Name it
AI Search(orAI Assistants, whichever label your team prefers).Set the condition:
Sourcematches regex (useSourcerather thanSource/Mediumfor reliability).Paste the regex below.
Save and reorder the channel above
Referralin the channel list. Ordering is critical here. GA4 assigns traffic to the first matching channel, so AI Search must come before Referral.
The comprehensive 2026 regex
If you want separate channels per major platform
Create one channel per platform for more granular reporting. Example for ChatGPT specifically:
For Perplexity:
For the Google AI family:
Separate channels make it easier to see when ChatGPT spikes versus Perplexity, which matters because they have different citation behavior and your optimization work will affect them differently. For the platform-specific citation analysis, see our brand visibility audit methodology.
Ordering matters
After creating AI Traffic channels, drag them above Referral in the channel list. GA4 evaluates channels top to bottom and assigns each session to the first matching channel. If Referral sits above your AI channels, AI traffic will continue to be lumped into Referral. (If you use Claude for analytics work, the MCP connector guide for marketing teams covers how to wire GA4 into Claude so this kind of attribution check runs against live data.)
Method 2: Explorations for Ad-Hoc Analysis
Channel groups are production-ready. Explorations are for the analysis work the channel groups do not cover. Four Exploration configurations cover 90% of AI analysis needs.
Exploration 1: Landing pages AI platforms cite
Shows you which specific URLs on your site AI platforms are sending traffic to. Critical for understanding what content is working.
Technique: Free Form
Rows:
Landing page + query stringColumns:
Session sourceValues:
Sessions,Engaged sessions,Conversions(if configured)Filter: Session source matches regex (use the comprehensive regex above)
Date range: Last 90 days
Output is a table showing which pages get AI referrals, which platform sent the traffic, and what the engagement looked like. The pages that show up here are the pages AI platforms are citing. Expand the ones performing well, replicate the pattern across similar content.
Exploration 2: AI traffic trend line
Technique: Free Form with line chart visualization
Rows:
DateColumns:
Session sourceValues:
SessionsFilter: same AI regex
Date range: Last 12 months
Shows the trend of each AI platform over time. Great for board decks and trend identification (e.g., ChatGPT Atlas launching).
Exploration 3: AI conversion funnel
Only useful if you have conversion events configured in GA4.
Technique: Funnel Exploration
Steps: session start → key page view → conversion
Segment: sessions where Source matches AI regex vs. all other sessions
Compare the two segments side by side
Output shows whether AI-referred users convert at higher or lower rates than other channels. Adobe Digital Economy Index data from the 2025 holiday season suggested AI-referred retail traffic converted 31% higher than non-AI. Confirm the pattern holds for your business.
Exploration 4: Audience-level analysis
Create an Audience inside GA4 for AI-referred users. Condition: sessions where Source matches the AI regex within a trailing 30 days. Save as a persistent audience. Then run:
Cohort exploration: retention over time for AI-referred users vs. Organic Search users
Path exploration: where AI-referred users go next, which content they consume in sequence
User Lifetime exploration: LTV comparison across channels
Method 3: Channel Groups in Looker Studio
For dashboards that get shared with stakeholders, Looker Studio (formerly Data Studio) is the pattern to use. GA4’s Looker Studio connector does not pass your custom channel group automatically (as of April 2026), so build the channel in Looker Studio directly.
Create the AI channel calculated field
In your Looker Studio report, add a Calculated Field at the data source level (not the chart level, so every chart inherits it):
Name the calculated field Channel Group with AI. Use it as the primary dimension in all channel-comparison charts.
Note the double backslashes. Looker Studio’s REGEXP_MATCH requires double-escaped special characters compared to GA4’s regex engine.
The 5 core charts for an AI traffic dashboard
Sessions by channel over time (time series): Compares AI Search to every other channel. Use stacked area chart for visual density or line chart for per-channel readability.
AI platform breakdown (pie or horizontal bar): Sessions by specific AI platform. ChatGPT vs. Perplexity vs. Gemini vs. Claude and others.
Top AI-referred landing pages (table): Shows which pages AI is driving traffic to, sorted by sessions descending.
AI conversion rate vs. all channels (scorecard): Side-by-side CR comparison with month-over-month delta.
AI revenue (if ecommerce): Revenue per session from AI compared to other channels. Often 2-4x higher per-session revenue for ecommerce sites.
For how AI traffic economics play out specifically in D2C, see our Claude for ecommerce guide.
Method 4: BigQuery Queries for Advanced Teams
If your GA4 property exports to BigQuery (available on any property with BigQuery Export enabled), you have access to the raw events data and can run arbitrary SQL. Raw access unlocks analysis that is awkward or impossible in GA4’s UI.
Query 1: Daily AI sessions by platform
Replace your-project.analytics_XXXXXXXXX.events_* with your actual dataset. Modify the date range as needed.
Query 2: Landing pages with AI citation traffic and conversion rate
Swap 'purchase' for your own conversion event name (generate_lead, sign_up, etc.).
Query 3: AI platform share of voice over time
For tracking whether your brand is gaining share across AI platforms month over month:
Solving the Dark Traffic Problem
The honest limitation of GA4-based AI tracking: a meaningful share of AI-influenced traffic never shows up as AI in GA4 at all. Three specific dark-traffic leaks need separate handling.
Leak 1: Referrer-stripped sessions show as Direct
When ChatGPT or other AI platforms open outbound links with rel="noreferrer" or aggressive referrer-policy settings, the browser sends no Referer header. GA4 attributes the session to Direct. Mitigation: no client-side fix, but you can spot the pattern. Direct traffic that spikes in correlation with your AI visibility lift, or Direct traffic that lands on obscure deep pages rather than your homepage, is probably displaced AI traffic. Track the trend alongside your brand visibility audit output for better understanding.
Leak 2: Mobile apps strip everything
Traffic from the ChatGPT mobile app, Perplexity mobile app, Claude mobile app, etc., arrives through the user’s default browser with no referrer data. The session is indistinguishable from a direct visit. Mitigation: if you control the destination URL shared in AI responses (e.g., if you own the page ChatGPT is citing), you can proactively append UTM parameters, but this only works for links you place yourself, which is rare.
Leak 3: Google AI Overviews look like organic Google
AI Overviews sessions arrive from google.com with standard organic attribution. GA4 cannot separate them. The workaround: cross-reference Search Console query data with landing page patterns. Queries that show impressions-without-clicks at disproportionate rates, or that match Google’s AI Overview-trigger patterns (informational questions, comparison queries, long-tail conversational queries), are likely AI Overview exposures. You will not get per-session attribution, but you can get query-level AI Overview estimation. For context on the broader organic search disruption, see our analysis of Google’s September shake-up and how to recover.
Partial solution: proactive UTM tagging
For content you actively promote through AI channels (e.g., asking your team to share links from Claude conversations, or including links in external content AI platforms will likely cite), append UTM parameters. Example for a blog post you expect Perplexity to cite:
https://yoursite.com/your-post?utm_source=ai_visibility&utm_medium=ai&utm_campaign=q2_content_cluster
The tactic only helps for the subset of AI traffic originating from links you control, but it is strictly additive to the other methods.
Partial solution: conversion modeling
GA4’s data-driven attribution model can estimate AI-influenced conversions using machine learning across all channel touchpoints. The model is imperfect and consumes consent-signaled data only, but it catches AI-assisted conversions that deterministic attribution misses. Enable data-driven attribution in Admin → Attribution Settings, then compare modeled-conversion estimates to your AI Search channel’s deterministic conversions. The gap is your AI influence underreporting.
Segmenting AI-Referred Users
Raw session counts are the floor. The real value of AI attribution comes from understanding how AI-referred users behave differently than other channels.
Build the persistent AI audience
In GA4 Admin → Audiences → New Audience:
Name:
AI-Referred UsersConditions: include users where Source matches your AI regex at any time in the last 90 days
Save as a persistent audience
The audience populates automatically. Every future Exploration, Segment Comparison, and Predictive Metric can be filtered to or from this audience.
Key metrics to compare: AI-Referred vs. Organic Search
Metric | Why it matters |
Engaged sessions rate | AI users often have higher engagement because they arrived with context |
Pages per session | Higher on AI traffic typically; they often explore further |
Conversion rate | Adobe data showed AI conversions 31% higher in 2025 holiday retail |
Average order value (ecom) | Often higher for AI-referred; they are further down the research funnel |
Return visits within 30 days | Proxy for brand recognition stickiness |
Revenue per session | The unified metric; this is where Adobe measured +254% growth for AI referrals |
Teams that run this comparison for the first time routinely find that AI-referred users are disproportionately valuable per session. For the D2C-specific version of this analysis, our Claude for ecommerce guide covers the full workflow.
AI Traffic Dashboard Template (Looker Studio)
A production-ready Looker Studio dashboard for AI traffic has six components. Build once, share with stakeholders monthly.
Page 1: Executive summary
Scorecard: Sessions from AI Search (last 30 days, with month-over-month %)
Scorecard: AI conversion rate vs. site-wide (with delta)
Scorecard: AI revenue per session vs. site-wide (with delta)
Time series: AI Search sessions last 12 months
Pie: AI platform share (ChatGPT / Perplexity / Gemini / Claude / Copilot / Other)
Page 2: Content performance
Table: Top 25 landing pages receiving AI traffic (with sessions, engaged sessions, conversions)
Bar chart: Sessions by content topic cluster (using a calculated field that buckets URLs by path pattern)
Table: AI platform × Landing page heatmap (which platforms cite which content)
Page 3: Platform deep-dive
One page per major platform (ChatGPT, Perplexity, Gemini, Claude)
Each page: sessions over time, top landing pages, conversion rate, revenue
Page 4: Conversion analysis
Funnel chart: AI-referred users through conversion funnel
Table: AI conversions by campaign/landing page
Comparison: AI-modeled conversions (from data-driven attribution) vs. AI deterministic conversions
Sharing and permissions
Lock the data source so edits flow through a single admin. Grant view-only access to stakeholders. Schedule email delivery of the PDF version monthly for executive visibility.
8 Mistakes That Break AI Traffic Tracking
1. Using Source/Medium instead of Source
GA4’s “Source/Medium” field combines two values with a slash. Regex against this field is more fragile because medium changes based on UTM parameters, whereas the Source field is cleaner for domain matching. Use Source as the primary dimension in your channel group condition.
2. Forgetting to reorder the channel
A custom channel group that has AI Search listed below Referral will not classify anything as AI Search. GA4 assigns each session to the first matching channel top to bottom. Reorder AI channels to the top of the list every time you add new ones.
3. Only matching the www subdomain
www.perplexity.ai and perplexity.ai are different strings. Regex must match both ((www\.)?perplexity\.ai or a broader wildcard pattern). The comprehensive regex at the top of this guide handles subdomain variation; copy it verbatim if in doubt.
4. Forgetting to backslash the dots
chatgpt.com in a regex matches chatgpt followed by any single character followed by com, which would also match chatgptXcom or chatgptacom. Always escape dots with backslash: chatgpt\.com.
5. Running without updating the regex quarterly
New AI platforms launch, old ones get renamed, and subdomain patterns change. Review your regex every 90 days and add new platforms as they appear in your referral data. Our GEO guide covers the broader landscape shifts worth tracking.
6. Not tracking mobile apps separately
Mobile app traffic arriving as Direct is a meaningful share of AI traffic. If your audience skews mobile-heavy, the Direct bucket probably contains more AI traffic than your AI Search channel does. Track Direct traffic to AI-optimized landing pages separately.
7. Measuring AI traffic without measuring AI visibility
GA4 measures the sessions you do get from AI platforms. Nothing in GA4 measures whether AI platforms are citing you accurately, which competitors are cited alongside you, or how often your brand is mentioned in AI responses without a click. Pair GA4 tracking with a brand visibility audit methodology run monthly, and pair both with a schema markup playbook so AI platforms have the entity signals needed to cite you accurately when they do send traffic.
8. Treating low current volume as an excuse to skip setup
Typical AI traffic volume for a mid-market site in early 2026 is 0.5% to 3% of total sessions. Low absolute numbers. But Adobe’s holiday 2025 data showed 693.4% year-over-year growth, and trend lines matter more than point-in-time volume. Set up tracking now, even if the volume looks trivial, so you have a clean baseline for the growth to come. For the broader channel economics, see our analysis of whether AI search referrals are the new clicks.
Frequently Asked Questions
Why doesn’t Google just add AI Traffic as a default channel?
Google has publicly indicated they are working on it but has not committed a timeline. Until a native solution ships, custom channel groups are the right workaround. Reviewing your channel group every quarter catches new platforms before they materialize as unclassified Referral traffic.
How do I separate ChatGPT browsing vs. ChatGPT search vs. ChatGPT Atlas?
As of April 2026, all three route outbound clicks through chatgpt.com without reliable subpath differentiation in the referrer. Some sessions include subpath information (e.g., chatgpt.com/search, chatgpt.com/c/), which you can capture by using Page referrer rather than Session source in Explorations. For production channel groups, a single chatgpt.com match is more reliable.
What about Google AI Overviews specifically?
Google AI Overviews traffic arrives as standard Google organic and cannot be isolated in GA4. Use Search Console’s “Search appearance” filter (if your property is in the eligible set) combined with query-level analysis: impressions spiking without proportional click growth is a common AI Overviews exposure pattern. Our AI search readiness audit guide covers the full technical diagnosis.
Should I use utm_source=chatgpt.com if ChatGPT already auto-appends it?
ChatGPT sometimes auto-appends utm_source=chatgpt.com to outbound links. When present, the UTM parameter overrides referrer-based attribution. In practice, the behavior is inconsistent. For links you control (e.g., links you yourself share in AI conversations), append your own UTM parameters for deterministic attribution. For AI citations you do not control, the referrer-based channel group is your primary detection method.
Do I need BigQuery export, or is GA4 UI enough?
GA4 UI is sufficient for 80% of AI tracking needs. BigQuery unlocks the remaining 20%: custom joins, arbitrary date ranges, cross-property analysis, and integration with other data sources (CRM, warehouse, internal tools). If your team already uses BigQuery, enable GA4 export and run the queries above. If not, channel groups + Explorations + Looker Studio are the right stack. For teams using Claude to interpret the query results faster, our Claude for marketing guide walks through how to load GA4 exports into a Brand HQ Project for analysis.
How does this integrate with a consent management platform?
If your site uses consent mode v2 (recommended in the EU and increasingly in the US), sessions without consent are still counted but with anonymized data. AI referrals behave the same as any other traffic source under consent mode. The regex matching works on whatever referrer data GA4 receives after the consent layer resolves. No separate handling required.
Which channel do I optimize for first if I only have time for one?
ChatGPT. According to OpenAI + Harvard NBER research from September 2025, ChatGPT had 700-800 million weekly active users with 2 billion+ prompts daily. ChatGPT is the dominant AI platform by orders of magnitude and the most likely starting point for AI-driven visits to your site. Perplexity second (higher citation volume per response makes it disproportionately important for high-intent research queries), Gemini third, Claude fourth.
Your Next Move
Ship the custom channel group this week. Thirty minutes of setup. Paste the comprehensive regex above into GA4, reorder the channel above Referral, and you will see AI traffic as its own channel from that point forward.
In month two, set up the Looker Studio dashboard. In month three, if your GA4 exports to BigQuery, run the queries above monthly against a rolling 90-day window. At the end of quarter one, you will have a production AI traffic measurement system that compounds every month as AI volume grows.
Adobe’s data showed 693.4% year-over-year growth in AI-referred retail traffic during the 2025 holiday season, with AI conversions running 31% higher than non-AI and revenue per visit up 254%. Cloudflare Radar’s 2025 Year in Review documented AI bot traffic growing at triple-digit rates across every platform Cloudflare tracks. Forrester + 6sense 2025 research found 94% of B2B buyers using LLMs during their purchase journey. The traffic exists. The attribution gap is solvable with a few hours of setup.
If you want an expert set of eyes on where your brand actually shows up across AI search (and what fixes would move your numbers), get a free SEO and AEO audit from Passionfruit. We benchmark your visibility across ChatGPT, Perplexity, Gemini, and Google’s AI Overviews, pair it with your GA4 traffic data, and hand you a prioritized 30-60-90 day plan you can execute whether or not you work with us.
AI traffic attribution is not a reporting nicety. Rather, the measurement spine that lets you defend every AEO investment decision for the next 24 months.
