Default GA4 has no AI Assistants channel. A custom channel group with a Source matches regex rule, placed above the Referral channel, routes ChatGPT, Perplexity, Claude, Gemini, and Copilot sessions into a clean bucket and applies retroactively to historical data. The setup takes 15 minutes. It will recover roughly 30% of your AI traffic; the rest arrives without a referrer and stays in Direct.
This is the one universal step every brand should ship this week. It is not the whole answer – Loamly's February 2026 analysis of 446,405 visits put 70.6% of AI traffic in the Direct bucket because mobile apps and free ChatGPT do not pass referrer headers. But the visible 30% is enough to start sizing the channel, watching its trajectory, and stopping the "is this AI traffic or someone typing our URL" guesswork. The strategic context – why the click-side data is structurally incomplete and how to read leading indicators around it – lives in the dark SEO funnel. This piece is the implementation playbook it points to.
Why default GA4 hides AI traffic in three different buckets
GA4's default channel grouping was frozen before generative AI became a referral channel, and Google has not added an "AI Assistants" channel to the default set as of May 2026. The result is a three-way scatter. Sessions from ChatGPT with a usable referrer (mostly desktop Search-mode citation links) land in Referral as chatgpt.com. Sessions from free ChatGPT on mobile, Atlas in-app, or any conversational inline link arrive without a Referer header and land in Direct. Sessions from Google AI Overviews and AI Mode keep the Google referrer and land in Organic Search alongside regular SEO clicks.
None of those buckets is wrong on its own. The problem is that none of them is "AI." A single ChatGPT recommendation can produce sessions in all three buckets depending on the user's device and the link type, and a marketing leader reading the default Acquisition report has no way to read AI as a channel without restitching the data. The custom channel group is what stitches it.
The 15-minute setup, in five steps
You need Editor access on the GA4 property. The work is in the admin panel, not the reporting UI, so changes apply to every report and explore in the property at once.
Open Channel groups. In GA4, click the Admin gear at the bottom left, then under Data display click Channel groups. You will see the default group plus any custom groups already created.
Duplicate the default group. Click the three-dot menu on Default Channel Group and choose Duplicate. Rename the copy to something like "Default + AI" so every channel except the new one mirrors GA4's defaults. Editing the duplicate, not the default, keeps your original reports clean.
Add a new AI channel. Inside the duplicated group, click Add new channel. Name it "AI Assistants" (or "AI Tools" – the label is what shows up in reports). For the condition, choose Source as the dimension, "matches regex" as the operator, and paste the regex from the next section. Save the channel.
Reorder above Referral. GA4 evaluates channels top to bottom; the first match wins. If Referral sits above the new AI channel, every chatgpt.com session matches Referral first and your AI bucket stays empty. Drag the AI Assistants channel above Referral and above any custom Paid or Branded channels you have already configured.
Save and wait for backfill. Save the group. GA4 reprocesses historical data into the new grouping within minutes to a few hours, and the AI Assistants channel becomes available in any report that exposes a Default Channel Group selector – Reports, Explore, Advertising. No retro tagging is required.
The exact regex for May 2026
This is the Source-matches-regex pattern we use for Parse customers as of May 2026. It covers the platforms with non-trivial referral share plus a small tail for emerging models. Anchors at both ends matter: an unanchored chatgpt substring will pull in news articles about ChatGPT and inflate the bucket.
^(chatgpt\.com|chat\.openai\.com|openai\.com|gemini\.google\.com|bard\.google\.com|perplexity\.ai|claude\.ai|anthropic\.com|copilot\.microsoft\.com|edgeservices\.bing\.com|deepseek\.com|grok\.com|x\.ai|meta\.ai|you\.com|poe\.com|phind\.com|character\.ai|mistral\.ai|huggingface\.co|kagi\.com)$
Two clarifications about what the regex does and does not include. It targets Source (the platform's hostname), not Source / Medium, so it catches both chatgpt.com / referral and any session GA4 has stamped with Source: chatgpt.com from a utm_source parameter. It does not include google because Google AI Overviews share the google.com referrer with regular Organic Search; pulling Google in here would misclassify standard SEO traffic as AI. AI Overviews need a separate workaround, covered later in this post.
Parse tracks AI visibility across ChatGPT, Google AI Overviews, and Perplexity, which is what you fall back to once you accept that the click-side number will always be a fraction of the actual AI-influenced demand.
Channel order is the step that breaks most setups
Every public guide we reviewed flags channel order, and every Parse onboarding call where a customer reports "the channel is empty" turns out to be an ordering problem. GA4's matcher is not "best fit" – it is "first match." The default Referral rule simply checks Source != (direct) and has no domain exclusion, so any session with a referrer matches Referral before reaching a custom channel below it. Your AI Assistants bucket reads zero, the Referral bucket inflates, and on the surface nothing looks broken.
The fix is the drag-and-reorder step in the setup. Inside the channel group editor, click Reorder, then move AI Assistants above Referral. While you are there, also move it above any other custom channels that condition on broad Source patterns. If you maintain a Branded Paid or Influencer Referrals channel that uses domain-level rules, AI should sit at the top of that stack so it claims those sessions first. After saving, validate by opening Reports → Acquisition → Traffic acquisition, swapping the Default Channel Group for your new group, and confirming the AI Assistants row populates within 24 hours.
Verify the channel is firing within 24 hours
A channel group change takes effect immediately in real-time, but the reporting backfill is what most teams want to confirm. Two checks rule out the most common failure modes.
First, in Reports → Acquisition → Traffic acquisition, change the report's Default Channel Group to your new "Default + AI" group via the dropdown above the table. The AI Assistants row should appear within minutes. If it shows zero sessions over a 7-day window, your regex is fine but channel order is wrong, or the property has not received any AI-referred sessions yet. Cross-check by adding a Source filter for chatgpt.com to a Free Form Explore – if sessions exist in the Source view but not in the AI channel, ordering is the issue.
Second, in Explore → Free Form, build a quick table with Default Channel Group (Default + AI) as the row dimension and Sessions, Engaged sessions, Conversions as values. Filter the date range to the last 30 days. The AI Assistants row tells you the channel volume; the engagement and conversion columns tell you the quality. Most properties see disproportionate engagement from AI sessions versus their volume share, which is the first justification slide for the budget review.
If you want to know when AI changes its answer about your brand, start with a free brand check — it takes a minute.
What this captures, and what it cannot
Honesty about coverage is the difference between a useful dashboard and a slide that misleads the executive team. Two ranges to anchor expectations on.
The 70.6% figure is the number you should keep in mind when reading the new channel. Of the AI sessions you can prove are AI, roughly seven in ten arrive with no Referer header and stay in Direct. The 30% the channel does see breaks down predictably: most ChatGPT desktop Search-mode clicks, all Perplexity Comet visits, all Claude.ai citation links, all Copilot referrals on Edge desktop, and the vast majority of Gemini traffic. The 70% you do not see breaks down predictably too: free ChatGPT on mobile (no referrer at all), ChatGPT Atlas (WKWebView strips the Referer), conversational inline links on the paid tier (rel="noreferrer"), and copy-paste navigations. The custom channel group is the right floor; it is not the ceiling.
A common mistake at the next leadership review is to point at the channel and say "this is our AI traffic." It is not. It is the visible 30% slice. Treat it as a sample of a larger population, and pair it with branded search, demo-form self-attribution, and Share of Model so the conversation is grounded in directional change instead of a single, structurally low number.
Layer two: UTM tags and the demo-form backstop
Two cheap additions raise the visible share without requiring a re-platform. Both are worth doing the same week as the channel group, while the project is in flight.
The first is a UTM convention for any AI surface you can influence. Anything you publish that an AI assistant might cite – a press release, a guest post, a documentation page – gets a utm_source=ai plus a utm_medium of the platform when known. AI scrapers preserve URL parameters in the citation snippet, so when a user clicks through, the parameters travel into GA4 even if the Referer is stripped. ChatGPT also auto-appends utm_source=chatgpt.com on desktop citation links it generates itself; your channel group already catches that traffic via the Source field, but consistent first-party tagging extends the coverage to inline links that ChatGPT would otherwise leave bare.
The second is a "How did you hear about us?" field on every demo, signup, and pricing-contact form. Include named options for "ChatGPT," "Perplexity," "Claude," "Google AI Overview," "Gemini," and "Recommended by an AI assistant." This is the single highest-leverage measurement upgrade we recommend, because it captures attribution at the exact moment the buyer is most willing to share it, and it is robust to every referrer-stripping mechanism the platforms ship next quarter. Pair the field with GA4 conversion events on form submit, and you have a per-form AI attribution stream that complements the channel group.
What to do about Google AI Overviews and AI Mode
AI Overviews and AI Mode are the largest measurement gap and the one with the cleanest workaround. Both surfaces send traffic with google.com as the referrer and Organic Search as the channel, so the regex above intentionally excludes Google to avoid contaminating standard SEO reporting. To break out the AI portion, you build a second custom dimension downstream of Search Console.
Inside Google Search Console, the Performance report exposes a Search Appearance dimension with an "AI Overviews" filter as of late 2025, and AI Mode appearances are surfaced in the same panel. Export the AI Overviews queries weekly to BigQuery or a Looker Studio data source, then join against your GA4 landing-page data on URL. The output is a derived metric – AI Overviews referral share – that lives outside GA4's channel grouping but feeds the same dashboards. We use this layered model with Parse customers because it preserves the GA4 contract while giving the executive view a single AI line item. The strategic framing for what to do with that signal lives in how to report AI visibility to your CEO.
The first report to build after the channel goes live
Once the channel is firing, build one report and ignore everything else for two weeks. The point of the first dashboard is calibration, not optimization.
The minimum viable report has four lines on a weekly time series: AI Assistants sessions, Direct sessions, Branded organic sessions (Search Console queries containing your brand), and total conversions. Plot all four on a single chart with shared axes. The AI Assistants line is your visible-AI floor. The Direct line is the dark-AI ceiling – most of the rise in Direct over the last two quarters is AI you cannot see. The Branded organic line is the leading indicator of AI-influenced demand reaching your site through verification searches. Conversions tell you whether the entire stack is healthy independent of channel attribution.
Watching the four lines together for a month does more for budget conviction than any single attribution model. When AI Assistants and Branded organic both trend up while Direct stays flat, your AI work is converting a growing share of recommendations into in-channel clicks. When Branded organic rises and Direct rises together while AI Assistants stays flat, your dark-funnel share is growing faster than your visible share – a sign that mobile and Atlas are taking share. The weekly review cadence we recommend for this dashboard is in the weekly AI visibility review.
Frequently asked questions
Why does ChatGPT traffic show as Direct in GA4?
Four mechanisms strip the Referer header on the way out of ChatGPT: a strict-origin-when-cross-origin policy on chatgpt.com, rel="noreferrer" on paid-tier inline links, mobile WKWebView and Custom Tabs that drop headers during navigation, and copy-paste URLs that carry no referrer at all. Loamly's February 2026 analysis of 446,405 visits found 70.6% of confirmed AI sessions arrived without a Referer header, which GA4 has nothing to label and routes to Direct.
Will the custom channel group apply to historical data?
Yes. GA4 channel groups apply retroactively across the full retention window of the property. After saving the group, every Reports and Explore surface that exposes a Default Channel Group selector can be re-rendered against the new group, and the AI Assistants channel populates with all matching historical sessions. The session-level fields are unchanged; only the labeling logic is new.
Should I edit the default channel group instead of duplicating it?
Duplicate it. The default group is what every other Google product, integration, and external Looker Studio template assumes when it pulls "channel" data. Editing the default is reversible, but it changes the meaning of "Organic Search" and "Referral" for everything downstream. A duplicated "Default + AI" group keeps the original contract intact and lets you toggle between views per report.
What about Google AI Overviews and AI Mode? Why are they not in the regex?
AI Overviews and AI Mode use the google.com referrer and stay in Organic Search. Including google in the AI regex would misclassify standard SEO traffic. The cleaner workaround is to extract the AI-Overview impressions and clicks from Google Search Console's Search Appearance filter, join them to GA4 landing-page data outside the channel grouping, and surface the result as a separate dashboard line. This preserves both reports.
How do I tell whether a Direct session is actually AI?
At the session level, you cannot. At the cohort level, three signals stack: a rising share of Direct landing on deep pages (pricing, comparison, product detail) rather than the homepage, a higher conversion rate on Direct than on Organic, and growth in Direct that correlates with Share-of-Model gains in your Parse dashboard. Self-reported attribution on demo forms – with named AI sources as options – is the closest thing to a per-session signal and is worth shipping the same week as the channel group.
:::
The custom channel group is a 15-minute change that turns one of three default GA4 buckets into a usable AI line item. It will not solve the dark funnel – most of the AI sessions a buyer makes will still arrive without a referrer and stay invisible to your analytics – but it will tell you the floor, the trajectory, and the engagement quality of the visible slice. Pair it with a self-attribution form question and a weekly Branded Search line, and you have the click-side picture you need to brief your team this week and the leadership review next month.