GA4 AI Assist Channel: How to Track, Isolate, and Attribute AI Traffic
GA4's AI Assistant channel labels chatbot visits but misses Direct traffic and Google's AI surfaces. Here is how to isolate, attribute, and govern it.

For a long stretch, a visitor who clicked a link inside ChatGPT, Perplexity, or Claude arrived in Google Analytics 4 looking like any other site that happened to link to you. The session landed in Referral, mixed in with a partner blog and a forum thread, and nobody on the team could say what AI assistants were worth.
In May 2026, Google changed part of that. GA4 now identifies visits from chatbots like ChatGPT, Gemini, and Claude through a new AI Assistant channel in the Default Channel Group. That is real progress, and it creates a new risk: teams will open the report, see a row labeled AI Assistant, and assume the measurement problem is solved.
It's not. A label is not a measurement system. The native channel captures only sessions GA4 can identify from the referrer, and a channel row says nothing about whether AI-driven visits convert, which pages they land on, or how much influence they carry before a deal closes. This playbook covers the full job: finding AI traffic in your Google Analytics 4 property, isolating it to channel grouping you control, attributing honestly, in order to keep your whole setup accurate as the ecosystem shifts.
Why AI Traffic Hides in Referral, Direct, and (Other)
GA4 assigns every session to a channel using rules applied to source, medium, and campaign values. When no rule claims a session, it falls through to a catch-all. Three failure modes explain where AI visits have been going.

Referral. Most chatbot clicks arrive with a referrer header that names the assistant's domain. Before the native channel, all AI chatbot traffic landed in the Referral bucket by default, because GA4 had no AI-specific rule. This is the cleanest case: the data was present, just mislabeled. If you have ever pulled a Google Analytics Referral report to understand where your newest traffic comes from, you were looking at AI visits and partner links in the same table.
Direct. Some AI visits arrive with no usable referrer at all. A user who copies a link from a chat window and pastes it into a new tab, or who clicks through from a mobile app's in-app browser, often shows up as Direct. The new channel only captures what GA4 can identify through the referrer, and GA4 cannot label what the referrer header never told it.
Unassigned and (Other). Sessions with unusual or inconsistent source and medium combinations can fall into Unassigned. If your team has tagged links circulating in places assistants scrape and cite, such as an old campaign URL, odd tagging can push AI-originated visits here too.
The misconception worth naming is that Direct means "people who typed our URL." For a brand with any online footprint, Direct is better read as "sessions GA4 could not attribute." That bucket is exactly where AI-driven demand leaks, and it is why reading Direct growth as brand strength can be a costly mistake.
The Detect, Isolate, Attribute, Govern Loop
Treat AI channel measurement as a loop, not a one-time setup. Every step answers a different question, and skipping one produces a number you cannot defend.

- Detect. How much AI-driven traffic exists today, and where is it sitting? This is an audit of your current data.
- Isolate. How do we separate it into a stable, reportable channel that we control, whatever Google's default does? This is the channel grouping build.
- Attribute. What is that channel actually worth, once last-click bias is accounted for? This is where marketing attribution discipline applies.
- Govern. Who owns the source list, and when does it get rechecked? This keeps the setup honest as new assistants appear and referrer behavior changes.
The reason for a loop is that Step 4 feeds Step 1. A new assistant that starts sending traffic is invisible to a channel rule written last quarter, and it will sit in Referral until a scheduled audit catches it. Teams that treat this as a one-time project end up with a precise-looking report built on a stale list.
Detect: Audit Your Existing Chatbot Sessions First
Before building anything, establish a baseline. Whatever you build next will be judged against it.

In GA4, open Reports, then Acquisition, then Traffic acquisition. Change the primary dimension to Session source / medium, then use the search box to filter for the assistants you care about: chatgpt, perplexity, gemini, copilot, claude. Run the same exercise in the User acquisition report. The two answer different questions: User acquisition is scoped to new users, while Traffic acquisition is scoped to new sessions, so a returning visitor who first discovered you through an assistant can look very different in each.
Record four things for each assistant source you find:
- Sessions and engaged sessions. Engaged sessions tell you whether visitors did anything beyond bouncing.
- Key events. Whatever you have configured as a conversion, such as demo requests, purchases, or form fills.
- Top landing pages. AI assistants may cite deep informational pages, so the landing page distribution can look different from organic search.
- The channel GA4 currently assigns. If you still see these sources in Referral, the native channel has not reached your property yet or the source is not on Google's recognized list.
Then check the leakage side. Pull Direct and look at landing pages. A cluster of Direct sessions landing on deep, informational URLs that nobody would type from memory is a signal worth noting. It is an inference rather than proof, so treat it as a flag for follow-up and not as a count to add to your AI total.
One low-tech addition pays for itself: add a "How did you hear about us?" field to your main lead form, with an option for AI assistants. Self-reported attribution is imperfect, but it catches the buyers who read an answer in ChatGPT, then arrived through a branded search a day later. Analytics will never connect those two events.
Save a snapshot of this baseline. The most common mistake in this workflow is changing the configuration before recording what the data looked like, which makes every later comparison impossible to trust.
Check What GA4 Detects Natively vs. What You Must Define
Google's native AI Assistant channel is a default channel group rule. When a session's referrer matches a recognized AI assistant, GA4 sets the medium to ai-assistant and the campaign to (ai-assistant), and the session lands in the AI Assistant channel. No regex and no developer time are required, which is a meaningful improvement.

It also has limits, and an operator should know exactly where they sit:
- Referrer-dependent. Only sessions with a readable referrer header can be classified. Copied links and many in-app browser visits will still land in Direct.
- Google's AI surfaces are excluded. Google's own definition says the channel excludes Google's AI Overviews and AI Mode, which stay in Organic Search. The AI Assistant row therefore measures third-party assistants, not all AI-influenced search. Reporting it as "our AI traffic" understates the picture and misleads anyone who assumes it includes Google's own answer experiences.
- A partially disclosed list. Google's channel definition names examples such as ChatGPT, Gemini, Deepseek, Copilot, and Grok, but it does not publish the full referrer list in the article and says definitions may evolve as the market changes. You cannot audit exactly what the rule covers, so you cannot assume a given assistant is included. Google does accept requests to add sites and reviews them at least once a year, which is far slower than the market moves.
- Not editable. Default channel groups can't be edited. You cannot add a source it misses, and you cannot change the order in which its rules are evaluated.
- Rollout variance. Defaults roll out gradually. Confirm the AI Assistant row exists in your own Traffic acquisition report, under the Session default channel group dimension, before relying on it.
Native detection answers "did GA4 recognize this referrer?" A custom channel group answers "does this match our definition of AI-assisted traffic?" Those are different questions, and the gap between them is where niche assistants, regional tools, and emerging answer engines fall through. Use the native channel as your baseline reference and your custom group as the system of record.
Isolate: Build the Custom Channel Group
A custom channel group is a rule-based set of channels you define yourself, usable as a primary dimension in the same reports that support the default group. GA4 evaluates the channels in a group from top to bottom, and traffic is included in the first channel whose definition it matches. Rule order is therefore the whole game. This is the step teams get wrong.
Google's own documentation now includes this exact use case, so you are following the documented path and not a workaround.
The build, in sequence:
- In Admin, open Channel groups and create a new channel group. Starting from a copy of the default group keeps every standard channel intact.
- Add a new channel named "AI Assistants."
- Define it with a "matches regex" condition on Session source covering your target domains. A workable starting pattern is chatgpt\.com|chat\.openai\.com|perplexity|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai. Google's published example pattern is broader and covers ChatGPT, Gemini, Copilot, Claude, and Perplexity, so start from whichever fits your audit and tighten from there.
- Move AI Assistants above Referral. If Referral sits higher, it claims the session first and your channel never fires. Google's guidance is to place the channel above Referrals, and above others such as Organic Search if needed.
- Reorder, apply, and save the group.
A few operational notes. Custom channel groups can be applied to your reports retroactively, so you can compare your definition against past periods without waiting for new data to accumulate. The exception is the primary channel group: if you promote a custom group to primary, its criteria populate that dimension on future reporting only. Standard properties are limited to two custom channel groups, so name and scope yours deliberately instead of spawning experimental versions.
Be careful with the regex. An overly loose pattern, such as matching any source containing "gemini," can pull in unrelated traffic and quietly corrupt the channel. An overly tight one misses variants. Validate by comparing your custom channel's session count against the source-level numbers from your baseline audit. They should reconcile closely, and any large gap deserves explanation before anyone reports on it.
A trade-off to decide deliberately: if your custom definition and Google's native one overlap, you will see slightly different totals in each. Keep both live during the transition and document which one your dashboards use. Retire nothing until the numbers have reconciled for a full reporting cycle.
Segment and Explore: AI-Assisted Users, Landing Pages, and Paths
A channel gives you a total. Explorations tell you what those visitors do. Three views matter most.
A segment of AI-referred users. Build a user or session segment from your AI Assistants definition, then apply it in Explorations. This isolates behavior without relying on the channel dimension in every report you open.
Landing page analysis. Use a free-form exploration with landing page as the row, your AI segment applied, and sessions, engaged sessions, and key events as values. The pages that attract AI citations are rarely your homepage -- They tend to be definitional guides, comparison pages, and how-to content, though your own data is the authority on this. Knowing which ones tells you what to protect, refresh, and extend. It also tells your content team which pages are doing authority work that organic rankings alone would not reveal.
Path exploration. Look at what AI-referred visitors do after landing. If they arrive on a guide and leave, the page informs without converting. If they continue to a service page, the content is carrying commercial weight. Either way, you are now measuring a decision-relevant behavior rather than a vanity count.
A practical scenario: a services firm finds that its AI-referred sessions land overwhelmingly on two explainer articles, with a low rate of key events on those pages but a visible rate of branded search visits from the same users in later sessions. The explainers are not converting directly. They are seeding demand that closes elsewhere. Without segmenting, that firm would see the AI channel as weak and might cut the content investment that is feeding it.
Attribution Reality: Why Last-Click Undercounts AI Influence
This is the section that separates a tidy report from an honest one.
AI assistants behave like an early-stage research layer. A buyer asks a question, reads a synthesized answer, may click once, and often returns later through a branded search, a direct visit, or a typed URL. Under last-click logic, the credit goes to the final touch, and the assistant that shaped the shortlist gets nothing. This is the same structural bias that has long undervalued awareness channels, and it appears again with AI.
GA4 offers a way to test how much that bias matters. In the Advertising workspace, go to Attribution, then Attribution models. The Attribution models report, formerly called Model comparison, lets you compare how different attribution models change the valuation of your marketing channels. The Attribution reports offer three models: data-driven, paid and organic last click, and Google paid channels last click. Compare data-driven against paid and organic last click and look at what happens to the AI Assistant channel. If it gains credit under data-driven attribution and loses it under last-click, you have evidence that the channel participates earlier in the path than the default view shows. One practical check: confirm whether your custom channel group is available as a dimension in each Attribution report, and if it is not, read the native AI Assistant row there.
Three disciplines keep this honest:
- Read the Attribution paths report for the channel. Look for AI Assistant appearing early, followed by Organic Search or Direct at the close.
- Do not add Direct "estimates" to the AI total. The inferred sessions from your audit are a hypothesis, not data. Report them separately, labeled as such.
- Pair analytics with a human signal. Self-reported attribution from your lead form and sales call notes validates what the model suggests. When the two sources agree, you can act with confidence. When they diverge, you have found a measurement gap worth investigating.
The point of GA4 attribution work here is calibration rather than a perfect number. You are establishing a defensible range for what AI-driven demand is worth, so that budget and content decisions stop treating it as zero. Anyone building a broader marketing attribution practice should treat the AI channel as one more early-touch source to model, and the same discipline applies to the rest of the stack.
Reporting Views: A One-Page AI Channel Dashboard
Once the channel is defined and validated, give leadership one page that answers a single question: is AI-driven demand growing, and is it worth anything? Build it in a GA4 report, a saved exploration, or Looker Studio, with these components:
- Sessions and engaged sessions by week, AI Assistants channel only, with engagement rate shown beside the volume so growth is not mistaken for quality.
- Key events and revenue attributed to the channel, with the attribution model stated on the face of the chart.
- Top ten landing pages for the channel, which doubles as your content priority list.
- Assistant-level breakdown, using session source, so you can see which assistants matter to your audience.
- A comparison line against Organic Search, so the channel is read in context instead of in isolation.
Resist adding more. A dashboard with thirty tiles is a dashboard no executive opens twice. The design test: someone with two minutes should be able to answer whether AI is growing, what it converts, and which pages carry it.
Label the page with its limits. A single line stating that the channel counts referrer-identified sessions from third-party assistants only, that Direct may hide additional AI-driven visits, and that Google's AI Overviews and AI Mode sit in Organic Search, prevents the most common misreadings of the data.
Govern: QA and Quarterly Maintenance
The source list will drift. New assistants launch, existing ones change domains, and in-app browsers change how referrers pass. A channel built once and never revisited degrades quietly, and no alert will tell you.
Put four controls in place:
- Name an owner. One person is responsible for the channel definition. Without a named owner, the list is nobody's job.
- Run a quarterly review. Each quarter, re-run the Detect audit. Search source / medium for AI-related sources that are still landing in Referral or Direct, compare your custom channel against the native one, and update the regex. Google's own guidance is to update the regex whenever the URLs or the list of assistants changes.
- Keep a change log. Record every edit to the channel group with a date and reason. When a trend line breaks, the log tells you whether the cause is the market or your configuration.
- Reconcile on a schedule. Compare reported AI-channel conversions against CRM records and self-reported attribution. Divergence is information, not an error to suppress.
The failure mode here is false precision. A confident number that quietly excludes a quarter's worth of new sources misleads more than an honest range does.
The Bottom Line
GA4's native AI Assistant channel is a useful default, and it is the starting point rather than the finished system. It labels sessions from third-party assistants that carry a readable referrer. It does not tell you how many AI-driven visits are hiding in Direct, how Google's own AI surfaces are performing, what those visitors are worth, or whether the list behind it is current.
Run the loop. Detect what you have, isolate it in a channel grouping you control, attribute it with the bias of last-click in mind, and govern the definition on a schedule. Done once, this takes an afternoon. Maintained quarterly, it gives you something most teams still lack: a defensible answer to what AI-driven demand contributes to revenue.
If your AI-channel numbers still don't reconcile with revenue, a Performance Marketing Audit will show where attribution is leaking. Book a call.
- GA4
- AI Assistant Channel
- AI Search Traffic
- Channel Grouping
- Marketing Attribution
- Google Analytics 4
- Marketing Analytics



