For two years the standard advice for tracking ChatGPT traffic in GA4 was to build your own regex and hope the referrers held still. In May 2026 Google added a native AI Assistant channel to the default channel group, and a lot of teams took that as the problem solved. It is not, quite. The native channel is useful, but it has a start date, a list of assistants that changes, and the same blind spots every referrer-based method has. This piece covers what the native channel does, the custom channel group I set up alongside it, the traffic neither can see, and how I measure the same thing in my own first-party analytics. If you are setting up GA4 from scratch, start with my GA4 setup page; this article assumes the basics are in place.
What the native AI Assistant channel does
Google's GA4 documentation on default channel groups now lists an AI Assistant channel, described as traffic from sources like ChatGPT, Gemini, DeepSeek, Copilot or Grok, and it states that the channel excludes Google's AI Overviews and AI Mode. When GA4 recognises the referrer as an AI assistant, the session is grouped there instead of under Referral. According to Search Engine Journal (Itamar Blauer, July 2026), the channel was added on 13 May 2026 and rolled out gradually, reaching most properties by early June. Google's help page itself does not show a launch date, so treat the exact date as reported rather than official. You do not have to configure anything. In Reports, open Acquisition, then Traffic acquisition, set the primary dimension to Session default channel group, and look for the AI Assistant row. Two practical notes. First, Google has not published the full list of referrers it recognises, only examples, so you cannot audit the rule the way you can audit your own regex. Second, the example list has already changed: the same Search Engine Journal piece reports that Claude was named at launch and dropped from the examples by June.
Where the native channel falls short
Three gaps matter for reporting. It is not retroactive. Sessions before the rollout stay in Referral, so any year-on-year or quarter-on-quarter comparison that crosses May 2026 will show a jump that is partly just reclassification. Perplexity is the most disputed gap. Google's channel description does not name it, and Search Engine Journal reports that Perplexity sessions continued to land in Referral. Some other write-ups claim it is covered. Because Google does not publish the list, I check my own property: filter Traffic acquisition to session source contains perplexity and see which channel those sessions sit in. Do that check yourself rather than trusting any article, including this one. The third gap is coverage drift. New assistants appear, existing ones change domains, and the native rule updates on Google's schedule, not yours. None of this makes the channel useless. It is a sensible default for people who will never touch a regex, and it means AI visits no longer vanish into Referral by default. It just should not be the only lens if AI search is something you are actively investing in.
The custom channel group I add alongside it
GA4 lets you create custom channel groups under Admin, Data display, Channel groups. I copy the default group, then add an AI assistants channel above Referral with a condition on session source matching a regex of assistant domains: chatgpt.com, chat.openai.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com, deepseek.com, grok.com, meta.ai, mistral.ai, you.com and a few others. I also match a session source of chatgpt.com or similar coming from the utm_source parameter, because ChatGPT commonly appends utm_source=chatgpt.com to the links it shows, and my own analytics is set up to catch that tag. The advantage, as Search Engine Journal also points out, is that a custom channel group applies its rules across your whole date range, so older ChatGPT sessions that sat in Referral move into the new channel and your trend line becomes comparable. Order matters: put the AI rule above Referral and Organic Search, but keep gemini.google.com separate from google.com so you do not sweep ordinary Google organic traffic into it. Keep the regex in a shared document with a change date, the same discipline I apply to UTM naming.
What neither method can see
Referrer-based tracking only works when there is a referrer. Many AI visits arrive without one. Mobile assistant apps and in-app browsers often strip it, so the visit lands in Direct. Someone who copies a URL out of an answer and pastes it into a new tab is Direct too. And a large share of AI influence never produces a click at all: the person reads the answer, remembers your name, and searches for you later, which shows up as branded organic search or direct traffic weeks afterwards. Google's own AI features are the other blind spot. Google's channel documentation excludes AI Overviews and AI Mode from the AI Assistant channel, so visits from them are reported as Organic Search, mixed in with classic results. GA4 gives you no way to separate them. The honest framing for a leadership report is that GA4 shows a floor for AI-referred visits, not the total. I add a short note under every AI traffic chart saying exactly that, because the first question a CFO asks is why the number is so small.
How I measure it in my own first-party analytics
On my own site I run first-party analytics alongside GA4, and AI assistants get their own channel there too. Each visit is classified when it is recorded: if the referrer host matches a known assistant, or the utm_source matches one, the channel is AI assistants. The same rules exist as a SQL CASE expression, so I can regroup older visits whenever I add a new assistant, which solves the retroactivity problem without waiting for anyone. The admin report then shows visits from assistants over time, a breakdown by assistant, and the first page of each visit, which is effectively a list of the pages AI tools link to. Because the events table also records leads and bookings, I can see whether visitors from assistants convert, not just whether they arrive. I pair that with server-side logs of AI crawlers, described in my piece on tracking AI crawlers, so I can compare pages AI reads with pages AI sends people to. Results are tracked in my own dashboard. The method matters more than my numbers, and it works the same way for a client site.
A reporting setup I would recommend
For most B2B teams I suggest four layers. Keep the native AI Assistant channel as the default view, because it is what everyone else in the business will see. Add the custom channel group for trend analysis and for assistants Google may not cover. Add a self-reported attribution field on your main lead form, with an option for ChatGPT or another AI assistant, because people will tell you about influence that left no referrer. And watch branded search impressions in Search Console as a lagging signal of AI visibility. Then connect it to pipeline. A channel that sends a handful of visits but produces qualified leads deserves attention; one that sends hundreds of visits to a single glossary page may not. In the CRM, carry the original channel onto the contact and the deal, so AI-referred pipeline can be reported next to every other source. If your event setup is messy, fix that first, because no channel group survives broken lead events. My GA4 event taxonomy piece covers the common mistakes.
Sources
Google Analytics Help, [GA4] Default channel group: https://support.google.com/analytics/answer/9756891. Search Engine Journal, Itamar Blauer, GA4's AI Assistant channel undercounts your AI traffic: how to build one that doesn't (15 July 2026): https://searchenginejournal.com/ga4s-ai-assistant-channel-undercounts-your-ai-traffic-how-to-build-one-that-doesnt/580133
FAQ
Mostly. GA4 groups visits with a recognised AI assistant referrer, including ChatGPT, into the AI Assistant default channel. It only counts from the rollout in May 2026, and visits from apps that strip the referrer still land in Direct.
Google's documentation does not name Perplexity among its examples, and Search Engine Journal reports Perplexity sessions staying in Referral. Check your own property by filtering session source for perplexity, and add it to a custom channel group if it is missing.
In Organic Search. Google's channel definition explicitly excludes AI Overviews and AI Mode from the AI Assistant channel, so they are mixed in with ordinary organic results and cannot be separated in GA4.
Both. Keep the native channel as the shared default, and use a custom group for trend lines, because custom groups apply across your full date range and let you include assistants Google may not recognise yet.