ChatGPT and Perplexity Traffic in GA4: The AI Assistant Channel and 4 Gaps (2026)

Key takeaway: On May 13, 2026, GA4 added an AI Assistant channel to its default channel group, so sessions that arrive from a link in ChatGPT, Gemini and similar AI services are classified with no setup. You'll find it under Reports > Acquisition > Traffic acquisition. The channel only includes sessions whose referrer matches GA4's list of AI services, though, so visits from AI apps, clicks from Google AI Overviews and AI Mode, sources recorded under a service name alone, and data from before May 13 are counted in other channels. Read the AI Assistant row as a floor for AI traffic, and close the gaps by cross-checking session source / medium and building a custom channel group.

---

Three-line summary

• You can see AI traffic in GA4's standard reports. Since May 13, 2026, the AI Assistant row in the Traffic acquisition report shows sessions from ChatGPT, Gemini and similar services.
• That row is a floor, not the total. App clicks without a referrer, Google AI feature clicks, name-only sources and pre-May data are not in it.
• Cross-checking session source / medium and building a custom channel group narrows the gap. Custom channel groups also apply to past data.

GA4 classifies AI visits into the AI Assistant channel automatically

Since May 13, 2026, GA4 has classified sessions from AI services such as ChatGPT and Gemini into the AI Assistant channel of the default channel group, with nothing for site owners to configure.

Google Analytics Help's What's new page announced AI Assistant traffic measurement in its May 13, 2026 entry, naming ChatGPT, Gemini and Claude as examples. The Default channel group article defines the channel this way:

"AI Assistant is the channel by which users arrive at your site from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok." — Google Analytics Help: Default channel group

Classification is based on the referrer. According to the same article, when the referrer matches GA4's list of AI assistants, the medium is set to ai-assistant and the campaign to (ai-assistant). Sessions whose medium already exactly matches ai-assistant fall into the same channel.

Item	Detail
Added	May 13, 2026 (GA4 What's new)
Where to look	Reports > Acquisition > Traffic acquisition, dimension "Session default channel grouping"
Rule	Referrer matches the AI assistant list, or medium is ai-assistant
Examples in Google's docs	Release note: ChatGPT, Gemini, Claude / Channel definition: ChatGPT, Gemini, Deepseek, Copilot, Grok
Not published	The full list — Perplexity appears in neither set of examples

The unpublished list matters in practice. The examples are framed as "sources like," so the docs alone don't tell you whether a service missing from the examples is on the list. That is why you need to check your own data.

4 kinds of AI visits the AI Assistant channel misses

Because the AI Assistant channel identifies AI services by referrer, AI visits that arrive without a referrer or are recorded in a format outside the list are counted in Direct, Organic Search, Referral or other channels.

Type	Where GA4 counts it	Basis
Visits from AI apps or copied links	Direct	No referrer passed — industry checks (Ahrefs 2025, SearchPilot 2026)
Clicks from Google AI Overviews and AI Mode	AI Overviews in Organic Search, AI Mode mostly in Direct	SearchPilot summary (2026-05)
Sources recorded under a service name alone or a changed host	Varies by format	RanketAI's own check (2026-10)
Sessions before May 13, 2026	Referral or Unassigned	Google's docs don't mention reclassification

AI app clicks. When Ahrefs clicked links in each AI service by hand in May 2025, the Perplexity desktop app, Copilot in Windows and Grok passed no referrer (Ahrefs). A May 2026 SearchPilot roundup of earlier research says the ChatGPT and Claude mobile apps pass almost no referrer, and the Perplexity app passes one only about 70% of the time (SearchPilot). Both are link-level checks and summaries, not click-weighted statistics, so read them for the direction — "app visits can be counted as Direct" — rather than the exact ratios.

Google AI feature clicks. The same SearchPilot piece says AI Overviews clicks arrive with a google.com referrer and can't be told apart from regular organic search in either GA4 or Search Console. AI Mode strips the referrer from most clicks, so they mostly land in Direct, with some in Organic Search. This is a single summary, so confirm current behavior in your own data.

Data before May 13. Google's docs don't say whether sessions from before the AI Assistant channel launched are reclassified. To compare before and after, you need the custom channel group in Step 4 below.

RanketAI's own GA4 check: AI source values came in more than one format

When RanketAI checked its own GA4 in October 2026, AI service visits were recorded not only by host but also by service name alone, so a count based on a host list missed some of them.

On October 20, 2026, RanketAI reviewed all-time session source values in the GA4 property for its own site. A count that treated only values matching a host list (chatgpt.com, perplexity.ai, gemini.google.com and so on) as AI traffic was missing two formats:

Format	Session source value	What we saw
Service name only	perplexity, gemini	gemini sessions were classified as AI Assistant; perplexity sessions had an empty medium
Changed host	copilot.com	Missed by a count that used only copilot.microsoft.com

After adding the name values and copilot.com to the criteria, RanketAI's own count matched the session count of GA4's AI Assistant channel over a three-day window. RanketAI updated how it counts AI traffic based on this check.

This is one site, so it doesn't prove other sites will see the same formats. It does show that the assumption "AI traffic = a fixed host list" can fail on real data. Sessions with an empty medium may also be classified into a different channel, so check session source values directly instead of looking only at the channel row.

Step 1: Find the AI Assistant row in the Traffic acquisition report

Open Reports > Acquisition > Traffic acquisition in the left menu, and the AI Assistant row appears in the table for the default dimension, Session default channel grouping.

The Traffic acquisition report uses Session default channel grouping as its default dimension, so you can see the AI Assistant row without extra setup. Check three values in that row:

• Sessions and share of total — switch between the last 28 days and the last quarter, and record the trend.
• Engagement rate — compare it with other channels to judge whether AI visitors actually read your pages.
• Key events — check whether conversion actions such as inquiries and sign-ups happen in AI traffic too.

If no sessions were classified into the channel in the selected period, the row doesn't appear at all. Even then, don't conclude there is no AI traffic — move on to Step 2.

Step 2: Cross-check AI services with Session source / medium

Switch the dimension to Session source / medium and search for AI service values; if their total is larger than the AI Assistant row, the difference is AI traffic counted in other channels.

In the Traffic acquisition report, choose "Session source / medium" from the dimension selector, then type each value below into the table search box.

AI service	Values to check in session source
ChatGPT	chatgpt.com, chat.openai.com
Perplexity	perplexity.ai, perplexity
Gemini	gemini.google.com, gemini
Claude	claude.ai
Copilot	copilot.com, copilot.microsoft.com
Deepseek, Grok and others	Services in Google's example list — confirm with the values that appear in your report

ChatGPT leaves a UTM value independent of the referrer. OpenAI's Publishers and Developers FAQ says ChatGPT automatically adds the utm_source=chatgpt.com UTM parameter to referral URLs (OpenAI Help Center). So if you see a row where the session source is chatgpt.com but the medium is not ai-assistant, check which channel those sessions were counted in.

Keep the result in one table: total sessions per AI service, sessions in the AI Assistant row, the difference, and the medium values where the difference was counted (referral, (not set), (none) and so on). That becomes your baseline for next month's comparison.

Step 3: Watch landing pages in Direct traffic

Because links opened in AI apps can be counted as Direct, use growth in Direct traffic to detail pages and comparison pages, rather than the homepage, as a supporting signal.

People rarely type the long URL of a blog post or comparison page by hand. If sessions landing on detail pages without a referrer are rising, some may be visits from links opened in apps or messengers. In an Exploration, place the Landing page dimension next to Session default channel grouping, filter to Direct, and compare month over month.

This signal doesn't prove AI traffic, because messenger shares, bookmarks and email apps are recorded the same way. Treat it as stronger evidence only when Direct traffic rises on the same pages that show up as landing pages in the AI Assistant row over the same period.

Step 4: Build a custom channel group to compare with pre-May data

To compare AI traffic with the period before May 13, 2026, or to include AI services outside the list, build a custom channel group, which applies to past data retroactively.

Custom channel groups are created in Admin, under Data display, Channel groups. Standard properties can have up to two in addition to the predefined group.

"Custom channel groups can be applied to your reports retroactively." — Google Analytics Help: Custom channel groups

1. On the Channel groups page, copy the default channel group to create a new group.
2. Add a channel and give it a name such as "AI traffic (all)".
3. Set the condition to Source matches regex and enter the expression below.
4. Move this channel above Organic Search and Referral. GA4 includes traffic in the first channel whose definition it matches, given the current order.
5. After applying it, check the session source list for values you didn't intend to include.

.*(chatgpt|openai|perplexity|gemini|claude|copilot|deepseek|grok).*

The expression is a partial match, so other sites with the same word in their name (for example, a blog with "copilot" in its domain) can be included, which is why step 5 matters. Google's help also says to update your regex when URLs or the list of assistants you want to measure change. A custom channel group only regroups source values that were already recorded, so app visits recorded as Direct without a referrer still can't be separated this way.

AI traffic numbers alone don't show AI visibility

GA4 only counts visits from people who clicked after seeing an AI answer, so exposure where your brand was mentioned but nobody clicked doesn't show up in AI traffic numbers.

In a case where one site tested opting out of Google's AI features, the estimated click-through rate for AI feature impressions was about 0.9% (Search Engine Roundtable, a single-site case). AI answers reduce the reason to click, so low traffic doesn't mean no exposure; and when traffic rises, GA4 still can't tell you which questions' answers you appeared in. You can check the scale of exposure in Google's AI features with Search Console's generative AI reports.

Whether ChatGPT, Perplexity and Gemini mention your brand requires measurement at the question level. RanketAI's AI Brand Visibility Analysis measures brand mentions in ChatGPT, Perplexity and Gemini answers, and when you connect a GA4 property under Integrations (Starter plan and above), the dashboard's AI Inflow card shows AI sessions by service, landing pages and conversion actions. The two numbers are meant to be read side by side for the same period; they don't imply cause and effect. Page Structure Diagnostics checks whether your site is structured for AI to read, with no sign-in required.

Limits of this guide

The classification behavior in this guide reflects documents and sources checked on November 9, 2026, and how GA4 and AI services handle links can change without notice.

• The full AI Assistant list isn't published. Whether a service outside the examples is included can only be confirmed in your own data.
• Referrer behavior for apps and AI Mode comes from industry sources. We cite Ahrefs (2025) and SearchPilot (2026); RanketAI did not reproduce them.
• The own check covers one site. How often name-only sources appear, and in what format, can differ by site.
• OpenAI can change how it adds UTM parameters. If session source values drop suddenly, check for a matching rise in Direct traffic.

FAQ

Related reading

• Why AI Search Traffic Converts Higher — measuring GEO ROI and attribution
• ChatGPT's "Not Provided" Moment — ads-only conversion data and the organic measurement gap
• Google Search Console's New Generative AI Reports — checking exposure in Google's AI features with an official tool
• AI Search Answers Without Referring — why outbound clicks from AI search are falling
• Microsoft Clarity Citations Dashboard — free AI citation measurement