How do you track AI Overview and ChatGPT traffic in GA4?

By Andrej Ruckij · November 1, 2026

TL;DR: Read every AI traffic number in GA4 as a floor, because AI Overview clicks aren’t in it at all. Clicks from Google’s AI Overviews and AI Mode arrive with a Google referrer and are counted as Organic Search, and Google’s own channel definition excludes them from the AI channel. What you can see: since May 13, 2026, GA4 has a built-in AI Assistant channel for visits from ChatGPT, Gemini, Deepseek, Copilot and Grok. Perplexity and Claude aren’t on Google’s published list, so add a custom channel group with a regex for them and place it above Referral and Organic Search. For AI Overviews and AI Mode, use Search Console’s Generative AI performance report, which shows impressions only (data from May 18, 2026, per Neil Patel’s write-up), and estimate the clicks with the CTR studies: Pew measured 8% clicks on pages with an AI summary against 15% without. Check the GA4 numbers against your server logs.

Primores helps brands get found and recommended in AI answers: see how we work with brands.

Who this is for

For: a marketer or site owner with GA4 and Search Console access who wants a monthly number for “traffic from AI”, split by assistant, and a defensible estimate of what Google’s AI features do to organic clicks. You need the Editor role on the GA4 property to build channel groups.

Not for: apps without a website. For scale, Ahrefs’ 2025 study of 3,000 sites found AI chatbots sent 0.17% of the average site’s visitors, so expect small numbers.

How do I check AI Overview traffic in GA4?

The short answer is that GA4 can’t separate it. Google’s help page for the default channel group says the AI Assistant channel covers visits “from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok” and excludes Google’s AI Overviews and AI Mode. A click from an AI Overview carries a Google referrer, so it lands in Organic Search with every other Google click.

Google confirms the same from the other side. Its documentation for site owners says AI features “are included in the overall search traffic in Search Console” and “they’re reported on in the Performance report, within the “Web” search type.” AI Mode has counted towards those Search Console totals since June 2025, according to PPC Land’s report of Google’s help page, with no filter to see AI Mode alone.

So the honest setup has three layers, and each sees a different slice:

Where	What it shows	What it misses	Cost
GA4 default channel group, AI Assistant channel	Sessions and conversions from ChatGPT, Gemini, Deepseek, Copilot and Grok when a referrer or a recognised tag arrives	AI Overviews and AI Mode (in Organic Search), app visits with no referrer (in Direct), assistants not on Google’s list	Free
GA4 custom channel group with your own regex	Everything above plus Perplexity, Claude and any assistant you add, applied to past data too	Visits with no referrer or tag	Free
Search Console, Generative AI performance report	Impressions in AI Overviews and AI Mode, by page, country, device and date	Clicks, CTR, position and queries	Free
Server logs	Pages that ChatGPT-User and Perplexity-User fetch when someone asks a question	Whether a person clicked through	Free if you have log access
Paid AI-visibility trackers (e.g. Peec AI)	Whether you’re cited for a fixed list of prompts	Real traffic	From $95/month for 50 prompts (Peec AI, as listed)

What does GA4’s AI Assistant channel actually catch?

Google added it on May 13, 2026. The “What’s new” entry says GA4 “now provides a dedicated way to measure and analyze traffic originating from popular AI assistants.” It needs no setup. Sessions it recognises get the medium ai-assistant and the campaign (ai-assistant).

Three things to know before you trust it:

The list is short and unpublished. Google’s channel definition says the rule fires when “the referrer matches a list of AI Assistants” but doesn’t publish the list. The page names ChatGPT, Gemini, Deepseek, Copilot and Grok. Digital Applied’s June 2026 check of the documentation found Perplexity absent and still landing in Referral, and marked Claude “Not in current list”. A newer Source Group dimension (June 11, 2026) groups “emerging traffic sources like ChatGPT (OpenAI) and Perplexity”, but Swapnil Biswas, who checked Google’s wording on October 10, 2026, couldn’t find it in the dimensions reference and doesn’t recommend relying on it yet.
It isn’t retroactive. Sessions before May 13, 2026 stay in Referral and Direct. A custom channel group, by contrast, “can be applied to your reports retroactively”, per Google’s help page.
A tag can push a visit out of the channel. A-Track’s analysis points out that GA4 only fills the source from the referrer “when no other campaign or traffic source fields have been set.” A link with utm_source but no medium leaves the medium empty, and since AI Assistant membership depends on the medium, “the session can end up outside the channel, in Unassigned or in an unexpected channel.”

That last point matters for Gemini right now. Search Engine Journal reported on October 1, 2026 that Google had quietly added undocumented UTM parameters to some outgoing Gemini links. SEJ didn’t give the values; Vizup, an AI-visibility vendor, reports them as utm_source=gemini and utm_medium=organic. If that’s right, those visits match Google’s Organic Search rule (“Medium exactly matches organic”) and never reach the AI channel. Check your own landing URLs before you assume either way.

How to do it

In GA4: build an AI channel you control (about 20 minutes)

See what’s already arriving. Go to Explore, start a Free form exploration, put Session source in rows and Session medium in columns, and filter Session source to contain chatgpt, perplexity, gemini, claude, copilot, deepseek or grok, one at a time. Write down the exact source values you see. Then add Landing page + query string as a row, filtered to contain utm_source, to list which tags the assistants append. This is A-Track’s 20-minute check, and it tells you what your regex has to match.
Create the channel group. Go to Admin → Data display → Channel groups and click Create new channel group. That starts as a copy of the default group, so you keep every standard channel. Standard properties allow 2 custom groups, each with up to 50 channels.
Add the AI channel. Click Add new channel, name it AI assistants, click + Add condition group, choose the Source dimension, choose matches regex, and paste:
```
^(chatgpt\.com|chat\.openai\.com|gemini\.google\.com|gemini|copilot\.microsoft\.com|claude\.ai|(www\.)?perplexity\.ai|chat\.deepseek\.com|grok\.com)$
```
This is Swapnil Biswas’s anchored pattern, which he tested against twelve hostnames: it matched all eight assistants and none of the four non-assistant values. I added gemini on its own because that’s the tag value Vizup reports on Gemini links. Add any other exact source values you found in step 1.
Put it above Referral and Organic Search. Google’s rule is that “traffic is included in the first channel whose definition it matches given the current order of channels in the group.” Drag AI assistants above Referral, and above Organic Search too, or Gemini’s organic-tagged visits will be claimed by Organic Search first. Save. One note on Google’s help page says channel definitions can’t be edited after they’re saved, so check the regex before you do.
Read it. Go to Reports → Acquisition → Traffic acquisition, open the primary dimension dropdown and pick your new group. Because custom groups apply retroactively, set the date range back past May 13, 2026 and you’ll see AI visits that the default group filed under Referral. Compare sessions, engaged sessions and key events against Organic Search.

Don’t copy Google’s own example regex. The example on Google’s custom channel group page starts with ^.*ai|, which Biswas shows matches any source ending in “ai” under GA4’s full matching, and any source containing “ai” under partial matching such as BigQuery’s REGEXP_CONTAINS. He found 18 matches in Google’s 819-source list, including webmail hosts such as email.seznam.cz, because “ai” sits inside “mail”. It also misses DeepSeek and Grok.

Check it worked: in the exploration from step 1, switch the dimension to your custom channel group. Every source value you listed should sit in AI assistants, and nothing else (no mail hosts, no gov.ai) should.

In Search Console: see AI Overviews and AI Mode (about 10 minutes)

Open Search Console, go to Performance, and pick Generative AI performance (Search), or go straight to search.google.com/search-console/performance/search-analytics/ai. Google says it was rolled out “to all websites worldwide” as of August 31, 2026. It won’t appear if your site has too few AI impressions.
Read the impressions by Pages. Google defines an impression here as “how many times links to your site were shown to a user in a generative AI feature on Google Search.” Several links to your site in one answer count once. Neil Patel’s write-up says the data starts on May 18, 2026, with no backfill, and that the report has no clicks and no query data.
Export the page list. Then open the regular Performance → Search results report and compare the same pages’ clicks and CTR. Don’t add the two reports together: Google says the AI report “includes data from the Web search type in the Performance report”, so the impressions are already counted in the main report.

In your server logs: the cross-check (about 30 minutes the first time)

GA4 depends on JavaScript, consent and a referrer. Logs depend on none of those, but they show fetches, not people.

Filter your access logs for the user agents ChatGPT-User and Perplexity-User. OpenAI says that “when users ask ChatGPT or a CustomGPT a question, it may visit a web page with a ChatGPT-User agent”, and Perplexity says Perplexity-User “supports user actions within Perplexity.” So these hits mean your page was read to answer someone’s question.
Check the requests come from the published IP ranges: openai.com/chatgpt-user.json and perplexity.ai/perplexity-user.json. Anyone can fake a user agent; more in How to verify AI bots.
Count fetches per page per month and put them next to GA4’s AI-assistant sessions for the same pages. Pages with many fetches and few sessions are being used in answers without sending visitors. Keep GPTBot out of this count: OpenAI describes it as the training crawler, which says nothing about answers.

What can’t be tracked, as of November 2026?

AI Overview and AI Mode clicks. They sit inside Organic Search in GA4 and inside the Web search type in Search Console. The Generative AI report gives impressions only. I found no official way to count clicks from AI features separately.
App visits with no referrer. A-Track notes that from an assistant’s “mobile app, desktop app, extension, or an integration inside another piece of software” the referrer “is often absent or stripped”, so the visit is Direct. I found no independent measurement of how many AI visits end up there.
ChatGPT without its tag. ChatGPT adds utm_source=chatgpt.com to links in its answers, and that tag does most of the work: per The State of Brand, Adobe’s Cedric Huesler found that most ChatGPT clicks still arrive without a referrer. When the tag went missing for about a week after GPT-6 launched, ChatGPT sessions fell sharply in GA4 and Direct rose.
Gemini, inconsistently. Some Gemini links now carry tags and some don’t, and Google hasn’t documented which.
Anything an assistant said without a link. A mention that leads someone to search your brand later shows up as branded organic search or Direct, if it shows up at all.

Why did my AI traffic drop in October 2026?

If your ChatGPT sessions fell around October 9, it was probably a labelling problem. LLM Pulse, which sells AI-search tracking, reported ChatGPT sessions down about 46% between October 9 and 11 across the sites it monitors, while total traffic on those sites stayed flat (around −2%). Huesler’s sample of about 500 pages where, per The State of Brand, ChatGPT normally drove at least 5% of visits, lost about 31,000 ChatGPT visits a day while Direct gained about 26,000. The cause reported in each case was links that lost utm_source=chatgpt.com and opened with no referrer. Huesler reportedly said OpenAI confirmed the problem to him, but OpenAI has published no explanation; a Siecle Digital report, cited by ALM Corp, says the tag returned on October 15.

GA4 records the source at the time of the visit, so those sessions won’t be corrected. Add an annotation for October 8 to 15, and for that week read ChatGPT and Direct together. The same check works for any future drop: if total sessions held steady and only one AI source fell, look at tagging before you look at your content.

How much traffic do AI Overviews take? What the impact studies show

You can’t measure your own AI Overview clicks, so the studies are the next best thing. They agree on direction and disagree on size.

Study	Data	Finding
Pew Research Center (July 2025)	900 US adults’ real browsing, 68,879 Google searches, March 2025	Clicks on a traditional result on 8% of visits with an AI summary, 15% without. Links inside the summary clicked on 1% of visits.
Seer Interactive (November 2025, agency)	3,119 search terms, 42 client organisations, 25.1M organic impressions, June 2024 to September 2025	Organic CTR on AI Overview queries fell 61%, from 1.76% to 0.61%. Brands cited in the Overview got 35% more organic clicks than those not cited.
Ahrefs (February 2026, vendor)	300,000 keywords, desktop CTR from Search Console, December 2023 vs December 2025	An AI Overview went with a 58% lower average click-through rate for the top-ranking page, up from 34.5% in its April 2025 study.

Two caveats. Seer’s non-AI Overview queries lost 41% of their CTR over the same period, so not all of the drop is the Overview. And Google disputes the framing: Liz Reid, head of Google Search, wrote in August 2025 that “total organic click volume from Google Search to websites has been relatively stable year-over-year” and that clicks from AI Overview pages are “higher quality”, meaning people don’t quickly click back. She gave no figures. Pew’s study is the one I’d lean on, because it’s independent and based on what people actually did; more on which numbers are measured and which are vendor estimates in The SEO stats everyone quotes.

To estimate your own exposure: take the pages with the most impressions in the Generative AI report, and in the main Performance report compare their CTR for May to October 2026 with the same months in 2024, before AI Overviews were widespread. Pew found AI summaries on 60% of question-style searches and only 8% of one- or two-word searches, so filter queries to those starting with how, what, why or which and compare them with short queries. If CTR fell much more on the question queries, that’s the Overview effect on your site, give or take seasonality.

How should I read the numbers?

My read: the AI traffic figure in GA4 is a floor, and on most sites a small one. Ahrefs’ 0.17% average is from early 2025, so treat it as dated, but the gap between how often you’re shown and how often you’re visited is the real story. Seer’s data says being cited in an AI Overview is worth about a third more organic clicks than being left out of it, and Pew says almost nobody clicks the links inside the summary. Seer’s figure is a correlation, not proof that citation causes the clicks, but both studies point the same way: on Google, being cited goes with keeping more of the clicks, and the summary itself sends very few.

So I’d report three numbers each month, side by side, and never sum them:

AI assistant sessions and key events from the custom channel group, with Direct next to it, since that’s where untagged AI visits go.
AI Overview and AI Mode impressions from Search Console, by page.
ChatGPT-User and Perplexity-User fetches from the logs, by page.

Rising impressions and fetches with flat sessions means AI answers are using your pages, which is where the value is moving, and why tracking what AI says about you matters as much as the visits. If you want to know why an assistant mentions you or doesn’t, start with How can ChatGPT recommend my business? and, for Google’s side, Can you trust Google’s AI Overviews?

Sources

More of this every day on the channel: https://t.me/aidaily_business