How to Track AI Traffic in GA4 and Search Console

AI assistants like ChatGPT, Gemini and Perplexity already send real visitors to your website, but Google Analytics will not label most of that traffic for you. This guide shows how to track AI traffic in GA4 with a custom channel group and a regex, why so much of it hides as Direct, and what Google Search Console can and cannot show about AI Overviews and AI Mode.

I check this exact setup on every client account I run, because a Pune clinic or shop that shows up inside a ChatGPT answer or an AI Overview needs to know whether that visibility is turning into an actual visit, not just an impression. This guide covers the GA4 side first, then Search Console, then what I actually put in a monthly report.

What counts as AI traffic, and why GA4 does not label it for you automatically

AI traffic, for this guide, means a session that started because someone clicked a link inside an AI assistant's answer, a citation inside ChatGPT, Gemini, Perplexity or Copilot, or a link inside Google's AI Overview or AI Mode on the results page. It does not include someone who read a summary and never clicked through. That visit never reaches your website at all, so GA4 has nothing to record either way.

GA4's reporting was built around channels like Organic Search, Paid Search and Referral, categories that existed long before AI assistants sent anyone anywhere. On 13 May 2026, Google Analytics added a native AI Assistant channel to the Default Channel Group, tagging matching sessions with a medium of ai-assistant, according to the Google Analytics release notes. That is a genuine improvement, but Google has not published a complete, fixed list of every source it recognises, and its own announcement names ChatGPT, Gemini and Claude only as examples of "chatbots like" these. At the time of writing (September 2026), several independent trackers report that Perplexity in particular still lands outside the native channel. That gap is exactly why the custom channel group later in this guide is worth building even though the native one now exists, and none of it answers whether you are actually earning citations in the first place, which is a separate question I cover in my guide to ranking in Google AI Mode.

Where AI assistant traffic already shows up in GA4 today

Start in Reports, Acquisition, Traffic acquisition, GA4's report for seeing which channel or source sent a session. Set the date range to the last 28 days, then look at the table's first column, which defaults to Session default channel group. If the native AI Assistant channel has picked up any sessions, it appears as its own row here, sitting alongside Organic Search, Direct and Referral.

The more useful view for this kind of check swaps that default dimension. Click the dimension dropdown above the table and change it from Session default channel group to Session source, or to Session source / medium if you want the referrer and the medium together in one row. This is where individual domains show up by name: chatgpt.com, perplexity.ai, copilot.microsoft.com, gemini.google.com, claude.ai and any other AI product a visitor arrived from, each on its own line with its own session count.

Filter that table to a single AI source by typing part of the domain into the search box above it, then check the key events column next to it. A handful of sessions from perplexity.ai that produced a WhatsApp click or a form fill tells you more than a much larger number from a channel that converts at zero. If key events are not marked yet, my Google Analytics for beginners guide covers setting up WhatsApp, call and form key events before any of this traffic-source work is worth doing.

Why a large share of AI traffic still hides inside Direct

Even with the native channel and a Session source view, a real share of AI-referred visits will never show up tagged as AI anything. When an AI assistant's mobile app or in-app browser opens a link, it frequently strips the referrer header before handing the visitor to your site, so GA4 has no source to read and files the session under Direct instead, next to people who typed your URL straight into the address bar. Google's own channel definitions confirm that Direct requires an empty or missing referrer alongside an unset medium, which is exactly the gap many AI apps leave behind; see Google's default channel group definitions for the full rule set.

This is not a GA4 bug, and there is no setting that fixes it from your side. The browser or app the visitor used simply did not send the information GA4 needs. What you can do is watch Direct traffic for a shape that looks unusual for your business: a service page that never used to get direct visits suddenly getting them daily, or a jump in Direct sessions the same week your impressions or citations rose elsewhere. Neither proves the source, but both are worth a note in your monthly report.

One AI product helps here rather than hurting. ChatGPT's search citations, at the time of writing (September 2026), commonly carry a utm_source=chatgpt.com parameter, so a visit that arrives with that parameter in the address bar is confirmed AI traffic in GA4's Session source or campaign fields, whatever the referrer header says. This is not guaranteed on every plan or every kind of ChatGPT answer, so check your own site's incoming URLs for the parameter rather than assuming it applies. If you route tags through Google Tag Manager rather than pasting GA4 directly onto the page, my Google Tag Manager for beginners guide covers checking that a parameter like this survives the trip. My UTM parameters guide covers building and reading tags like this one from the other direction, for your own outbound links.

How to build a custom AI assistants channel group in GA4

A channel group you control is the only way to catch every AI source in one place, old and new, without waiting for Google to add each one. Building it takes about ten minutes and does not touch any historical source data; GA4 simply starts applying the new rule going forward and can reclassify existing sessions wherever the underlying source is still there. Google's own instructions for building a custom channel group cover every dimension and match type available if you want to go beyond a source-based rule.

Step 1: Open Channel Groups in GA4's Admin settings

In the property column of Admin, open Data display, then Channel groups. This is where GA4 keeps the default channel group and any custom ones you add. Standard properties can hold two custom groups alongside the default, so this will not be your only chance to build one later.

Step 2: Create a new channel group and add a channel called AI assistants

Click Create new channel group, which starts as a copy of the default group so every existing channel keeps working. Click Add new channel, name it AI assistants, and leave it empty for the next step.

Step 3: Set the condition to Session source matches regex

Inside the new channel, click Add condition group, choose Source as the dimension, and set the match type to matches regex. Paste in a regex covering the AI assistant domains you want to catch, escaping every dot so it does not match unrelated text.

Source matches regex:
chatgpt\.com|chat\.openai\.com|perplexity\.ai|copilot\.microsoft\.com|gemini\.google\.com|claude\.ai|meta\.ai|grok\.com|chat\.deepseek\.com|you\.com

Treat that pattern as a starting point rather than a finished list. Add a new domain with a vertical bar and an escaped dot whenever you spot one in your Session source report that this pattern does not already cover.

Step 4: Move the AI assistants channel above Direct and Unassigned

Drag the new channel above any channel that could otherwise catch the same sessions first, particularly Direct, Unassigned and Referral, since GA4 applies channel rules top to bottom and stops at the first match. Leave the built-in AI Assistant channel below your custom one if you want the broader regex to take priority, or above it if you would rather trust Google's native match first.

Step 5: Save and apply the channel group in Traffic acquisition

Click Save, then open Reports, Acquisition, Traffic acquisition and select your new grouping from the dimension picker in place of Session default channel group. New sessions start classifying immediately, and historical sessions reclassify retroactively wherever the underlying source data still exists.

Step 6: Recheck the regex every few months and add new assistants

AI products launch and rebrand often enough that a list frozen in September 2026 will be incomplete by early next year. Add newly discovered domains to the same regex rather than building a second channel, and note the update date somewhere you will actually see it. A comment in the sheet you report from works fine.

A five-minute Exploration that shows AI sources without waiting on the channel group

If you want a same-day answer instead of a new channel group, GA4's Explore section gets there faster. Open Explore in the left-hand navigation, start a blank exploration, and add Session source as a dimension and Sessions plus Key events as metrics.

Drag Session source into Rows, sort by Sessions descending, and scan the list for domain names you recognise, such as chatgpt.com, perplexity.ai or gemini.google.com. Add a filter for the date range you care about, and if you already run the channel group from the previous section, add Session default channel group as a second dimension so you can see the raw domain and the channel it landed in, side by side.

This view is disposable by design. Nothing you build in Explore changes how future sessions get classified, so use it to sanity-check the channel group's regex, spot a new AI domain before you add it to the rule, or answer a one-off question for a client without touching the property's settings at all.

What Google Search Console actually shows about AI Overviews and AI Mode traffic

Search Console is the other half of this picture, and it works differently enough from GA4 that treating the two the same way leads to wrong conclusions. If you have not set up Search Console yet, my Google Search Console for beginners guide covers verification and the Performance report before you need any of what follows here.

Google is explicit about this in its own documentation. Its AI features and your website page states that no special markup is needed to appear in AI Overviews or AI Mode, that ordinary SEO fundamentals apply, and that sites appearing in these features are included in the overall search traffic in Search Console, reported inside the Performance report's Web search type. In plain terms, a click from an AI Overview and a click from an ordinary blue link both land in the same Web totals, and the main report does not let you split them apart.

Position works the same way. Google treats an AI Overview as a single block on the results page, the way it already treats several other search features, so every link cited inside one AI Overview for a given query reports the same position value, whichever citation a person actually clicked. A page cited first in the block and a page cited fifth look identical in the Performance report's Average position column.

There is one genuine improvement, recent enough to be worth flagging by date. Google rolled out a dedicated Generative AI performance report inside Search Console, announced in June 2026 and made available to every property worldwide since 31 August 2026, according to Search Console Help. Two caveats sit behind that rollout date: a property only sees the report once it has received enough impressions in AI features to populate one, and the same 1,000-row limit that caps the ordinary Performance report applies here too. Where you do have access, it shows impressions, meaning how often your pages appeared inside AI Overviews or AI Mode, broken down by page, country, device and date. At the time of writing (September 2026), the report does not include clicks, click-through rate, average position or a breakdown by the actual query that triggered an appearance, and Search Console Help does not commit to a date for adding them. Treat it as a visibility check, not a performance report.

Three workarounds for the gaps Search Console leaves open

None of the three workarounds below recovers exact numbers Google does not publish, but each one gives you a usable signal instead of a shrug.

Search Console's Queries tab accepts a regex filter, so you can isolate the long, conversational searches typical of AI Mode, the kind nobody types into a normal search box. A filter such as the one below on the query dimension surfaces long question-style searches. Skim the list monthly rather than trusting it as a precise AI Mode count, since plenty of long questions trigger an ordinary results page too.

Query matches regex:
^(what|how|why|can|should|is|does).{40,}

Watch for the pattern that shows up when a query starts triggering an AI Overview: impressions on that query hold steady or keep climbing in the Web performance report while clicks and click-through rate slide, even though average position has not moved. That combination, unchanged or rising impressions against falling clicks on an unchanged position, is the closest free signal Search Console gives you that an AI feature has started absorbing the click. It is also, in plain language, what a zero-click search looks like from inside your own property. My guide to how AI Overviews affect SEO goes through this pattern by query type in more depth.

Keep a simple manual log. Once a week, ask ChatGPT, Gemini, Perplexity and Google's AI Mode the handful of questions your actual customers would ask, the way you would check a rank tracker, and note whether your business gets mentioned, linked or left out entirely. There is no dashboard for this on any platform at the time of writing, so a spreadsheet with the date, the prompt, the assistant and what came back is the only record you will have. A dedicated AEO monitoring tool can automate parts of this; my AEO tools guide covers what is actually worth paying for versus doing by hand.

What to report to a client every month

Strip this down to five lines a client can actually read, the same instinct behind the ten-row sheet in my SEO KPIs guide. Report GA4 sessions and key events from the AI Assistant channel and your own custom channel, side by side so any gap between them is visible. Report Direct sessions and note whether the trend looks unusual, without claiming certainty about the cause. Report impressions from the Generative AI performance report next to impressions from the main Web performance report, so growth in one against flat numbers in the other tells its own story. Report the falling-clicks, flat-position pattern on any money query where you have spotted it. Finally, report anything new from the manual prompt log, in plain words, not a score.

Resist the pressure to invent a number Google has not published, a made-up AI visibility score or a percentage of traffic you cannot actually verify. Google's own guide to optimising for AI features warns site owners to be wary of any third-party tool that claims to use Google's internal ranking or AI data, since no outside tool actually has access to it. My guide to measuring digital marketing ROI covers the wider version of this same discipline: report what the platform actually gives you, and say plainly when something cannot be measured yet instead of filling the gap with a guess.

Honest limits: what none of this can tell you

Being honest about the ceiling here matters more than the setup itself. None of this proves your page ranks anywhere inside an AI answer, because there is no rank 1 in AI Mode or inside any AI-generated answer the way there is in a list of ten blue links. The realistic goal is being cited, linked or mentioned, and even a confirmed citation today is no guarantee of the same citation next week, since these answers regenerate per query rather than sitting in a fixed position.

GA4 and Search Console together also cannot separate an AI-referred visitor who was already planning to buy from one who clicked out of curiosity. A WhatsApp click from an AI Assistant channel session still needs the same follow-up and the same honest look at close rate as a lead from anywhere else, covered in my guide to tracking where leads actually come from. If you are weighing whether any of this search-side work still pays off at all, does SEO still work in 2026 answers that wider question with numbers rather than a hunch. Treat every number in this guide as directional, useful for spotting a trend, not precise enough to build a single decision on by itself.

Common mistakes that quietly break AI traffic tracking

I see the same handful of mistakes on almost every account that has tried to track AI traffic on its own.

• Building the channel group once and never adding new AI domains, so a product that launches in November shows up as Unassigned in December.
• Assuming every Direct session is AI traffic. Most Direct sessions are still bookmarks, typed URLs and stripped links from ordinary apps, not AI assistants.
• Ordering the custom channel below Direct or Unassigned in the channel list, so the broader default rule catches the session first and the AI assistants channel never fires.
• Reporting the Generative AI performance report's impressions as if they were clicks, then wondering why the lead numbers do not match.
• Treating a single AI citation as proof of a ranking, when the honest description is a citation that can disappear on the next regeneration of the same answer.

Most of these take minutes to fix once you know to look for them. The setup in this guide will not give you a perfect number, Google has not published enough for anyone to have that, but it replaces a guess with a trend you can actually defend in a monthly report.

Frequently asked questions

Does GA4 have a built-in way to track ChatGPT or Gemini traffic?

Yes, since 13 May 2026 Google Analytics has shipped a native AI Assistant channel inside the Default Channel Group, tagging matching sessions with a medium of ai-assistant. Google's own announcement names ChatGPT, Gemini and Claude only as examples, not a complete list, and several independent trackers report gaps for other AI products at the time of writing. A custom channel group with your own regex still catches more than the native channel alone, and the two can run side by side without conflict.

Why does AI assistant traffic show up as Direct in Google Analytics?

Because many AI apps and in-app browsers strip the referrer header before handing a visitor to your site, and GA4 can only classify a session using the data the browser actually sends it. With no referrer and no UTM parameter, GA4 has nothing to match against any channel rule, so the session falls into Direct alongside typed URLs and bookmarks. There is no setting on your side that fixes this, though a sudden rise in Direct traffic on a page you know gets AI citations is worth noting in a monthly report.

What regex should I use to catch AI assistants in a GA4 channel group?

Start with a Session source regex covering the major assistants: chatgpt.com, chat.openai.com, perplexity.ai, copilot.microsoft.com, gemini.google.com, claude.ai and similar domains, joined with the vertical bar operator and escaped dots. Treat that list as a starting point, not a finished one. New AI products launch often enough that revisiting the regex every few months and adding whatever is missing matters more than getting the first version perfect.

Can Search Console show me clicks from AI Overviews separately from normal search clicks?

No, not at the time of writing (September 2026). Google's own documentation states that AI Overview and AI Mode traffic is included in the regular Web search totals inside the Performance report and is not broken out separately. Search Console's newer Generative AI performance report shows impressions only, broken down by page, country and device, with no clicks, click-through rate, position or query data included yet, and Google has not committed to a date for adding them.

Does getting cited in an AI Overview mean I am ranking number 1?

No, and that framing does not really apply here. There is no rank 1 inside an AI Overview or any AI-generated answer the way there is in a list of ten blue links, since every link Google cites inside the same AI Overview reports an identical position value in Search Console. The realistic goal is being cited, linked or mentioned at all, and even that can change from one regeneration of the answer to the next, so treat a citation as a good sign, not a fixed ranking.

Related guides

• Zero-click searches: meaning, 2026 data and what to do
• Google Analytics for beginners: GA4 setup guide
• Google Search Console for beginners
• How Google AI Overviews impact SEO
• UTM parameters: meaning, examples and a naming system
• SEO KPIs to track
• How to rank in Google AI Mode
• How to measure digital marketing ROI

Your next step

Build the GA4 channel group this week using the regex above, then check back in a month once real sessions have flowed through it. If AI Overviews are already eating into your click-through rate on specific queries, read zero-click searches: meaning, 2026 data and what to do next for the wider strategy, or get in touch if you want a second pair of eyes on your account.