AI traffic in Google Analytics: how to find and track it in GA4

If a customer clicks a link in ChatGPT and lands on your website, Google Analytics 4 (GA4) can show you that visit in a row called AI Assistant.¹ That only works when GA4 can tell where the visitor came from. Think of GA4 as a door counter in your shop. It records the people who walk in and where they came from. It can’t count anyone who only read your name and never came through the door. This guide shows where to look, how to build your own report if you want one and what the numbers can’t tell you.

Our AI SEO agency can help connect those recorded visits to enquiries. Checking how your business appears in generated answers is a separate task, explained in what is GEO. An answer that names you without a click never reaches your analytics, so AI visibility needs its own check.

Here’s an example. Say you run an electrical business in Newtown. A homeowner asks ChatGPT for a local electrician, clicks the link to your website and sends an enquiry. In GA4 that visit can land in the AI Assistant row. A second homeowner reads your name in the same answer but doesn’t click. They search for you on Google later and visit that way, so GA4 would most likely file the visit under Organic Search with no AI label. Both people came because of the answer. Only one of them shows up as AI traffic.

What does this mean for your business?

You can start with what GA4 already gives you and build nothing. Here are five steps in the order we’d take them.

1. Open Reports, then Acquisition, then Traffic acquisition. Look for the AI Assistant row and add Session source / medium so you can see which assistant sent each visit.
2. If the list is short, leave the built-in channel as it is. Build a custom channel group only when you want your own list of sources.
3. Count enquiries with a form-submission key event, which is GA4’s name for an action you mark as important. Don’t count clicks on a contact button.
4. Treat the AI count as a floor, because some AI visits show up as Direct or go missing.
5. Keep mentions apart from visits. Google’s own AI features are reported in Search Console. GA4 doesn’t count an answer nobody clicks, so check those mentions separately with saved answers.

Where do you find AI traffic in Google Analytics?

Four words to know. A source is the website or app a visit came from. A medium is the type of traffic, such as organic or referral. A channel group is a set of rules that sorts visits into named rows. A referrer is the page address a browser passes along when someone clicks a link.

Open Reports, then Acquisition, then Traffic acquisition. Look for a row called AI Assistant. That row holds visits from assistants that GA4 recognises.⁵¹ The report sorts visits into rows by a column called session default channel group.

1. Sign in to Google Analytics and select Reports in the left menu.
2. Select Acquisition, then Traffic acquisition.⁵
3. Set the date range to cover at least four full weeks.
4. Find AI Assistant in the first column.
5. Click the + next to the column heading and add Session source / medium.⁵

Step 5 shows which assistant sent each visit. Google defines the session source as the site or app the traffic came from, so ChatGPT visits appear under the source name the assistant passes.⁵ Your own rows decide what you build later, so write down every source name you see.

If you can’t see an AI Assistant row, there may be no recognised visits in that date range. If the whole Traffic acquisition report is missing, Google says it may have been removed from the navigation. An Editor or Administrator can add it back.⁵

What does the AI Assistant channel count?

It counts visits from sources such as ChatGPT, Gemini, DeepSeek, Copilot and Grok. It leaves out Google’s AI Overviews and AI Mode.¹ Google’s list is introduced with the word “like”, so it’s a sample and not a full roll call.

A release note dated 13 May 2026 describes the change. Visits whose referrer matches a recognised assistant get the medium “ai-assistant” and the campaign “(ai-assistant)”.² That release note doesn’t say whether older data was reclassified, so don’t assume your earlier months were.²

The table shows where five common kinds of visit end up.

The first two rows follow Google’s channel definitions.¹ The Direct row follows its definition of Direct traffic.⁶ The last two rows matter because both are real effects of an AI answer, yet neither will carry an AI label.

The picture below stacks three of those routes one above the other. A recognised assistant click goes to AI Assistant. A click from Google’s AI Overviews or AI Mode goes to Organic Search. A click with no usable referrer probably lands in Direct. The note at the bottom is the one to remember, because Direct doesn’t prove an AI source.

Where a recorded visit goes

Recognised assistant click

↓

AI Assistant

Google’s channel definition

Google AI Overview or AI Mode click

↓

Organic Search

Google’s channel definition

No usable referrer

↓

Likely Direct or another channel

Inference from the Direct definition

Direct does not prove an AI source.

Why does some AI traffic show up as Direct or go missing?

GA4 can only label a visit it can attribute. Google describes Direct traffic as visits without a clear referral source.⁶ Google lists several causes. They include links without tracking tags, redirects that strip the tags, URL shorteners, typed addresses, links in offline documents such as PDFs and ad blockers.⁶ None of those causes is specific to AI, so an unlabelled visit isn’t proof that an assistant sent it.

That leaves a gap you can’t close from the report. Treat your AI count as a floor, which means only the visits that were identified. Don’t scale it up by a guessed share.

Kavera note: A link in an AI answer gives a reader a route to the website. In our Sydney trades index, ChatGPT’s 50 completed answers contained 144 business appearances and at least 140 had a link to the business’s own website. Perplexity linked hipages.com.au in 12 of 50 completed answers. These are link observations and not recorded website visits.

Bots are a separate question. GA4 automatically excludes traffic from known bots and spiders, using Google research and a list maintained by the Interactive Advertising Bureau. Google says you can’t turn this off or see how much was removed.⁸ Programs that read your pages for AI products aren’t people clicking through, so they aren’t what this report is meant to count. Our guide to AI crawlers and robots.txt covers how to see and manage them.

How do you track Google AI Overviews and AI Mode?

You track them in Search Console and not with the AI Assistant channel. Google’s channel definitions place visits from AI Overviews and AI Mode in Organic Search.¹ Google also says those features are reported within the Web search type in Search Console, so their clicks sit inside your normal Web totals.¹⁰

Search Console now has a Generative AI performance report. It counts impressions, meaning how many times links to your site were shown in AI Overviews and AI Mode. It splits them by page, country, date and device.⁹ It doesn’t report clicks. It may not appear until your site has had enough impressions in those features.⁹ Our guides to Google AI Overviews and AI Mode explain what each feature shows.

For a local business the practical read is simple. GA4 tells you about visits from assistants such as ChatGPT. Search Console tells you how often Google’s AI answers showed your pages. Keep the two apart and don’t add them together.

How do you tell whether AI visitors become enquiries?

Mark the successful enquiry action as a key event, which is an action you tell GA4 matters, such as a sent enquiry form. Then select that specific event from the All events menu in the Key events column and read its count against the AI channel.⁷ Test that one successful submission creates one event and one received enquiry before you use the count to judge performance.

For a service business the event is the enquiry form being sent or a booking made. A form that was opened, a button that was clicked or a page that was visited isn’t an enquiry. If your site can’t fire an event when the form succeeds, ask your developer to add it.

When the AI row holds only a dozen sessions, a single enquiry can swing the percentage a long way. Read the counts first. Compare successful submissions with the enquiries your team actually received before you draw conclusions about lead quality.

What can GA4 not tell you about AI visibility?

GA4 doesn’t count unclicked mentions inside an AI answer or judge whether that answer describes your business accurately. Check those separately with saved answers to a fixed question list. Our measure AI visibility guide explains the routine using the 50-question Sydney trades AI search method.

Kavera note: In two checks about two months apart, our client SHORE went from absent to first in one Sydney builder shortlist (see our ChatGPT SEO guide). That is a shortlist position from one saved question. It says nothing about visits until a click reaches the site, so we record the answer and the visit as two separate measures.

Use GA4 for the click and the enquiry. Use saved answers for the mention, because neither replaces the other.

When do you need a custom channel group?

You can skip this part if the built-in channel already answers your question. A custom group is optional.

You need one in any of these cases:

• You want your own written list of AI sources.
• Google’s list misses a source you care about.
• You want a report you can explain line by line to someone else.

A custom channel group can be applied to your reports retroactively, so it can cover months that were collected before the built-in channel existed.³

Three limits apply. You must be an Editor or above at the property level to create or edit channel groups.³ A standard property can hold two custom groups in addition to the default one, so don’t spend both on this.³ Last, don’t make a custom group your property’s primary channel group just to get this report. Setting a primary group changes how the main channel dimension fills from that point forward.³

How do you create a channel group for AI assistants?

In Admin, under Data display, open Channel groups and create a copy of the default group.³ Then add one channel that matches AI assistants by source. A source rule uses a regular expression (regex), which is a short text pattern that tells GA4 which source names to match.

1. In Admin, under Data display, click Channel groups.³
2. Click Create new channel group. This starts you with a copy of the default group.³
3. Enter a group name, such as “Default plus AI sources”.
4. Click Add new channel and name it “AI assistants (custom)”.³
5. Click + Add condition group, choose Source and choose matches regex.³
6. Paste the pattern below and click Save channel.
7. Click Reorder and move AI assistants (custom) to the top if it should capture every source in your list. It must precede any other matching channel, including Organic Search, Referral and AI Assistant. Click Apply.³
8. Click Save group.³
9. Open Traffic acquisition and choose your new group as the primary or secondary dimension.³

^(.*\.)?(chatgpt\.com|chat\.openai\.com|perplexity\.ai|perplexity\.com|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|grok\.com|chat\.deepseek\.com)$

Step 7 matters because GA4 places each visit in the first channel whose rule it matches, so order decides the result.³ Google’s own example also tells you to move the AI channel above Referral.³

Google’s help page gives a much broader pattern. It matches any source containing letters such as “gpt” or “copilot”, so an unrelated domain that happens to contain them would be counted as AI.³ The pattern above is our own editorial choice and not an official list of assistants. It names domains and nothing else.

Google’s page also says to update your pattern when the assistants or their addresses change.³ Put a reminder in your calendar to check it each quarter against the source rows in your report.

Which sources should the pattern include?

Include only the sources you see in your own Traffic acquisition rows, plus any assistant you know your customers use. GA4 matches the whole source value by default, so the pattern must describe the whole name.⁴ The caret and dollar sign mark the start and end of that name. A backslash before each dot stops the dot acting as a wildcard.⁴

These are the results our pattern gives for a handful of source names. We ran each name against the pattern ourselves.

Bare bing.com is left out on purpose. It’s a search engine, so a visit from it isn’t evidence of an AI answer. Copilot sits at its own address, so the pattern lists that address instead.

If a source you care about isn’t in the list, add it with a vertical bar between names. Then run the same pass again with one name that should match and one lookalike that shouldn’t.

How do you check the report is working?

Check the source definition, channel totals and one known visit before you use the report for decisions. Keep the date and result of each check so you can investigate a mismatch.

1. Check a known visit. If an assistant links to your site, follow the link and record the time and source. Look for it after report processing. If it is absent, check consent, tracking and filters before concluding that the channel rule failed.
2. Compare channels. Put the built-in AI Assistant row next to your custom channel for the same dates. If the totals differ, open the source rows and find the visits that sit in one and not the other.
3. Test the pattern. Run one source that should match and one lookalike that should not, as in the table above.

Here’s an example with made-up numbers. Say the built-in row shows 12 sessions over 28 days and your custom channel shows 15. Open the source rows. Three sessions came from a source Google’s list doesn’t recognise. Now you know exactly which source to name in your notes. If the extra three came from a domain that merely looks similar, tighten the pattern instead.

What should you do next?

Open Traffic acquisition today, add Session source / medium and write down the AI sources you can see. If the list is short, leave the built-in channel as it is and add the form-submission key event. Build the custom group only when you need your own source list.

Our SEO audit is free. Implementation is separate unless retained. See the SEO packages for terms. Monthly plans start at Suburb SEO, $890/month + GST. If you want a team to set up the measurement and the pages behind it, talk to our AI SEO agency. No SEO package guarantees rankings, leads or revenue.