AI search traffic in Google Analytics: how to track ChatGPT and AI referrals in GA4

Can GA4 show traffic from ChatGPT and other AI tools?

Yes, partly. GA4 records a visit as AI related when the browser tells it which site the visitor came from, or when the link carries campaign tags, and that source matches a rule. Google's own documentation lists an AI Assistant channel in the default channel group as of October 2026. Visits that arrive without that information are filed as Direct or Unassigned, so GA4 always undercounts AI driven traffic. Treat what you see as a minimum.

This matters because AI tools change how people find you. A buyer may read an answer that names your company, never click, and later search your brand name or type your address. GA4 will record that visit as organic search or direct. The AI answer did its work, but the analytics cannot see it. That gap is a limit of the tool, not a sign that GEO has no effect.

Below we explain how channels work, how to build a report, what to check, and what the numbers cannot tell you. For the wider measurement picture, see how to measure GEO performance.

How GA4 decides which channel a visit belongs to

Every GA4 session has a source, a medium and sometimes a campaign. The default channel group applies a fixed set of rules to those values and puts the session in the first channel whose rule it matches. Google documents these rules in its help article on traffic source dimensions and default channel definitions.

Several channel definitions matter for AI traffic. We checked the following against Google's Analytics Help pages in October 2026.

Channel	How Google defines it	What it means for AI traffic
AI Assistant	Visits from AI assistants. The medium is set to a reserved AI assistant value when the referrer matches Google's list of AI assistants. Google's AI Overviews and AI Mode are excluded.	Your cleanest view of clicks from chat tools, limited to the tools on Google's list
Organic Search	Non ad links from search sites, and Google states this includes AI Overviews and AI Mode	Clicks from Google's AI answers are mixed in with ordinary organic clicks
Referral	Non ad links on other sites or apps, where the medium is referral, app or link	AI tools not on the recognised list may land here, with the tool name as the source
Direct	Source is direct and the medium is not set or none, such as typed addresses and saved links	Visits where the AI tool passed no referrer end up here
Unassigned	Traffic that matches none of the other rules	Odd source and medium combinations end up here

Two points deserve attention. First, the AI Assistant list is Google's, not yours. Google can add or drop tools from it, and trade press has reported that the named examples changed after launch. Check the current definition in your own property rather than trusting an old article, including this one. Second, Google AI Overviews and AI Mode clicks are counted under Organic Search, so you cannot separate them from classic search clicks in the default channels alone.

Step by step: see AI sources in your reports

1. Open the traffic acquisition report. In GA4, go to the Reports area, then Acquisition, then Traffic acquisition. Set the date range to the last 90 days or since you started tracking.
2. Check the default channel group first. Look for the AI Assistant row. If it is there, note sessions, engaged sessions and key events.
3. Switch the primary dimension to session source and medium. This shows the raw values behind the channels. Scan for AI related hosts, for example chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai. Host names change, so check what actually appears for you.
4. Look at landing pages. Add landing page as a secondary dimension and filter to the AI sources. Learn which pages AI tools send people to.
5. Check engagement and key events. Compare engagement rate, time and conversions with other channels. Small numbers are common, so read patterns not single days.
6. Save the view. Create a saved exploration or a report you can reopen each month.

Menu names in GA4 move around over time. If a path above does not match your screen, search the help centre for the report name.

Build a custom channel group for AI sources

The default channel is useful, but a custom channel group gives you control. Google's help centre describes custom channel groups: you define channels with rules on fields such as source, medium and campaign, and traffic goes into the first channel whose definition it matches in the current order. That order matters. Put your AI rule above Referral, or the same visits will be caught by Referral first.

A typical setup is one channel called AI tools, with a rule where source matches a regular expression listing the hosts you have seen. For example, a pattern like chatgpt\.com|perplexity|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai. Treat this as a starting point, not a complete list. Review the source list every quarter and add hosts you find.

• Create the group in the Admin area under channel groups, as described in Google's help.
• Put your AI channel above Referral and Direct in the order.
• Keep Google AI Overviews and AI Mode out of it. You cannot see them separately in referrer data, since they arrive through Google Search.
• Document the regular expression and the date you changed it, so later readers know what changed.
• Remember that custom channel groups are for reporting, and you should check Google's current guidance on whether they apply to past data in your setup.

Where the data stops: dark traffic and missing referrers

Analytics gets the visitor's origin from the browser's referrer information or from tags added to the link. Anything that removes both leaves a session with no clear source. That is usually called dark traffic. For AI search, there are several causes.

• No click at all. The buyer reads the answer and never visits. GA4 cannot record a visit that did not happen.
• Brand search after an answer. The buyer sees your name, then searches for it. That visit appears as organic search, with your brand as the query.
• Typed or copied addresses. Visits from a pasted link or a remembered name land in Direct.
• Apps and in app browsers. Some apps strip or never send a referrer, and the visit appears as Direct or Unassigned.
• Privacy settings and browsers. Some setups limit referrer data.
• Mixed up tools. A tool not on Google's recognised list may show as Referral or under a source that looks unrelated.
• Consent choices. Where visitors decline analytics cookies, you may not see them at all.

Because of these gaps, a drop or rise in AI referral sessions may reflect a change in how a tool sends referrers, rather than a change in real traffic. Always compare a change in AI referrals with changes in direct traffic, branded search and enquiries. If they move together, the story is more believable. If only one number moves, be careful.

Add measures that analytics cannot give you

Pair GA4 with three simple sources of evidence.

1. A how did you hear field. Add an optional question to contact and booking forms, with AI tools as a listed choice and a free text box. Count answers monthly.
2. Sales call notes. Ask new customers whether they used an AI tool when choosing. Record the answer in your CRM in a consistent way.
3. Server logs. Logs show which bots fetch your pages. Some AI products fetch pages on behalf of a user, for example ChatGPT-User, and those requests are not human visits. They usually do not run your analytics scripts, so they are not in GA4. Log analysis helps you check crawler access, as covered in what to allow in robots.txt.

None of these is precise. Together they give a more honest account than GA4 alone, and they cost very little to set up.

How to report AI traffic without overclaiming

Report AI referral sessions as a floor with a note on what is missing. Show the same chart for the previous period, and put direct traffic and branded search next to it. State the date the AI Assistant channel or your custom group started to apply. Avoid per session value claims unless you have enough conversions to support them.

Pair the traffic view with visibility data from a fixed prompt set, as described in AI visibility metrics. Visibility shows whether you are named. Traffic shows whether some of those namings turned into clicks. Our GEO reporting template puts both on one page.

Which pages do AI visitors land on? Use landing page reports

Knowing the total number of AI sessions is less useful than knowing where those visitors arrive. Landing pages show which of your pages engines send people to, and that tells you which pages are doing citation work. In GA4, open a report that shows landing page as a dimension, add a comparison or filter for your AI channel or your regular expression on session source, and sort by sessions.

1. Open Explore and start a Free form exploration. Set the date range to at least 90 days.
2. Add the dimensions Landing page plus query string, Session source and Session medium. Add the metrics Sessions, Engaged sessions and Key events.
3. Place Landing page in rows and Sessions in values. Add a filter where Session source matches your AI pattern.
4. Export the top 20 rows to a spreadsheet and write what each page is for: guide, product, pricing, contact.
5. Compare with your list of pages that engines cite in your prompt tests. Pages that are cited often but get few visits may lack a clear next step. Pages that get visits but are never cited in your tests may be reached through other routes, such as links shared in chats.

Use the results to decide where to spend effort. If AI visitors land on a guide with no call to action, add a next step. If they land on a page that has gone out of date, fix it first. This ties directly to content freshness for AI search.

Tag the links you control

For links you place yourself, you can remove the guesswork. If you publish a link to your site in a place where AI tools may read it, such as your own profiles, press kit or public documents, add consistent campaign parameters so that visits are traceable. Use a simple naming rule and keep it in a shared sheet. Do not add tags to every internal link on your site, since tagged internal links can overwrite the true source of a session in your reports.

Troubleshooting: why your AI traffic looks wrong

Example, hypothetical: a B2B services site over one quarter

Imagine a fictional UK engineering consultancy. In January GA4 shows 14 sessions from AI sources and the form asks nothing about discovery. The team builds the custom channel, adds a how did you hear field, and rewrites its three highest value service pages to answer buyer questions directly. By March GA4 shows 41 AI sessions, direct traffic is flat, branded search is up, and four of nine new enquiries mention an AI assistant in the form. The honest report states that 41 sessions is a small base, that the rise may reflect both better visibility and better tracking, and that the four form mentions are self reported. It does not state that the rewrite caused the rise. These numbers are invented for illustration.

Other data sources to place beside GA4

• Google Search Console: shows search performance for Google. Check Google's documentation for how AI features are counted and what it lets you filter.
• Bing Webmaster Tools: relevant for Microsoft Copilot and Bing. Look at what reports Microsoft currently offers for AI related visibility, and note the date.
• Server or CDN logs: show AI crawler requests, which is a different thing from user clicks. See which AI crawlers to allow.
• CRM and call tracking: show which leads mention AI tools.
• Prompt test records: show whether you are named, regardless of clicks.

None of these replaces another. Together they show visibility, access, visits and outcomes. If you want help setting up the tracking and connecting it to visibility, the baseline week of the GEO Pilot checks it.

What to do next

This week, open your traffic acquisition report, look for AI sources, build the custom channel group and add the how did you hear field to your forms. That takes an hour or two and gives you a baseline you cannot recreate later.

On the GEO Pilot, a fixed $999 project delivered inside 30 days, we check this setup as part of the baseline, then implement the fixes on your site and retest the same buyer prompts in week 4. You approve changes, and we do not promise any traffic result.