How to see traffic from ChatGPT, Perplexity and Gemini in GA4

Which referrers the assistants send

When someone clicks a link inside an AI answer, their browser usually tells your website where they came from. GA4 records that as the session source. At the time of writing, the sources you will see are:

Two things are missing from that table, and both matter.

First, some visits arrive with no referrer at all. Mobile apps, some browser privacy settings and some in-app browsers strip the referrer, and GA4 then counts the visit as Direct. You cannot recover these from GA4 alone, so every number in this article is a floor, not the full picture.

Second, clicks from Google's AI Overviews and AI Mode are not reported separately. Google treats them as part of Google Search, so they show up as google / organic in GA4 and as ordinary Search results in Search Console. At the time of writing, neither tool has a filter that separates an AI Overview click from a classic blue-link click. If much of your AI exposure is on Google, GA4 will not show it as "AI traffic", and a small chatgpt.com number does not mean AI sends you nothing.

How to see AI referrals in GA4 today

There are two ways: a quick filter on a standard report, and an exploration you can save and come back to.

Option 1: filter the Traffic acquisition report

1. Open GA4 and pick the property for your website.
2. In the left menu, click Reports, then Acquisition, then Traffic acquisition.
3. Above the table, click the dropdown that says Session default channel group and change it to Session source.
4. Click Add filter at the top of the report (the small "+" next to "All users").
5. Set the dimension to Session source, the match type to contains, and the value to chatgpt. Click Apply.
6. Repeat with perplexity, gemini, copilot and claude, or type each into the search box above the table.

This shows one source at a time, and the filter disappears when you leave the report. It is fine for a quick check and not for a monthly review.

Option 2: an exploration with a regex

An exploration matches all five sources at once and keeps the setup.

1. In the left menu, click Explore, then Blank.
2. Name the exploration, for example "AI assistant referrals".
3. Under Dimensions, click the "+" and add Session source, Landing page + query string and Session default channel group. Click Import.
4. Under Metrics, add Sessions, Engaged sessions, Key events (GA4's name for conversions) and Total users. Click Import.
5. Drag Session source into Rows and the four metrics into Values.
6. Under Filters, add one. Choose Session source, set the condition to matches regex, and paste chatgpt|perplexity|gemini|copilot|claude. Click Apply.
7. Set the date range to the last 90 days, so small numbers have a chance to add up.

You now have one table with every assistant on its own row. Swap Session source for Landing page + query string in the Rows slot to see which pages receive the visits. Explorations are saved automatically, so the next person on the team can open the same view.

How to make it permanent with a custom channel group

The standard "Session default channel group" dimension will never have an AI channel unless Google adds one. You can define your own grouping, and then every standard report can show "AI assistants" as its own line next to Organic Search, Paid Search and Referral.

1. Click Admin (the gear at the bottom left).
2. Under Data display, click Channel groups.
3. Click Create new channel group. Name it, for example "Channels with AI".
4. GA4 copies the default channels into the new group. Click Add new channel.
5. Name the channel AI assistants. Under conditions, choose Source, set it to matches regex, and paste chatgpt|perplexity|gemini|copilot|claude. Save the channel.
6. Back on the group's page, drag AI assistants so that it sits above Referral in the list. Save the group.
7. Open any acquisition report, click the primary dimension dropdown, and pick Session Channels with AI (or whatever you named the group).

Why the order matters

GA4 assigns a session to the first channel in the list whose rules match. Referral's rule is roughly "any source that isn't search, social, email or paid". A visit from chatgpt.com matches Referral, so if AI assistants sits below Referral, Referral claims the visit first and the AI channel stays empty. Put it above Referral and the AI rule gets the first look.

Channel groups are not retroactive: GA4 applies the new grouping from the day you save it, so build it now even if you will not read it for a month. You can also make the new group the default on the same Admin page, so the standard reports use it without anyone changing a dropdown.

How to read the numbers

The honest expectation for most 10 to 200 person companies is that AI referrals are a small share of sessions. That is not a reason to ignore them. A visitor who arrives from an AI answer has already asked a question, read a recommendation and chosen to click your name. That is closer to a branded search than to a casual referral.

So read the segment differently from the rest of your traffic:

• Do not rank it by sessions. Compare engagement rate, key events per session and pages per session against Organic Search. If AI visitors convert at a higher rate, which is common for this kind of intent, a small number of sessions can still deserve real attention.
• Watch the trend, not the week. Weekly numbers bounce around. A 90-day view with a month-over-month comparison is the right resolution, and longer ranges also reduce GA4's thresholding, where Google hides small rows to protect privacy.
• Note the split by assistant. ChatGPT usually dominates by volume. Perplexity visitors tend to land on deeper, more specific pages because Perplexity cites several sources per answer. Gemini draws on Google's index, so a change there often shows up in AI Overviews too.

A worked example. Fernwick, a scheduling tool with a two-person marketing team, sets up the channel group in October. By January it sees around three dozen AI sessions a month, mostly from ChatGPT, landing on its pricing page and on a comparison page it wrote against two larger rivals. Those sessions sign up at twice the rate of organic search. The right conclusion is not "AI is 1% of traffic"; it is "the comparison page is being cited and the people it sends are ready to buy, so write the other three comparison pages".

Which pages get the visits, and why

The landing page is the most useful dimension in this report, because it shows which of your pages the assistants actually link to. That is the bridge between the visits in GA4 and the citations you see when you run your prompts against the engines.

1. In the exploration above, put Landing page + query string in Rows and keep the AI source filter.
2. Sort by sessions. The top ten pages are the ones the answers link to most.
3. Compare that list with the pages the answers cite when you track your prompts. Pages cited often but visited rarely are usually cited for a fact; pages that get visits are the ones answers present as "here is where to go".

The gap between the two lists is where the work is. If your pricing page is cited but your product page is not, the answer tells people what you cost without telling them what you do. A tool that tracks citations per prompt, Echo's Citations view for instance, shows the cited pages side by side with their visits. The article on how ChatGPT decides which pages to cite explains what makes a page citable in the first place.

What about UTM links in the answers?

Nothing. You cannot add UTM parameters to the links inside an AI answer, because the assistant links to your page as it found it in its index or its live search. There is no campaign to tag. The referrer is the only signal you get, and the channel group above is the right way to use it.

The limits of what GA4 can tell you

Be clear with your team about what this report does not show.

• No query text. GA4 does not know what the person asked the assistant. You know the landing page and nothing else about the conversation. To learn which questions lead to your pages, you have to run the questions yourself, which is what tracking a prompt set is for.
• No impressions. Search Console tells you how often you appeared in Google results. There is no equivalent for ChatGPT or Perplexity. Being named in an answer without a click leaves no trace in GA4, and most mentions do not produce a click.
• Direct hides some of it. Visits with a stripped referrer land in Direct.
• AI Overviews are invisible as a channel. They are counted inside Google organic.

Put together, GA4 answers "what did people do after an answer sent them here", and nothing before that. For "how often do answers name us", measure AI visibility directly, and use GA4 to prove that it turns into visits and sign-ups.