Track ChatGPT traffic in GA4: a free AI traffic filter builder

How to track ChatGPT traffic in GA4, step by step

To track ChatGPT traffic in GA4, first read what the AI Assistant channel already reports, then add a source filter where you need more detail. There are three places a filter can go: a custom channel group, an exploration and a report comparison. They do not accept the same input, so the builder writes a version for each.

See what GA4 already labels

1. Sign in to Google Analytics and select Reports in the left menu.
2. Select Acquisition, then Traffic acquisition. The table opens on Session default channel group. Look for the row named AI Assistant.
3. Change the table's dimension to Session source / medium.
4. In the search bar above the table, type ai-assistant and press Enter. Each row left is a source GA4 gave that medium.

Google's help page on reports and explorations says this search field uses a contains match and is not case sensitive. It takes one phrase, not a regular expression. To see ChatGPT alone, search for chatgpt.com instead.

Create an AI assistants channel in a custom channel group

1. In Admin, under Data display, click Channel groups. You need the Editor role or above.
2. Click Create new channel group. It starts as a copy of the default group. Give it a name.
3. Click Add new channel and name it, for example AI assistants.
4. Add a condition, choose Source, select matches regex and paste the expression from the builder.
5. Click Save channel.
6. Click Reorder and drag the new channel above Referral, then click Apply. If the group has another channel that would match the same visits, put yours above that one too.
7. Click Save group, then choose the group as the dimension in the Traffic acquisition report.

These are the steps on Google's custom channel groups page. The order matters because Google puts a visit in the first channel whose definition it matches. Google also says custom channel groups can be applied to your reports retroactively, so the new channel shows past visits as well as new ones. A standard property can hold 2 custom groups and a 360 property 5, with up to 50 channels in each.

For a row per assistant, add one channel for each line of the "One channel per assistant" box, all of them above Referral.

Filter an exploration to AI assistants

1. In Explore, under Start a new exploration, select the Free form template.
2. In the Variables panel, click the plus sign beside Dimensions, select Session source and click Import. Import the metrics you want in the same way, such as Sessions and Key events.
3. Double-click Session source to add it to the rows, and add the metrics to the table.
4. At the bottom of the Settings panel, in Filters, select Session source.
5. Select the regex match type, paste the expression and click Apply.

Google's explorations page says you can add up to 10 filters and that all of them must be met. Filters are joined with AND, so ten assistants cannot go in ten filters. They have to be one expression, which is why the builder joins them with the OR sign.

Compare AI visitors with everyone else in a report

1. Open a report and click the comparison control at the top right, then Create new.
2. Choose the dimension Session source and the match type contains.
3. Add each line of the "Values for a contains condition" box as a value. Google says several values in one condition are joined with OR.
4. Click Apply.

Google's comparisons page lists two kinds of match type, exact and contains, and no regular expression. It also says only dimensions that have values appear in the list. A contains match is looser than the expression: the value meta.ai also matches a longer source that merely has those characters in it. If Session medium is offered, a single condition of Session medium exactly matches ai-assistant is the shortest comparison there is.

AI traffic sources: the assistants on the list and how each was confirmed

The builder covers 10 AI assistants. Each address on the list was confirmed on 13 October 2026 in one of two ways: the assistant's maker links to it from its own page, or a request to the address returned the assistant itself or a redirect to it. Lists copied between blog posts were not used as a source.

The third column says where Google names the assistant: in the description of the AI Assistant channel, in the release note for it, or in the example on the custom channel groups page. "Not named" means none of those pages mentions it, so whether GA4's own channel includes it is not something you can read anywhere. Your expression will still find its visits.

The list is short on purpose. An address that could not be confirmed at its source is left out. If your referral report shows an assistant that is missing, type its hostname into "Add your own" and it joins the expression. Addresses change, too: two of the rows above are earlier addresses that now redirect.

Do AI visitors convert? Compare the segment before you act

Once the segment exists, the useful question is not how many AI visitors you get but what they do. Compare the segment's conversion rate with your other channels, on the same key event and the same dates. A small segment swings widely from week to week, so check whether a gap is real before you change anything because of it.

The table shows why. Both halves have the same two conversion rates. Only the number of sessions differs. Every figure is made up to show the arithmetic and is not a benchmark.

In the first month the AI Assistant segment converts at 3.50% against 2.50% for Organic Search. It looks like a clear lead, but with 14 key events a two-sided test gives a p-value of 0.2138, well above the 0.05 that the usual 95% level asks for. With four times the sessions and the same two rates, the p-value is 0.0129. The rates did not change. The evidence did.

1. Work out each rate. Key events divided by sessions, per channel. The conversion rate calculator does it, and the guide to conversion rate in GA4 shows where the numbers are.
2. Test the gap. Put both channels into the statistical significance calculator. If the result could be chance, keep collecting before you draw a conclusion.
3. Look at where they land. Add Landing page to the exploration. The assistant chose the page, not you, and it may be one you never designed as an entrance.
4. Fix the pages, not the channel. You cannot edit what an assistant says. You can change what its visitors find. Run those landing pages through the free CRO audit, which scores the first screen, calls to action, forms, trust, mobile, speed and AI agent readiness.

A low rate from AI visitors is not always poor traffic. Someone who has already read a summary of your page may arrive with one narrow question, and a page that buries the answer loses them. That is a page problem, and it is one of the few parts of this channel you control.