How to Measure AI Search Traffic in Google Analytics

Everyone is talking about traffic from AI assistants, and almost nobody can tell you how much of it they actually receive. Ask a typical site owner and the answer is a shrug. The visits are there, but they are scattered across "referral", "direct" and "unassigned" in the analytics, so the channel looks smaller, or simply invisible, compared with what is really happening.

By the iSuggest.ai Team · Updated for 2026

This guide shows how to find AI-referred visitors in Google Analytics 4, group them into a channel of their own, and read the numbers without fooling yourself. No special tools are needed, only access to your analytics and about an hour. For the background on why this channel matters, see our complete guide to GEO.

Why measure it at all

You cannot improve what you cannot see. Industry analysis reports that traffic referred by AI search engines grew about 16x between 2024 and 2026 (SEO Sherpa), and that the assistants are growing at different speeds: one analysis reports ChatGPT's absolute traffic up 27% year on year, Gemini up 231% and Claude up 320% (PPC Land). Those are figures from specific datasets, and growth from a small base can still be a small number. Which is exactly why you should look at your own data: it tells you whether this channel is a rounding error for you or something worth feeding.

Measurement also closes the loop on everything else you do. If you fix crawler access, improve your answer pages or earn some coverage, the only way to know whether it moved anything is to watch a metric. Referral sessions from assistants are one of the few concrete numbers available.

What AI traffic looks like in analytics

When someone clicks a link inside an assistant's answer, their browser usually sends a referrer, the address of the site they came from. Analytics reads that and records the session's source. Depending on the assistant and the way the link was opened, you might see any of the following.

A recognisable referrer domain. For example, an assistant's own web address appearing as the session source with a medium of "referral".
A tagged link. Some assistants append a source parameter to the links they show, which analytics then reads as the source. This is the easiest case.
No referrer at all. If the link was opened from an app, copied and pasted, or the referrer was stripped for privacy, the session lands in "direct" or "unassigned" and you cannot see where it came from.

The practical consequence is that any measurement you do will undercount. Treat the figure you get as a floor, not a total. That is still useful, because a floor that rises over time is a real signal.

Source names and behaviours change as products change, so do not trust a list from a blog post, including this one, without checking it against what appears in your own reports. The method below is built to let your data tell you.

Step 1: Look in the places the traffic already shows up

Before building anything, see what is there. In Google Analytics 4, open the acquisition reports and look at sessions by source and medium.

Go to the reports section and open the traffic acquisition report.
Change the primary dimension to session source and medium, so you see pairs such as a domain and "referral".
Set the date range to the last 90 days, so small sources have time to appear.
Sort by sessions and scroll. You are looking for assistant-related names: the web addresses of the chat assistants and AI search products you know of.

Write down every source that looks relevant. Do not worry yet about being complete; you will refine the list in a moment. If you find nothing, that is a finding too: either the volume is tiny, or the visits are landing in direct and unassigned.

Step 2: Check the landing pages

A second view helps catch what the source list misses. Open the landing page report and look at which pages receive traffic with a "referral" medium. Pages that attract assistant referrals tend to be the ones that answer a specific question clearly: guides, comparisons, pricing explanations. If one of your answer pages has a steady trickle of referral visits from sources you do not recognise, investigate the source before dismissing it.

You can also add a secondary dimension for the full referrer or page referrer where available. That sometimes reveals a more specific origin than the source field shows.

Step 3: Build an AI channel with a custom channel group

GA4 puts every session into a default channel such as organic search, direct, referral or social. AI visits end up in "referral" mixed with everything else. A custom channel group lets you define your own channel and report on it by name.

Open the admin area and find the channel group settings for your property.
Create a new custom channel group and give it a clear name, such as "Channels with AI".
Add a new channel called "AI assistants".
Set its condition: session source matches regex, followed by a pattern listing the assistant domains you found in step 1, separated by the pipe character.
Move the new channel above the default referral channel in the ordering, because channels are evaluated from top to bottom and the first match wins.
Save the group.

A pattern might look like the following, which is only an illustration of the shape. Replace it with the sources you actually saw in your data.

chatgpt|openai|perplexity|claude|gemini|copilot

Keep the pattern loose enough to catch variations and tight enough that it does not sweep in unrelated sites. A fragment such as "claude" is short, so check the matches in your reports to make sure nothing unrelated has been included. If a source with a similar name turns out to be something else, make the pattern more specific.

One caution: in GA4, a custom channel group applies to reports going forward and in the reporting interface, and the details of how it treats past data have changed over time. Check the current documentation for your property before relying on historical comparisons.

Step 4: Make an exploration for flexible digging

The channel group gives you a tidy number. An exploration gives you freedom. Create a free-form exploration with these settings.

Rows: session source and medium, or page referrer.
Columns: none at first; add device or country later.
Values: sessions, engaged sessions, engagement rate, key events or conversions, and average engagement time.
Filter: session source matching the same regex you used for the channel.

Save it and put it in your regular reporting routine. The point of the exploration is to look at behaviour, not only volume: do visitors from assistants stay, read and convert differently from those from search? A small number of visitors who engage deeply may be more valuable than a larger number who leave at once.

Step 5: Find the hidden share in direct and unassigned

You cannot recover the visits that arrive without a referrer, but you can look for clues that some of them are AI-driven.

Deep landing pages in direct. People type your homepage into a browser. They do not type the address of a specific guide. Direct sessions that land on a deep article are more likely to be pasted or app-opened links.
Spikes after a mention. If a page suddenly gains direct visits at about the time you were named somewhere, a link may have been opened without a referrer.
Branded search growth. People who learn of you in an assistant often search your name afterwards. A rise in branded queries in Search Console, without a matching rise in other activity, is a hint.

None of these is proof. Use them to adjust your reading of the numbers ("the real figure is probably higher than reported") and not to invent a number. Never add an estimated share to a figure and then report it as measured.

Step 6: Tag the links you control

You cannot tag the links assistants show, but you can tag the ones you place. If you share a link in your own content, a profile, a directory entry or an email signature, add campaign parameters so analytics knows where it came from. For your iSuggest.ai directory snapshot or any other listing you maintain, a consistent source and medium makes those visits easy to separate from everything else. Choose a convention and write it down, so that you do not end up with five spellings of the same source.

Step 7: Report it honestly

A simple monthly report is enough. Include these items, and keep the wording modest.

Sessions from the AI assistants channel, this month and last month.
Its share of total sessions, with the caveat that it is a floor.
Engagement rate and key events for the channel compared with organic search.
The top five landing pages for the channel.
One sentence on what you changed and what you plan to change next.

Small numbers are normal. A site with a few dozen AI referral sessions a month is not failing, it is where most sites are. What you want to see is direction over several months, and engagement that is at least comparable to other channels.

A worked example

Imagine a small company that publishes guides and sells a subscription. This is an illustration, not a case study. The owner looks at the traffic acquisition report and, in the last 90 days, sees a handful of sources whose names are assistant domains, each with a small number of sessions. Together they are well under one percent of total traffic.

The owner builds a custom channel called "AI assistants" with a regex that covers those sources and moves it above referral. The next month the report shows the same small share, now in one line. Looking at landing pages, most of the visits arrive on two guides that answer a specific question clearly. Engagement rate for the channel is similar to organic search.

The owner takes three actions. First, improve those two pages and link them to related guides. Second, write two more pages in the same style. Third, put a quarterly reminder in the calendar to refresh the regex with any new sources. After a few months, the channel's share has edged up and the owner has evidence to show which kind of page earns it. No magic, just a visible metric.

Common mistakes when measuring AI traffic

Treating the channel as the whole picture

AI visibility also works through brand awareness: someone sees your name in an answer, then searches for you or types your address later. That value lands in other channels. Use the AI channel as an indicator, not as the entire accounting.

Using an out-of-date list

New assistants appear and old ones change their domains. Review the sources every quarter, and look at the unassigned and referral lists for anything new.

Overreacting to small samples

A single week with twelve sessions does not prove a trend. Look at rolling 28 or 90-day totals.

Ignoring quality

Volume is not the only measure. A channel with fewer visits but higher engagement and more conversions may deserve more attention than its size suggests.

Reporting guesses as facts

If you estimate hidden traffic, label it as an estimate and keep it out of your headline numbers.

Measure the cause as well as the effect

Analytics tells you what happened after a visit. It cannot tell you why you were or were not mentioned. For that you need to look at the machine-readable side of your site: can crawlers reach it, is the text in the HTML, is the identity clear, is the structure sound? Our post on how to check if GPTBot can crawl your website covers the first of those, and zero-click searches and how to still get traffic explains why some visibility never shows up as a visit at all.

It also helps to run the same set of real questions through a few assistants each month and record whether you are named. Pair that with the analytics channel and you have both an input measure (are we mentioned?) and an output measure (do people arrive?). Answers vary between runs, so look for patterns over weeks and not single results.

Start with a baseline

The free iSuggest.ai audit gives you the cause-side baseline: access, rendering, markup, headings, identity and more, from a machine reader's point of view. You can read what an audit measures to see each item. Audits are free and not metered, so you can record a score now and repeat it after each change, then set it alongside the analytics channel you have just built. For the rhythm, see how often to re-audit your website.

When your pages are ready, you can publish a factual snapshot to the iSuggest.ai AI directory, with pages for ChatGPT, Claude and Gemini. Publishing spends credits, which are non-refundable, and improves findability without guaranteeing citations or visits. See how it works, run an audit on your homepage today, and sign in at your account page when you are ready to publish.

How to Measure AI Search Traffic in Google Analytics (Step by Step)