How to Track Traffic Coming From AI Assistants

Traffic from AI assistants shows up in your analytics as referral traffic, from domains like chatgpt.com, perplexity.ai, and copilot.microsoft.com. It is measurable today. It is just buried under a default report that lumps it in with everything else.

An hour of setup turns that noise into a number you can watch.

Why the default view hides it

Most analytics tools group traffic into broad channels: organic search, direct, referral, social. AI assistants do not have their own bucket yet, so their visits scatter — some into referral, some into direct when the link carries no identifying information.

It is like tracking visitors to a shop by counting which street they came from, while a new footbridge opens that nobody added to the map. People are arriving. Your report just calls them "other."

Step one: build a referral list

Start with the hosts that currently send meaningful traffic:

• chatgpt.com and chat.openai.com
• perplexity.ai
• copilot.microsoft.com and bing.com/chat
• gemini.google.com
• claude.ai

Treat this list as living. New assistants appear, and hostnames change. Review it every quarter rather than assuming it is complete.

Step two: give the group a name

In GA4, create a custom channel group with a rule that matches those hostnames in the referral source, and label it something unmistakable like AI assistants. Every report that offers channel breakdowns will then show it as its own line.

If you use a privacy-focused analytics tool instead, the equivalent is usually a saved filter on referrer domain. Same idea, fewer clicks.

Step three: measure behavior, not just volume

Raw sessions from AI assistants will look small at first. Resist judging the channel on that number alone. What matters is what those visitors do.

Compare, for your AI group against organic search:

• Pages per session. Assistant referrals often arrive pre-informed and go straight to a service or contact page.
• Time on page. Short visits may mean the assistant already answered the question — not necessarily a failure.
• Conversion rate. Form submissions and email clicks are the honest test.

A channel delivering a tenth of the sessions at three times the conversion rate is not a small channel. It is a concentrated one.

Step four: watch the questions, not only the clicks

Some AI answers cite you without sending a visit. The user got what they needed inside the assistant. That is invisible in analytics and still valuable — your company was named as the credible source.

To keep an eye on it, run a short manual check each month. Ask three or four assistants the questions your clients ask. Record whether you appear, what is said about you, and whether the details are correct.

Keep the results in a simple spreadsheet with the date. Over six months you will have something no dashboard gives you: a record of how machines describe your business, and whether that description is improving.

What to do with the data

If AI referrals convert well, the lesson is usually about content shape rather than volume. Look at which pages receive those visits and what they have in common — typically a direct answer near the top, concrete specifics, and clean structure. Then apply that shape elsewhere.

If assistants cite you but the details are wrong, the fix is almost never analytics. It is the inconsistent information on your own site and profiles. Measurement points at the problem; it does not solve it.