What do AI referrals look like in GA4?

They arrive as referral traffic from chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, and copilot.microsoft.com. Google AI Overview clicks are the exception — they are attributed to normal Google organic search, so they are indistinguishable from regular search traffic.

Why is my AI referral traffic so low if AI visibility matters?

Because the highest-value outcome produces no traffic at all. A homeowner who reads "Carter Plumbing, (813) 555-0100" in an answer and calls that number generates a lead with zero sessions. Traffic undercounts AI's impact structurally.

Should I add UTM parameters for AI?

You cannot — you do not control the links the assistant writes. What you can control is asking every caller how they found you and recording the answer. That is the only reliable attribution available today.

Is AI traffic worth anything if the volume is small?

The volume is small and the intent is unusually high. Sessions from assistants typically arrive further down the decision than generic search traffic, because the assistant already did the filtering. Judge it on close rate, not session count.

Guides/Measurement & buyingupdated 2026-08-04 · 4 min

Is AI Search Actually Sending You Customers? How to Tell

The short answer

Most AI-driven leads never appear as AI traffic. Assistants often show your phone number without a click, and the clicks that do happen frequently land as direct traffic. To measure it you need referral filters for the assistant domains in GA4, a "how did you hear about us" field, and call tracking that captures the answer.

A contractor asked us last month whether any of this actually produces work. Fair question. Here is the honest answer, and how to check it for yourself.

The measurement problem, stated plainly

Three of the four ways AI visibility produces a customer are invisible in your analytics.

1. The zero-click call. An assistant answers "who should I call for a burst pipe in Tampa" with three business names and phone numbers. The homeowner taps a number. You get a call. Your website records nothing, because nobody visited it.

This is the most common and most valuable path, and it is completely invisible to GA4.

2. The brand-then-search. The assistant names you. The homeowner then searches your name on Google to check you out, and clicks through from there. Your analytics credits Google organic search. The AI did the work; Google gets the credit.

3. The direct visit. They read your name, type your domain, arrive. Attributed to direct traffic, alongside everyone who has your card on their fridge.

4. The actual referral click. They click a link in the answer. This one shows up properly, and it is the smallest of the four.

So the honest framing: the traffic number you can measure is the smallest part of the effect. Anyone selling you AI visibility on a promise of traffic growth is either measuring the wrong thing or hoping you will.

What you can measure in GA4

Set this up once. It takes about twenty minutes.

Create an AI referrals comparison

In GA4, go to Reports → Acquisition → Traffic acquisition. Add a filter or comparison on Session source matching any of:

chatgpt.com
perplexity.ai
claude.ai
gemini.google.com
copilot.microsoft.com
bing.com (for Copilot-adjacent traffic)

Save it as a comparison so you can pull it any time.

Make a segment for those sources

Then watch three numbers against your site average:

Sessions — the raw count. Expect it to be small.
Engagement rate — usually higher than average, because the assistant pre-qualified the visitor.
Conversions (form submits, click-to-call taps) — this is the number that matters.

Track click-to-call as a conversion

If you have not already, make every tel: link fire an event and mark it a key event. For a local service business this is the most important conversion on the site and it is frequently untracked.

Note what you cannot see

Google AI Overview clicks arrive as ordinary Google organic traffic. There is no separate source. You cannot isolate them in GA4, and Search Console does not break out AI Overview impressions either.

The practical tell: impressions flat or rising while clicks fall on a query where you know an AI Overview now appears. That gap is the Overview absorbing the click.

The measurement that actually works

Analytics will not solve this. Asking people will.

Add a "How did you hear about us?" field to every form, with an explicit option list that includes "ChatGPT / AI assistant." Not a free-text box — people leave those blank. A dropdown with the option visible.

Train whoever answers the phone to ask it, and to write the answer down. One line in your CRM. "AI/ChatGPT" as a lead source option.

Six weeks of that data tells you more than six months of GA4 reports. It is also the only method that captures the zero-click calls, which are the majority of the effect.

What good looks like

There is no published benchmark for local service businesses, and anyone quoting you a precise industry percentage is making it up. What we can say from the scans we run:

AI referral sessions are typically a low single-digit percentage of total sessions for a local service business today.
Those sessions convert at a noticeably higher rate than average, because the filtering already happened.
Self-reported "found you through AI" lead counts are consistently higher than measured AI sessions — often several times higher — which is exactly what the zero-click theory predicts.

Track your own numbers. Yours are the only ones that matter.

The leading indicator

Because the trailing indicators are so hard to measure, watch the leading one instead: how often the assistants name you.

That is directly measurable, it moves before revenue moves, and it is the thing your work actually changes. Ask the questions your customers ask, count the mentions, and track that number monthly.

If mentions go from 2 of 25 to 14 of 25 over a quarter, the leads follow. If mentions stay at zero, no amount of GA4 configuration will find revenue that was never created.

What to do this week

Set up the AI referrals comparison in GA4.
Make sure tel: link clicks are tracked as a key event.
Add "ChatGPT / AI assistant" to your "how did you hear about us" dropdown.
Tell whoever answers the phone to ask and log it.
Get a baseline mention count so next month's number means something.

Getting the baseline

Our free scan gives you that baseline in about fifteen seconds — twenty-five real customer questions across ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews, with a mention count per assistant and every verbatim answer shown.

Run it today, run it again in a month, and you will know whether the work is working.

Common questions

What do AI referrals look like in GA4?: They arrive as referral traffic from chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, and copilot.microsoft.com. Google AI Overview clicks are the exception — they are attributed to normal Google organic search, so they are indistinguishable from regular search traffic.
Why is my AI referral traffic so low if AI visibility matters?: Because the highest-value outcome produces no traffic at all. A homeowner who reads "Carter Plumbing, (813) 555-0100" in an answer and calls that number generates a lead with zero sessions. Traffic undercounts AI's impact structurally.
Should I add UTM parameters for AI?: You cannot — you do not control the links the assistant writes. What you can control is asking every caller how they found you and recording the answer. That is the only reliable attribution available today.
Is AI traffic worth anything if the volume is small?: The volume is small and the intent is unusually high. Sessions from assistants typically arrive further down the decision than generic search traffic, because the assistant already did the filtering. Judge it on close rate, not session count.

Stop guessing

See exactly what the five AIs say about your business.

Twenty-five real customer questions across Claude, ChatGPT, Gemini, Perplexity, and Google AI Overviews. Free, about fifteen seconds, no account.

Run the free scan