How to measure AI traffic to your website

You measure AI traffic by segmenting the known assistant referrer domains — chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com — into their own channel in GA4. That captures the visible slice; a large share of AI-driven visits arrives with no referrer at all, so treat the channel as a floor, not a total.

Which referrer domains count as AI traffic?

Start with the five that account for most assistant-referred sessions:

Legacy sessions may still show chat.openai.com, and new surfaces appear regularly — review your referrer report quarterly for domains you haven't classified yet.

How do you set up an AI channel in GA4?

1. Go to Admin → Data display → Channel groups.
2. Copy the default channel group (you can't edit the original) and open the copy.
3. Add a new channel named "AI / LLM" and move it above Referral and Organic Search, so it wins the classification.
4. Set the condition to Source matches regex with:

.*(chatgpt\.com|openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com).*

5. Save, then compare the channel against Referral over the next weeks. Channel groups are not retroactive, so also build an exploration with a session-source regex filter using the same pattern to see history.

Fifteen minutes of setup, and "how much AI traffic do we get?" becomes a report instead of a guess.

Why does AI traffic look smaller than it is?

Because much of it is dark. Visits from native mobile apps, from some in-app browsers, and from links a user copies out of a chat all arrive referrer-less and land in Direct. Some assistant surfaces strip or generalize referrers by design. So your AI channel is systematically undercounted, and your Direct channel quietly absorbs the difference. If Direct rose over the same period AI assistants took off in your category, some of that is misfiled AI traffic.

The practical response: watch the trend of the AI channel rather than its absolute share, and corroborate with landing-page patterns — AI-referred users tend to enter on deep, specific pages, not the homepage.

Is AI traffic actually worth anything?

The early numbers say it converts unusually well:

• Criteo's early data from a sample of 500 retailers showed users arriving from LLM platforms converting at roughly 1.5x the rate of other referral channels. Small sample, narrow window — but the direction is consistent.
• Industry estimates in late 2025 put retail conversion from AI referrals at roughly 11%, up about five percentage points year over year.
• Walmart's EVP of product and design Daniel Danker told Wired that ChatGPT drove roughly twice the rate of new-customer acquisition compared to search engines — even as Walmart found in-chat checkout converted at a third the rate of sending shoppers to its own site.

The pattern behind all three: a visitor who arrives from an AI answer already did their comparison inside the chat. They show up pre-qualified, later in the funnel, ready to act.

Why traffic still understates AI's influence

The biggest limitation isn't dark traffic — it's that the most important AI interactions never produce a visit at all. A buyer asks an assistant to compare five vendors, gets a synthesized answer, and contacts the winner directly, or buys elsewhere on the assistant's recommendation. Your analytics records nothing either way. Where your traffic went and where your influence went are now different questions.

That's why traffic measurement needs a partner metric: share of voice in AI answers — how often assistants mention you across the questions your buyers ask. Traffic tells you what happened after the answer; share of voice tells you what the answer said. Run both, and you can finally connect "we're being recommended less" to the revenue line before the quarter ends.

And if your share of the answers is lower than it should be, the causes are measurable too — run a free Legible report to see which of the eight machine-facing dimensions is holding your visibility down.