Measuring AI-Referred Traffic in Your Analytics

What Google Analytics tracks automatically now

Google Analytics 4's default channel groups are the rules that sort every visit into a category — Organic Search, Paid Search, Referral, Direct, and so on — based on where the visitor came from. Google's own documentation for these channel groups added a new one in May 2026: a session is placed in the "AI Assistant" channel when its medium exactly matches "ai-assistant," which Google sets automatically when the referrer matches a specific, named list of AI assistants — ChatGPT, Gemini, Deepseek, Copilot and Grok.

Two things about this are easy to miss. First, it is not retroactive: sessions recorded before the channel existed keep whatever classification they originally got, usually Referral, so a before/after comparison of "AI Assistant" traffic against older data will understate how much you actually had. Second, the list of recognized sources is short and specific, not a general "anything that looks like AI" rule — a referrer has to be on Google's list to get the label, and anything not on it is sorted the old way, by its domain and whatever source/medium signals normally apply.

Practically, this means a session from ChatGPT.com should now show up correctly labeled with no work on your part, going forward from when your property started receiving the update. It does not mean every AI-driven visit is now accounted for — which is the actual gap worth understanding.

The gap: Perplexity, and anything not on the list

Perplexity is the clearest example of what falls outside the automatic label. It is not one of the five sources Google's own AI Assistant channel documentation names, so a visitor arriving from a Perplexity answer is not tagged "ai-assistant" — that session is sorted the ordinary way, typically landing in Referral with a source of perplexity.ai, indistinguishable at a glance from any other referring website.

The same applies to any assistant that launches after your property's channel definitions were last set, or any AI product that routes clicks through a URL structure Google's matching rules don't recognize. New assistants appear often enough that treating the default channel list as complete, rather than as a snapshot from May 2026, will quietly under-count your AI traffic as the AI landscape moves.

The fix is a custom channel grouping — a rule you define yourself, alongside the default one, that groups sessions by referrer domain (perplexity.ai and any others you want to track individually) into a channel of your own naming. This does not touch the default "AI Assistant" channel or need to replace it; it sits next to it and catches what the built-in rule doesn't. Similarweb's traffic research confirms why this matters in practice: AI referral traffic overall grew 117.4% year over year through May 2026, and ChatGPT alone accounted for more than 80% of AI referrals to the top 1,000 global domains — meaning most of your AI traffic likely is captured by the default channel, but the fastest-moving remainder, from newer or smaller assistants, is exactly the part a static list will always be one step behind on.

Why AI-referred sessions behave differently once they arrive

A visitor who clicks through from an assistant's answer has usually already had several questions answered before they ever land on your site — what you do, whether you're a good fit, often even a comparison against alternatives. That changes the shape of the session that follows: fewer of them arrive in the first place relative to search, because most assistant answers satisfy the question without a click at all, but the ones that do arrive tend to be later in a decision and more specific about what they're looking for.

This is consistent with what happens to conversion numbers once a business does start tracking this channel separately: Similarweb's research, drawing on Adobe's traffic data, reports AI-sourced traffic to US retail sites converting 54% better than non-AI sources by May 2026 — not because AI traffic is inherently more valuable, but because the visits that do arrive have already been pre-qualified by the conversation that preceded them.

The practical implication is not to expect AI-referred traffic to look like search traffic at a smaller scale. Expect fewer sessions, and judge them on conversion and engagement rather than volume — a channel that sends you ten visits a month but three of them convert is telling you something a raw traffic comparison to Organic Search will hide.

What this number can't tell you — and the honest way to use it anyway

The largest source of AI influence on a buying decision never appears in analytics at all: someone asks an assistant a question, gets an answer that mentions or recommends your business, and either already has enough information to act — a phone call, a visit, a search for your name on Google instead of a link — or decides against you, in either case without a single click landing in your traffic reports. OpenAI's own documentation on how ChatGPT search works confirms the same shape from the platform side: an assistant's response shows the sources it consulted and the citations most relevant to the answer, but a citation being shown is not the same event as a person clicking it, and most of what an assistant says about you is never clicked through regardless of whether it names you fairly.

That doesn't make the analytics channel worthless — it makes it one input rather than the whole picture. Use it to confirm the assistants that do send clickable traffic are tagged correctly, to catch the gap (Perplexity and anything newer than your channel definitions), and to compare how AI-referred sessions convert against every other channel you already track. Use a separate measurement — direct tracking of what assistants actually say when asked about you, which on-site analytics structurally cannot see — for the fuller picture of your AI visibility.

Treat the traffic number as a lagging, partial confirmation that your visibility work is landing, not as the primary way you'd find out it wasn't. A business that only checks analytics for AI traffic and sees a small number has learned that few people clicked through — not that few people asked, and not that the assistant's answer was wrong.

• Confirm your GA4 property is on the updated default channel groups, then check the AI Assistant channel for any traffic at all.
• Add a custom channel grouping keyed to perplexity.ai and any other AI referrer domains you see in raw Referral data.
• Compare conversion rate and session depth for AI-referred traffic against your site average, not against Organic Search volume.
• Revisit your custom list every quarter — new assistants launch faster than default channel definitions get updated.

FAQ