AI referral traffic is now one channel: ChatGPT

Almost all of your AI referral traffic comes from one place: ChatGPT. A new Previsible report, published on 6 July and built on 6.77 million AI-driven sessions across 166 GA4 properties, puts ChatGPT at 92.4% of trackable LLM referral traffic. The other four assistants split what is left. So the practical move is to stop treating AI search as a five-platform problem and start measuring ChatGPT as its own channel.

That one number changes where you spend and what you track. Here is how to read it.

The number that reframes your AI referral traffic

Previsible looked at 166 sites that stayed live for the full window, November 2024 through May 2026, and counted every session an AI tool sent them. Total AI-driven sessions grew 9.9x over those 19 months, reaching 644,478 in May 2026. ChatGPT alone grew 12.8x, from 47,606 sessions to 610,910. It is not only winning. It is pulling further ahead while the whole market expands. Search Engine Land covered the same dataset and reached the same first instruction: optimize for ChatGPT before anything else.

The other assistants are not close

Gemini is the quiet number two, up 3.2x with little drama. Claude is the one to watch: it grew 64x and overtook Perplexity in March. Everything below that is shrinking. Perplexity has fallen 61% from its March 2025 peak. Copilot collapsed 96% from its high last August. If your plan for AI search has a Perplexity workstream and a Copilot workstream, you are spending time on surfaces that are getting smaller.

The channel is bigger in some industries

The growth is not even. E-commerce AI traffic grew 37x from almost nothing. Insurance grew 18.9x. If you sell in a category where buyers research hard before they commit, assume the AI channel is already a larger slice of your traffic than a site-wide average suggests, and look at it before you decide it is too small to matter.

What to do with your AI referral traffic

Three moves, in order.

Name ChatGPT as a channel. Out of the box, GA4 files chatgpt.com under referral or unassigned, so it hides inside your other traffic. Build a custom channel group that isolates the AI assistants, and check it every week the way you check organic. You cannot manage a channel you cannot see. If you have not built reporting you trust yet, start with analytics you will actually use to make decisions.

Optimize the pages ChatGPT already sends people to. Track AI traffic by page type, not site-wide. In the study, product and pricing pages act as the main entry points, because people arrive mid-decision rather than at your homepage. Make those pages state your pricing and specs as plain text an assistant can read and quote, not as an image or a number buried in a script.

Watch one challenger, skip the rest. Keep an eye on Claude, since it is the only assistant gaining on ChatGPT. Put down the Copilot and Perplexity optimization guides for now. When a surface is down 61% or 96%, it is not where your next customer is coming from.

Referral sessions are the visible tip

One caveat keeps this honest. The report counts sessions where someone clicked through to a site. Most AI answers never send a click at all. The person reads the summary, gets what they need, and moves on. So the referral number understates how often your brand actually shows up, and the citation itself is the win even when no visit follows. Pair your GA4 channel with citation tracking, and judge AI search on assisted revenue and branded search lift, not sessions alone. Google's Search Console AI performance report is the other half of that picture, showing impressions inside AI Overviews and AI Mode that never turn into a click.

The headline is simple: AI referral traffic is now one channel with one name on it. Measure it that way, and the work gets clearer.

If you want help turning this into a reporting setup and a page-by-page plan for AI search, that is the kind of thing we do.