ChatGPT commands 92% of AI referral traffic. Here’s what 6.77 million sessions reveal.

Twelve months in the past, the trade was betting on which AI platform would win discovery. Perplexity regarded just like the search-native challenger. Copilot regarded just like the enterprise Malicious program. Neither guess paid off.

Previsible (disclosure: I’m its CPO and co-founder) simply printed its third AI Traffic Study, analyzing 6.77 million LLM-driven classes. The information exhibits consolidation. Month-to-month LLM classes grew 9.9x, reaching 644,478 in Might 2026. And 92.4% of that site visitors comes from one platform.

The plateau was a pause

In mid-2025, AI site visitors seemed to be peaking in some sectors. It wasn’t.

Classes rose from 65,249 in November 2024 to 396,278 by August 2025, then dropped sharply in November 2025, earlier than hitting new highs of 428,203 in February 2026 and 644,478 in Might.

That November dip wants context.

Classes fell 50% in a single month, pushed virtually fully by ChatGPT referrals dropping from 448,412 to 213,345. Different platforms held regular. This was possible a model-related change; we’ve seen modest product tweaks massively swing referral site visitors, like final fall when many websites misplaced half their ChatGPT site visitors as a result of the model began favoring Wikipedia and Reddit. Classes recovered to 442,609 by December.

The lesson: one vendor’s product selections can halve your AI site visitors in a single day. Plan for the volatility.

Consolidation, not competitors

After we final printed in December 2025, ChatGPT held roughly 84% share, adopted by Perplexity at 8.9%, Gemini at 4.5%, Copilot at 2.1%, and Claude at 0.6%. Six months later, the sphere has collapsed towards the chief.

Throughout the complete dataset, ChatGPT instructions 92.4% of trackable LLM referral site visitors, rising 12.8x over 19 months with no signal of slowing. It’s the one LLM sending significant referral quantity at scale. Optimizing for “AI visibility” with out prioritizing ChatGPT means optimizing for an abstraction.

Necessary framing: this measures standalone LLM referral site visitors. AI discovery inside Google’s personal outcomes, together with AI Overviews, virtually actually drives extra AI site visitors than all standalone platforms mixed, however it operates on a distinct measurement paradigm and is excluded right here.

The challengers flipped

The shock isn’t on the high. It’s who’s shifting beneath.

Claude

Claude grew 64x, from 133 classes in November 2024 to eight,528 in Might 2026, and overtook Perplexity in March 2026 for the primary time. It stayed forward.

Claude was flat via 2025, then accelerated 4x in two months as its agentic instruments and enterprise integrations gained adoption. The enterprise benefit the trade anticipated Copilot to win could also be materializing for Claude as an alternative.

In case your viewers contains technical patrons, builders, or skilled companies, Claude visibility is now materials, and the window for early positioning is open.

Gemini

Gemini is the quiet quantity two: 3.2x development with virtually no volatility. Its Workspace and Android integration imply referral numbers possible undercount its actual discovery footprint.

Perplexity & Copilot

Perplexity peaked at 17,507 month-to-month classes in March 2025 and has fallen 61% since. Copilot collapsed 96% from its August 2025 peak, from 8,651 classes to 339.

Neither is a development guess for site visitors acquisition anymore. Each are shifting towards retaining customers inside their very own experiences: browsers, brokers, and modes the place they don’t must ship you site visitors in any respect.

The place LLMs ship customers, and why it ought to change your roadmap

The research’s most actionable discovering isn’t market share. It’s touchdown pages.

ChatGPT sends 28.8% of its site visitors to inner search outcomes pages. Throughout industries, roughly 25% of AI-referred site visitors lands on inner search.

The mannequin trusts your area however can’t decide the correct web page, so it sends customers to your search field and lets them navigate. This sample persists throughout verticals and time intervals, suggesting it’s structural to retrieval-augmented era reasonably than a short lived quirk.

Take into consideration what which means. The mannequin did the exhausting work of selecting your area. Your inner search UX now determines whether or not that high-intent go to converts or bounces.

For many websites, inner search is a uncared for navigation characteristic, not an acquisition floor. That has to alter.

The vertical view tells a number of completely different tales:

• SaaS site visitors lands on search pages (34.6%).
• Writer site visitors lands on information pages (54%), but towards 120+ million natural classes, writer penetration is 0.11%; publishers produce the content material LLMs cite and seize virtually not one of the ensuing site visitors.
• Ecommerce site visitors lands on product pages, with buy intent already shaped.
• Schooling site visitors lands immediately heading in the right direction pages (52%), bypassing advertising and marketing content material.
• Well being site visitors lands on About pages (42.1%), with customers evaluating the supply earlier than the content material.
• Authorized site visitors spreads throughout weblog, about, contact, and placement pages: the complete analysis arc.

The platforms have personalities, too:

• ChatGPT and Gemini are search-pattern fashions: area belief, page-level uncertainty.
• Perplexity and Claude are content-selection fashions that decide particular pages and over-index on long-form.

In case your technique is determined by editorial content material driving certified site visitors, Perplexity and Claude matter disproportionately to their share.

What to do now

• Optimize for ChatGPT first. Develop elsewhere when quantity justifies it.
• Monitor Claude. It overtook Perplexity in March. Early positioning compounds.
• Deal with product pages as AI entry factors. Product pages seize 43% of e-commerce LLM site visitors. Structured, comparable product information is a discoverability requirement now.
• Make pricing machine-readable. “Contact us for pricing” provides AI techniques nothing to summarize, examine, or suggest.
• Prioritize inner search. It’s an acquisition instrument, not a navigation characteristic.
• Monitor AI site visitors by web page kind, not site-wide. Your website common hides the place AI site visitors concentrates. Your pricing web page may run 3x your site-wide penetration.

The following query is the one no one has answered: conversion charge by LLM platform. Which platforms ship customers who purchase, and which ship customers who bounce?

We constructed this dataset to reply that. If the final 19 months are any indication, the solutions will change sooner than most groups are prepared for.

Concerning the information

166 GA4 properties, November 2024 via Might 2026, spanning SaaS, ecommerce, finance, authorized, well being, insurance coverage, schooling, publishing, and ticketing. All 166 properties are current all through the complete 19-month window, so the trajectories replicate behavioral change reasonably than pattern growth.

The report

Yow will discover the full report at previsible.io.