ChatGPT captures 92% of AI referral traffic from 6.77M sessions

Twelve months ago, the conventional wisdom pointed toward a fragmented future for AI-powered discovery. Perplexity was hailed as the search-native disruptor. Copilot was positioned as the enterprise gateway. Neither prediction materialized.

Previsible, where I serve as CPO and co-founder, has released its third AI Traffic Study, drawing on 6.77 million LLM-driven sessions. The findings reveal a clear pattern of consolidation. Monthly sessions originating from large language models grew by a factor of 9.9x, reaching 644,478 in May 2026. Critically, 92.4% of that volume flows from a single source.

The plateau was a pause, not a peak

By mid-2025, there were signs that AI traffic had topped out in certain sectors. That turned out to be a false signal. Sessions climbed steadily from 65,249 in November 2024 to 396,278 by August 2025. A sharp drop followed in November 2025, but the trajectory resumed its upward path, hitting 428,203 in February 2026 before surging to the May 2026 record.

That November decline demands scrutiny. Sessions fell by 50% in a single month, a plunge driven almost entirely by ChatGPT referrals dropping from 448,412 to 213,345. Other platforms remained stable. The culprit was likely a model-level adjustment; we have observed how modest product changes can dramatically shift referral patterns. A similar event occurred last fall when many sites lost half their ChatGPT traffic after the model began favoring Wikipedia and Reddit. Sessions recovered to 442,609 by December.

The takeaway is stark: a single vendor’s product decision can cut your AI traffic in half overnight. Plan for volatility.

Consolidation, not competition

When we last analyzed the data in December 2025, ChatGPT held roughly 84% share, followed by Perplexity at 8.9%, Gemini at 4.5%, Copilot at 2.1%, and Claude at 0.6%. Six months later, the field has collapsed toward the leader.

Across the full dataset, ChatGPT now commands 92.4% of trackable LLM referral traffic, growing 12.8x over 19 months with no signs of slowing. It remains the only LLM sending meaningful referral volume at scale. Optimizing for “AI visibility” without prioritizing ChatGPT means optimizing for an abstraction.

A critical framing: this measures standalone LLM referral traffic. AI discovery within Google’s own results, including AI Overviews, almost certainly drives more total AI traffic than all standalone platforms combined, but it operates under a different measurement paradigm and is excluded here.

The challengers flipped

The real surprise is not at the top. It is who is moving underneath.

Claude

Claude experienced a 64x growth, rising from 133 sessions in November 2024 to 8,528 in May 2026. It overtook Perplexity in March 2026 for the first time and has stayed ahead. Claude was flat through 2025, then accelerated 4x in two months as its agentic tools and enterprise integrations gained traction. The enterprise advantage the industry expected Copilot to claim may be materializing for Claude instead. If your audience includes technical buyers, developers, or professional services, Claude visibility is now material, and the window for early positioning is open.

Gemini

Gemini is the quiet number two, showing 3.2x growth with almost no volatility. Its integration with Workspace and Android means its referral numbers likely undercount its true discovery footprint.

Perplexity & Copilot

Perplexity peaked at 17,507 monthly sessions in March 2025 and has fallen 61% since. Copilot collapsed 96% from its August 2025 peak, dropping from 8,651 sessions to 339. Neither platform is a viable growth bet for traffic acquisition anymore. Both are shifting toward keeping users inside their own experiences: browsers, agents, and modes where they do not need to send you traffic at all.

Where LLMs send users, and why it should change your roadmap

The most actionable finding in the study is not market share. It is landing pages.

ChatGPT sends 28.8% of its traffic to internal search results pages. Across industries, roughly 25% of AI-referred traffic lands on internal search. The model trusts your domain but cannot pick the right page, so it sends users to your search box and lets them navigate. This pattern persists across verticals and time periods, suggesting it is structural to retrieval-augmented generation rather than a temporary quirk.

Think about what that means. The model did the hard work of choosing your domain. Your internal search UX now determines whether that high-intent visit converts or bounces. For most sites, internal search is a neglected navigation feature, not an acquisition surface. That has to change.

The vertical view tells several distinct stories:

SaaS traffic lands on search pages (34.6%).

Publisher traffic lands on news pages (54%), yet against 120+ million organic sessions, publisher penetration is just 0.11%; publishers produce the content LLMs cite and capture almost none of the resulting traffic.

Ecommerce traffic lands on product pages, with purchase intent already formed.

Education traffic lands directly on course pages (52%), bypassing marketing content.

Health traffic lands on About pages (42.1%), with users evaluating the source before the content.

Legal traffic spreads across blog, about, contact, and location pages: the full evaluation arc.

The platforms also have distinct personalities:

ChatGPT and Gemini are search-pattern models: they trust the domain but show page-level uncertainty.

Perplexity and Claude are content-selection models that pick specific pages and over-index on long-form formats. If your strategy depends on editorial content driving qualified traffic, Perplexity and Claude matter disproportionately to their share.

What to do now

Optimize for ChatGPT first. Expand elsewhere when volume justifies it.

Monitor Claude. It overtook Perplexity in March. Early positioning compounds.

Treat product pages as AI entry points. Product pages capture 43% of e-commerce LLM traffic. Structured, comparable product data is a discoverability requirement now.

Make pricing machine-readable. “Contact us for pricing” gives AI systems nothing to summarize, compare, or recommend.

Prioritize internal search. It is an acquisition tool, not a navigation feature.

Track AI traffic by page type, not site-wide. Your site average hides where AI traffic concentrates. Your pricing page might run 3x your site-wide penetration.

The next question is the one nobody has answered: conversion rate by LLM platform. Which platforms send users who buy, and which send users who bounce? We built this dataset to answer that. If the last 19 months are any indication, the answers will change faster than most teams are ready for.

About the data

The study covers 166 GA4 properties from November 2024 through May 2026, spanning SaaS, ecommerce, finance, legal, health, insurance, education, publishing, and ticketing. All 166 properties are present throughout the full 19-month window, so the trajectories reflect behavioral change rather than sample expansion.

The report

The full report is available at previsible.io.