ChatGPT Owns 92% of AI Referral Traffic — Here's What 6.77M Sessions Tell Us

A year ago, everyone had a different bet on which AI platform would dominate search discovery. The smart money was split between Perplexity (the search-native challenger) and Copilot (the enterprise Trojan horse hiding inside Microsoft 365). Neither bet paid off. Instead, a study published last week by Previsible — tracking 6.77 million LLM-driven sessions across 166 websites over 19 months — shows something more decisive than anyone expected: ChatGPT controls 92.4% of all AI referral traffic. Not a plurality. Not a comfortable lead. Near-total dominance.

If you've been spreading your "AI optimization" efforts across five different platforms thinking you're being thorough, this data should change your priorities today.

How We Got Here: The Numbers Behind the Consolidation

The Previsible data covers November 2024 through May 2026 — 166 GA4 properties spanning SaaS, e-commerce, finance, legal, health, insurance, education, publishing, and ticketing. The total volume of AI-referred monthly sessions grew 9.9x over that window, from 65,249 in November 2024 to 644,478 in May 2026.

That growth wasn't smooth. There was a violent dip in November 2025 when ChatGPT referrals dropped 50% in a single month — from 448,412 to 213,345 — apparently because the model shifted to favoring Wikipedia and Reddit for a period. The other platforms barely flinched. Sessions recovered to 442,609 by December and then kept climbing to new highs.

That November dip is actually the most important thing in the whole study. Not because it's frightening — though it is — but because of what it tells you about the architecture of your risk. One company's product decision can halve your AI referral traffic overnight. That's not a hypothetical. It happened. Build your strategy knowing that's possible.

Where the Other Platforms Ended Up

Back in December 2025, ChatGPT held about 84% share with Perplexity at 8.9%, Gemini at 4.5%, Copilot at 2.1%, and Claude at 0.6%. Six months later:

• ChatGPT: 92.4% share, 12.8x growth over 19 months, no sign of slowing.
• Gemini: Quiet #2, grew 3.2x with almost no volatility. 5,598 sessions/month → 18,119 in May 2026.
• Claude: Grew 64x (133 sessions → 8,528), overtook Perplexity in March 2026 and stayed ahead.
• Perplexity: Peaked at 17,507 monthly sessions in March 2025, now down 61% to 6,788.
• Copilot: Peaked at 8,651 sessions in August 2025. Now at 339. That's a 96% collapse.

The industry expected Copilot to win the enterprise play. It didn't — Claude appears to be taking that ground instead, driven by Claude Code and its expanding professional workflow integrations. Meanwhile, Perplexity is making a strategic pivot toward keeping users inside its own browser and agent experiences, which means it's not trying to send you traffic anymore. It's trying to replace the need to visit your site at all.

The Finding That Should Actually Change Your Site Architecture

Market share data is interesting but it's not actionable on its own. The most operationally useful finding in the whole study is this: ChatGPT sends 28.8% of its traffic to internal search results pages. Across all platforms and verticals, roughly 25% of AI-referred traffic lands on your internal search.

That's not a bug or a quirk. It's structural. ChatGPT trusts your domain enough to recommend it, but it often can't confidently identify the specific page the user needs — so it routes them to your search box and lets them navigate from there. The model did the hard part. It chose you over every other option. Then it dropped the user at your front door and said "find what you're looking for."

If your internal site search is slow, returns garbage results, or shows a 404 when it can't find something, you're wasting high-intent traffic that an AI assistant specifically sent you. Most teams treat internal search as a navigation feature. The data says it's an acquisition surface.

Industry Breakdowns: The Picture Varies Wildly by Vertical

One thing the aggregate numbers hide is how differently AI traffic behaves across industries. The fastest-growing verticals are the ones that started from almost nothing:

• E-commerce: Grew 37x from near-zero. Users are landing on product pages with purchase intent already formed — they've already asked the AI "what's the best [product]" and gotten your brand as an answer. The page they hit needs to close the deal.
• Insurance: 18.9x growth to 1.51% AI penetration — the highest rate of any vertical in the study.
• Education: 5.4x growth, with 52% of LLM traffic landing directly on course pages. Not blog posts. Not marketing pages. The actual course. Users are asking "where can I learn X" and skipping everything else.
• Health: The only vertical where AI penetration actually declined (from 0.23% to 0.17%). If you're in health, you're dealing with heavier content restrictions and brand trust filters inside LLMs.

Where Traffic Lands Tells You What Users Are Actually Asking

The study breaks down landing pages by vertical and it's genuinely illuminating:

SaaS: 34.6% of AI traffic lands on internal search pages. Same structural pattern as the overall data — the model knows your brand but can't nail the specific feature page the user wants.

Publishers: 54% of traffic goes to news/article pages, which makes sense. But here's the brutal irony — publishers produce the content LLMs train on and regularly cite, yet they capture only 0.08% of their total sessions from AI sources. You're basically running a free training data operation for AI companies while getting almost nothing back in direct referral traffic.

Legal: The most evenly spread distribution — blog (28%), about page (12%), contact (12%), location pages (10%). The AI is routing users through the full evaluation arc: what do they do, who are they, how do I reach them, where are they located.

Health: 42% of AI-referred health traffic lands on About pages. When someone asks an LLM a health question, the first thing they do after clicking through is evaluate whether the source is credible. Your About page in health isn't an afterthought — it's where trust is established or lost.

Claude Is Real Now, Not Theoretical

A lot of people still treat Claude as a rounding error. The data says otherwise. Claude grew 64x over the study period. It was basically flat through most of 2025 — hovering between 1,700 and 2,000 sessions/month — then jumped 4x in two months, hit 9,501 sessions in March, and overtook Perplexity. It's stayed ahead since.

The reason matters: Claude's growth tracks its expansion into agentic tools (Claude Code, Claude Cowork), enterprise workflows, and professional integrations. The enterprise AI discovery play that everyone expected Copilot to win? It appears Claude is taking it instead.

More importantly, Claude behaves differently than ChatGPT. Where ChatGPT and Gemini are "search-pattern models" that trust domains but approximate page selection, Claude is a "content-selection model" — it picks specific pages and over-indexes heavily on long-form educational content, guides, and research. If your content strategy relies on in-depth editorial work, Claude referrals are disproportionately valuable to you even at their current volume. Early positioning in Claude's training data and citation patterns compounds in ways that are still very open.

If you're selling to developers, technical teams, or professional services buyers, Claude visibility isn't optional anymore. It's just not measured yet.

What the Data Actually Tells You to Do

There's a tendency in SEO coverage to turn every study into a 27-step framework. The Previsible data is actually more direct than that. Here's what it points to:

1. Stop spreading yourself thin across AI platforms

ChatGPT is 92.4% of the trackable AI referral traffic. Optimize for it first. Expand to Claude and Gemini when their volume in your specific vertical justifies the investment. Right now, building elaborate Perplexity or Copilot optimization strategies is spending budget on two platforms with 61% and 96% traffic declines respectively.

2. Treat your product pages as AI entry points

E-commerce AI traffic lands almost entirely on product pages with purchase intent already formed. That means your PDPs need structured, machine-readable data — clear specs, transparent pricing, comparison-ready attributes. "Contact us for pricing" doesn't just frustrate humans. It gives AI systems nothing to summarize, compare, or recommend. If the model can't read your pricing, it either skips you or gives users an inaccurate answer — both bad outcomes.

3. Fix internal search before anything else

This is the highest-ROI change most sites can make right now. If 25% of your AI-referred traffic is hitting your search box, and that search returns poor results, you've got a leaky acquisition funnel that no amount of AI optimization will fix. Get your internal search working before you worry about optimizing your About page for LLM crawlers.

4. Segment your AI traffic by page type, not site-wide

Your overall AI penetration might be 0.3%. But your pricing page might be running 3x that. Your course catalog might be 5x. Averaging everything hides where AI traffic is actually concentrated and where you should focus. Pull your GA4 AI traffic by page type, not just by source/medium, and look for the concentrations.

Bottom Line

The AI traffic race has, at least for now, a clear winner — and it's not close. If you're an SEO practitioner or a site owner trying to figure out where to put energy, the data gives you a cleaner answer than most studies do: optimize hard for ChatGPT, watch Claude carefully, and don't let your internal site search remain an afterthought. The model is already choosing you. Make sure what happens next doesn't waste the referral.

The big unanswered question the study flags — which AI platform's traffic actually converts? — is worth tracking in your own GA4 right now. Segment your LLM sessions by source and look at conversion rates, session depth, and goal completions. That's the data that tells you whether your ChatGPT traffic is actually worth chasing, or whether Claude's smaller volume is quietly punching above its weight in pipeline generated.