Which AI Engines Actually Send Traffic? ChatGPT vs Claude vs Perplexity

ChatGPT sends roughly nine out of every ten AI referral visits. Across 6.77 million LLM-driven sessions on 166 sites from November 2024 to May 2026, Previsible's 2026 AI Traffic Report put ChatGPT at 92.4% of trackable AI referral traffic. Claude, Gemini, and Perplexity split what's left. If you're deciding where to spend GEO effort, that split is the whole game.

ChatGPT sends about 90% of AI referral traffic

Two independent GA4 datasets land in the same place. Previsible measured 92.4% across 166 properties. Trakkr's AI Search Traffic Index, which aggregates 1,429 GA4 properties and updates daily, showed 90.9% for the 30 days ending July 12, 2026. When two separate panels using real referrer data agree within two points, the number is trustworthy.

Here is where each engine sat in May 2026, straight from Previsible's session counts:

The gap is not close, and it's widening in ChatGPT's favor. Its share was around 84% in December 2025 per Previsible's earlier read, so it has gained ground even as the overall channel grew nearly 10x.

Claude is the fastest riser, and it just passed Perplexity

Claude grew 64x over the window, from 133 sessions in November 2024 to 8,528 in May 2026, and overtook Perplexity in March 2026. That's the steepest curve of any engine here. It's still small in absolute terms, under 2% of referrals, but the direction is clear. If you publish technical, developer, or research content, Claude is the engine worth watching. Our Claude Skills coverage tracks the workflow side of that ecosystem.

Why Perplexity and Copilot send less, not more

Both are shrinking as referrers. Perplexity peaked in March 2025 and has fallen 61%. Copilot collapsed 96% from its August 2025 peak, down to a few hundred monthly sessions. The pattern Previsible flags is a product decision: these tools increasingly answer in-app and keep the user inside their own surface instead of linking out. A tool can be popular and still send you almost nothing.

Gemini is the quiet number two, but usage isn't traffic

This is the trap. If you read a headline saying ChatGPT holds "52.7%" of AI traffic, that study is measuring a different thing. Similarweb's data reported by PPC Land counts visits to the AI chatbots themselves, which is usage, not referrals out to publisher sites. Those two metrics diverge hard:

Gemini is the clearest case. It draws about 27% of chatbot visits but sends only about 5% of referral clicks, because Google resolves most queries inside AI Overviews and its own results without an outbound link. When someone tells you "which AI is winning," ask whether they mean who gets used or who sends traffic. For a publisher, only the second column pays rent.

AI referral traffic is real, but it's still about 1% of the web

Keep the scale honest. AI referrals average roughly 1.08% of all website traffic, growing about a point a month, per Conductor's analysis of 3.3 billion sessions. It's a fast-growing sliver, not the main channel yet. The strategic case for GEO rests on the growth rate and the visitor quality, not on today's volume. And the true figure is usually higher than your dashboard shows, because a large share of AI traffic arrives with no referrer and gets bucketed as "direct."

How to see which engine actually sends you traffic

Default analytics undercounts this. To measure it:

• Build a custom channel grouping in GA4 that maps chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, and copilot.microsoft.com referrers to an "AI" channel. Out of the box, GA4 scatters them across Direct and Referral.
• Check both metrics separately. Track referral sessions (who sends clicks) apart from citation share (who mentions you). They move independently.
• Weight by ChatGPT first. With 90%+ of the referral channel, ChatGPT is where measurable traffic lives today. Don't split effort evenly across five engines that don't send even shares.
• Watch Claude's slope. It's the one climbing fastest. A quarterly re-check is enough.

The work that earns citations across all of them is the same underlying discipline. Front-load answers, cite sources, and build real authority, which we cover in the LLM SEO guide and the roundup of AI SEO tools. What the traffic data changes is prioritization, not tactics: optimize for the engine that actually sends the clicks, and treat the rest as upside.