ChatGPT vs. Perplexity vs. Gemini: Which LLMs Are Driving Real Conversions?

A data-driven look at where AI traffic is actually coming from, where it's actually going, and which platform's users are actually buying.

The pitch decks are everywhere. "AI is replacing search." "Optimize for ChatGPT or get left behind." "LLM traffic converts 10x better than organic." If you've been within five feet of a marketing newsletter in the past year, you've heard some version of it.

Most of it is wrong. Some of it is very wrong. And some of it is, surprisingly, right — but only for a specific kind of business in a specific stage of the buyer journey.

This article does something most of the breathless coverage doesn't. It looks at the actual numbers. From the largest academic study yet conducted on the question, from first-party analytics across hundreds of websites, from independent traffic measurement firms, and from the platforms themselves.

The short answer: AI referral traffic is real, it's growing fast, and it converts unusually well for certain categories. But it's also tiny in absolute volume, structurally invisible in standard analytics, and much less revolutionary than the agencies selling "GEO services" want you to believe.

The longer answer is more interesting.

Part 1 · Volume Where AI Traffic Actually Comes From

Let's start with the easy part: market share. As of early 2026, the AI referral market is dominated by one platform to a degree that's hard to overstate.

According to SE Ranking's analysis of global AI referral traffic, ChatGPT accounts for roughly 78% of all AI-driven visits to websites worldwide. Perplexity and Gemini fight for distant second and third place, with Gemini having recently overtaken Perplexity following the rollout of Gemini 3 in late 2025. Statcounter's April 2026 data put Gemini at 8.65% of AI chatbot referrals against Perplexity's 7.07%.

Claude, despite enormous mindshare in technical and enterprise circles, sends almost nothing in measurable referral terms. SE Ranking's global data put Claude at 0.17% of AI referral traffic. That number has been climbing — Statcounter saw it rise nearly tenfold from April 2025 to March 2026 — but in absolute terms it's still a rounding error.

The headline finding from Similarweb's 2026 generative AI report is even starker: AI platforms in aggregate now account for roughly 0.24% of global web traffic. Yes, AI referral volume nearly doubled between 2025 and 2026, but the starting baseline was so small that even rapid growth has produced a channel that, for most websites, contributes less than 1% of total visits.

Part 2 · Retention AI is Built to Answer, Not to Route

The fundamental tension in any conversation about "AI traffic" is that AI assistants are explicitly engineered to keep users on the assistant. ChatGPT's web search integration, Gemini's AI Mode in Google, and Perplexity's citation-rich answer cards all exist to satisfy a query without sending the user anywhere. When that succeeds, no website gets a click — and no analytics platform records a referral.

Pew Research found that fewer than 1% of users click on links surfaced inside Google's AI Overviews. SE Ranking's data shows organic click-through rates collapsing from 1.41% to 0.64% on informational queries when an AI answer appears. Adobe Analytics measured a 1,200% surge in AI-sourced traffic to U.S. retail sites in early 2025, then watched the channel plateau by year-end as the platforms got better at answering questions in-place.

Gemini is the most extreme case. Google's AI Mode now answers an enormous share of queries directly inside the search interface, where the user never leaves Google's properties. Some analysts describe this as "attribution collapse" — the discovery is happening, but it's happening invisibly, and your analytics will never see it.

This matters because it changes what "conversion rate" actually measures in the AI context. A high conversion rate from a tiny pool of clicks isn't necessarily evidence that AI traffic is high quality. It might just mean the only people who clicked through were people who were already going to buy. Hold that thought.

Part 3 · Conversions Who Actually Buys?

The Bullish Case

The bullish case comes from First Page Sage, an SEO agency that analyzed anonymized data from 150+ B2B and B2C client companies between May 2025 and April 2026.

Seer Interactive ran a similar analysis on a single B2B software client between October 2024 and April 2025 and found ChatGPT converting at 15.9% versus Google organic's 1.76%. Ahrefs, looking at their own first-party data, reported that AI search traffic accounted for 0.5% of their total website visits but drove 12.1% of all signups — a 23x conversion advantage. Microsoft Clarity, looking across more than 1,200 publisher sites, measured AI traffic converting to signups at 1.66% versus 0.15% from organic search, an 11x premium.

These numbers look spectacular. Some of them really are. But all of them share two characteristics worth holding onto.

First, they're measured against businesses that sell information products, software, or content-heavy services. A user who arrives at Ahrefs from a ChatGPT conversation about "best SEO tools" is not in the same buying state as a random Google visitor. They're at the bottom of the funnel, already comparing options, with their wallet halfway out. Of course they convert at 23x the rate of cold organic traffic. This is intent compression, not channel magic.

Second, they're measured against the small slice of AI-referred traffic that is actually attributable. Roughly 70% of AI referral traffic arrives without referrer headers — invisible to Google Analytics, misclassified as "direct," and never counted in any conversion calculation. The studies above are reporting conversion rates on the visible 30%.

The Bearish Case

The bearish case comes from the largest and most rigorous analysis published to date: a working paper by Maximilian Kaiser of the University of Hamburg and Christian Schulze of the Frankfurt School of Finance & Management, released in October 2025 and currently under peer review. If you only read one study on this topic, read theirs.

Kaiser and Schulze also found that ChatGPT accounts for more than 90% of all e-commerce traffic originating from LLMs, with Perplexity at 4.1%, Gemini at 2.6%, Microsoft Copilot at 2.1%, and DeepSeek and Grok at statistically irrelevant levels. So when people say "AI traffic" in an e-commerce context, they almost always mean ChatGPT traffic. The "multi-LLM optimization" market is, for now, mostly theoretical.

How Can Both Be True?

They're measuring different things. Kaiser and Schulze measured unselected, average-case e-commerce traffic across 973 sites of varying sizes, categories, and brand strength. The agency studies measured self-selected traffic from clients who had already invested in generative engine optimization, often in B2B and information-product categories where buyer intent is naturally compressed at the click stage.

Both findings can coexist. In the average, AI referral traffic underperforms. In the right category, with the right content, with the right audience, it dramatically outperforms. The question isn't "does AI traffic convert?" — it's "does your category make AI traffic worth optimizing for?"