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.

In the authors' own words: "Results contradict widespread expectations of LLM superiority." ChatGPT traffic is "clearly relevant to users" but "does not yet translate into comparable sales outcomes."

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?"

Part 4 · Platforms Who Each Engine Actually Serves

🟢 ChatGPT — the Volume Play

ChatGPT is the volume play. It accounts for somewhere between 64% and 90% of AI referral traffic depending on which study you read, sends the most absolute clicks by an enormous margin, and reaches roughly 700 million weekly active users globally as of mid-2025. Its users tend to engage in long, conversational queries — averaging 23 words per query versus 4 for traditional Google search.

ChatGPT now drives 20% of Walmart's referral traffic, nearly 15% of Target's, and 10% of eBay's. The First Page Sage industry breakdown shows ChatGPT shoppers concentrated in travel & hospitality (18%), retail and CPG (16%), IT services (14%), lifestyle & health (13%), food & beverage (13%), and home services (12%).

The downside is consistency. ChatGPT's recommendations vary noticeably between sessions and have shifted dramatically over time as OpenAI tunes citation weights — the platform's referral traffic dropped 52% in a single week in July 2025 after OpenAI adjusted its citation system to prioritize Reddit and Wikipedia over commercial sites. If you build a business around ranking inside ChatGPT, you're building on shifting ground.

🔵 Perplexity — the Precision Play

Perplexity is the precision play. It sends a small fraction of ChatGPT's volume but consistently shows the highest revenue per session and the highest average order value among the three major platforms. Some analyses have found Perplexity delivers 57% higher AOV than ChatGPT despite ChatGPT's overwhelming volume advantage in the same dataset.

This isn't an accident. Perplexity is structurally a citation-first platform: every answer comes with sources, users routinely click into those sources to verify, and the people who use Perplexity skew toward research-heavy decisions. They're shopping for B2B software, professional services, technical equipment, and considered purchases — not impulse items.

For a business selling considered purchases to research-heavy buyers, Perplexity may be the highest-value AI channel per visit, even though it's nowhere near the highest-volume one.

🟡 Gemini — the Distribution Play

Gemini is the distribution play, and it's the most strategically interesting platform of the three right now. Gemini grew 157% in monthly visits between April and September 2025, then more than doubled its referral volume between October 2025 and January 2026 — a 115% surge in two months — following the Gemini 3 rollout.

But Gemini converts the worst of the three majors. First Page Sage measured Gemini referral conversion at around 3%. Seer Interactive's B2B client data showed the same. The reason is structural: Google's AI Mode answers most queries inside the search interface, so the small slice of users who actually click through to a website tend to be the ones whose questions weren't fully answered — a less-committed cohort by definition.

There's a Google-specific irony here. Gemini's 388% year-over-year growth in referral traffic to external sites versus ChatGPT's 52% suggests Google is, for now, more willing to send users elsewhere than OpenAI is. Whether that holds as Google's monetization strategy evolves is an open question.

Part 5 · Implications What the Data Means for Marketing Decisions

Part 6 · Answer So Which LLM Drives Real Conversions?

If you sell information products, B2B software, or considered purchases to research-heavy buyers: Perplexity delivers the highest per-visit value despite low volume, and Claude — despite sending almost nothing — produces the highest reported conversion rates of any AI platform when its users do arrive.

If you sell consumer goods at scale and need volume more than precision: ChatGPT is the only AI platform with enough referral mass to matter, and its users are meaningfully more engaged than typical Google visitors.

If you operate inside Google's ecosystem and care about being cited in answer boxes regardless of click-through: Gemini is the platform whose distribution is growing fastest and whose visibility increasingly determines whether you exist in the AI-mediated discovery layer at all — even though its referral conversion rate is the worst of the three.

If you sell average-case e-commerce to a mass audience without strong brand equity or category authority: none of them are sending you meaningful revenue today, and the largest first-party academic study to date suggests they won't reach parity with organic search for at least another year.

The companies that are winning this transition aren't the ones chasing the highest reported AI conversion rate. They're the ones treating LLM visibility as a discovery and influence layer that runs parallel to traditional channels rather than replacing them — building entity authority across multiple platforms, making sure their brand shows up in the citations and source lists AI assistants pull from, and accepting that for the next 12 to 24 months, AI referral traffic will be a small but disproportionately valuable slice of their funnel rather than a dominant channel.

The agencies selling you "AI search optimization services" want you to believe the channel is bigger than it is and that the conversion advantages are universal. They're not. The academics studying it from first-party data want you to understand that the hype has run ahead of the numbers. They're right too. Both things can be true at once.

The smart move isn't to pivot your marketing budget to LLM optimization. It's to treat AI visibility the way you treated mobile in 2010 or video in 2014 — as a real, growing, structurally important channel that you start investing in early but don't bet the company on until the data tells you to.