Here's the number that launched a thousand Slack threads: ChatGPT referred sessions equal to 0.62% of Google organic sessions across a 12-month GA4 sample of 100 websites and more than 145 million sessions, according to Marie Haynes's analysis published in October 2026. Six sessions per thousand. Add every other LLM referral source and the combined figure climbs to 0.73%.

That ratio has already been misquoted as a share of total traffic, as proof AI doesn't matter, and as a reason to pour more budget into "GEO" programs. All three readings are wrong, and the measurement problem underneath them is the part that should worry marketing ops.

The Denominator Matters More Than the Headline

The "less than 1%" figure compares ChatGPT-referred GA4 sessions to Google organic sessions. It is not ChatGPT's share of all website traffic. That distinction sounds pedantic until someone on your leadership team reads the headline and concludes either "AI is irrelevant" or "SEO is dying." Neither conclusion follows from this ratio, and both can trigger budget decisions you'll spend quarters unwinding.

Haynes's dataset didn't fully specify reporting dates, industry mix, or source-identification rules, as Search Engine Land noted when amplifying the findings. Treat it the way you'd treat any single-source benchmark: useful for calibration, dangerous as a planning input without your own data beside it.

A separate 31-website sample from Theory Road reported AI assistants sending sessions equal to 23.8% of Google organic over a 90-day window, with ChatGPT making up 92.9% of those AI sessions. The gap between 0.62% and 23.8% tells you more about sample composition than about any universal truth. Your site's ratio will depend on content type, audience, and how clean your GA4 source/medium definitions are.

Conversion Rates Look Good Until You Count Conversions

Among the 55 sites in Haynes's sample that tracked conversions, ChatGPT visitors converted at 13.8% versus 3.02% for Google organic. That's a 4.5x rate advantage. And it doesn't matter much, because Google organic still drove 27 times as many total conversions as all LLMs combined.

Haynes put it plainly: "The AI search that is actually sending people to businesses today isn't ChatGPT. It's Google."

The B2B SaaS picture is less flattering. A Usermaven study of 150 B2B SaaS sites and 15.38 million sessions found identifiable AI referrals at 0.88% of all sessions, with same-session conversion at 1.76% for AI referrals versus 2.02% for organic search. No rate advantage. Tiny volume. ChatGPT accounted for 65% of those identifiable AI referral sessions.

Small-volume conversion rates are noisy by definition. If you're modeling CAC or LTV off AI referral conversion rates today, you're building on sand.

What GA4 Can't See Is the Real Problem

Sessions attributed to chatgpt.com / referral capture only the clicks. They miss every case where a buyer reads an AI-generated answer, doesn't click through, and later searches your brand name or navigates directly. That influence is real, it affects demand, and it's invisible in your GA4 source report.

This is the same attribution gap that plagues LinkedIn and podcast measurement: the channel shapes the decision, but the click shows up somewhere else. Teams that already run brand-lift holdouts or track branded search volume as a leading indicator have the muscle to measure this. Teams relying on last-click GA4 reports will undercount AI's influence and overcredit Google organic in the same motion.

The source mix within AI referrals is shifting, too. Previsible's AI Traffic Index for SaaS shows ChatGPT at 89.9% of AI referral traffic, with Gemini at 6.8%, Claude at 2.0%, and Perplexity at 1.2%. If you're only filtering for chatgpt.com in your GA4 setup, you're missing the tail, and the tail is where early signals of a shift will appear.

The Ops Move: Instrument Before You Optimize

Before anyone writes a "GEO strategy" deck or dismisses AI traffic as noise, get the measurement right. Define source/medium values for chatgpt.com, gemini, claude, and perplexity as distinct GA4 channels. Report sessions and conversions together, because a conversion rate on 40 sessions is a curiosity, not a signal. And separate "AI referral volume" from "AI-influenced demand" in every executive slide, so leadership can make decisions without confusing the two.

The hypothesis worth testing: if we instrument AI referral sources cleanly and track branded search lift alongside referral sessions, then we'll have a directional read on whether AI visibility is generating demand we currently can't see. Success equals a repeatable monthly report that shows both volume and influence. Guardrails: don't reallocate SEO budget based on AI referral data alone until you have at least two quarters of clean, segmented data. Stop-loss: if the measurement project takes more than two sprints to implement, scope it down to ChatGPT-only and expand later.

Six sessions per thousand is a small number. But the measurement gap around those six sessions is where the real risk sits: not in what AI sends today, but in what it influences that you can't yet count.