ChatGPT drives 92% of AI referral traffic — here's why you can't plan your AI visibility around that number

What the report found

Previsible's 2026 State of AI Discovery Report, published July 6, 2026 and covered by Search Engine Land, analyzed 6.77 million sessions that arrived from standalone LLM tools across 166 GA4 properties held constant over the full 19-month window. The headline numbers:

• ChatGPT drives 92.4% of trackable standalone-LLM referral traffic — and grew 12.8x over the 19 months. Total monthly LLM sessions across the sample rose from 65,249 (November 2024) to 644,478 (May 2026), a 9.9x increase.
• Claude grew 64x (133 sessions in November 2024 to 8,528 in May 2026) and overtook Perplexity in March 2026 to become the clear #2 standalone referrer for many sites.
• Gemini is the quiet #2 by some cuts, up 3.2x with little volatility; Perplexity fell 61% from its March 2025 peak; Copilot collapsed 96% from its August 2025 peak.
• Google's AI Overviews were excluded — Previsible notes they “almost certainly” drive more AI traffic than all standalone platforms combined, but they answer in-place and aren't tracked the same way, so they aren't in these totals.

Trap one: referral clicks are not recommendations

The obvious misread is “92% is ChatGPT, so optimize for ChatGPT and ignore the rest.” But referral traffic only counts the narrow slice of AI answers that both cite a source and get clicked — and the engines differ enormously in how often they link out at all. ChatGPT's dominance of clicks is partly an artifact of surface design, not proof it's where your buyers hear your name. In our own 1,120-answer engine comparison, Claude recommended the audited business in 38% of answers versus ChatGPT's 16% — more than double. A business that plans purely around referral share would over-invest in the engine that recommends it least and under-invest in the one that recommends it most. Being recommended and being clicked are two different channels; this report measures the second one.

Trap two: the biggest AI surface isn't in the data

Previsible is admirably explicit that Google's AI Overviews and AI Mode are excluded — and that they likely dwarf every standalone tool combined. That is not a footnote; it's the whole map. Any “AI traffic” number built from GA4 referrers systematically undercounts the surface where a customer asks a question, reads the AI's answer, and never clicks anyone. For local businesses — med spas, law firms, home-services companies — that in-place answer is often the entire interaction. If the AI names a competitor and books them, you don't see a lost click in GA4; you see nothing at all. The channel that decides the outcome is precisely the one a referral report can't see.

This is the same blind spot Google's own new Search Console AI-features report leaves: it counts impressions, not whether you were the recommended answer, and only for Google's surfaces. Traffic dashboards and impression reports each instrument one sliver. Neither tells you the number that moves revenue — how often, across the engines your buyers actually use, the model names you.

The takeaway

Read the Previsible report for what it is: strong evidence that AI referral traffic is real, growing nearly 10x in 19 months, and heavily concentrated in ChatGPT's clicks. Just don't let a click-share chart set your AI-visibility strategy. Measure the thing that actually decides who gets the customer — your recommendation rate across ChatGPT, Perplexity, Gemini, and Claude, including the answers that never send a click. A free 60-second check samples ChatGPT and Perplexity (the $79 audit adds Gemini and Claude, 25 questions sampled three times each) and reports how often each one names your business right now, and if you're comparing tools to track it over time, here's how the category stacks up.