SEO consultant Marie Haynes has published 12 months of Google Analytics data from 100 websites showing ChatGPT sends just 0.62% as many sessions as Google organic search — 0.73% when all large language models are counted. Writing in Search Engine Journal, Haynes found that while ChatGPT visitors converted at 13.8% versus 3.02% for Google organic, Google still drove 96.5% of all conversions across the channels analyzed, or 27 times more than all LLMs combined. She argues the industry’s fixation on generative-engine optimisation is a distraction from where the real return lies: Google’s own AI surfaces and the coming agentic web.
The dataset covers more than 145 million sessions across 100 websites of varying sizes. The headline figure — “That’s not a typo. 0.62%,” Haynes writes — lands as a corrective to years of predictions that chatbots would siphon search traffic. Haynes notes that ChatGPT’s exponential growth in website referrals actually ended in August 2025, around the release of GPT-5, and has been declining since.
The trend data is as striking as the snapshot. Haynes had Antigravity tally ChatGPT referrals across the 100 properties over time and found the exponential growth ended in August 2025; since then, ChatGPT has sent websites fewer and fewer visitors. The chatbots-are-eating-search narrative, in other words, peaked more than a year ago.
Conversions Tell A Different Story
Of the 100 sites, 55 tracked conversions in GA4, and there the picture is more nuanced. ChatGPT visitors converted at 13.8%, more than 4.5 times the 3.02% rate of Google organic visitors — suggesting that when people do arrive from an AI assistant, they arrive with intent. But volume dominates: Google organic still drove 96.5% of total conversions, and in Haynes’s sample, 27 times more conversions than all LLMs combined.
Where The Real Opportunity Is
Haynes’s argument is not that AI doesn’t matter for search — it is that the AI search actually sending people to businesses today is Google’s, not ChatGPT’s. Between AI Overviews, AI Mode and a core ranking system now largely driven by AI, she writes, millions of people use AI search every day without ever leaving Google. The opportunity she points to is agentic: Google’s Universal Cart and the Universal Commerce Protocol for agent-driven purchasing, WebMCP for exposing site functions to AI agents, and Gemini-powered intelligent eyewear arriving this fall.
She is particularly bullish on intelligent eyewear. Gemini-powered Android XR glasses could turn everyday moments — spotting a shirt, cooking a recipe — into search queries, expanding search from screens into the physical world. “Hey Gemini — look at that shirt she’s wearing, find me one in my size and buy it for me” is the kind of ambient commerce query no chatbot tab will ever see.
Why It Matters
For the GEO industry — the AI-visibility tracking tools that have raised serious money measuring brands’ presence inside chatbot answers — this data is an awkward mirror. If chatbots drive under 1% of the sessions Google does, the budgets flowing into LLM-mention tracking are chasing a rounding error while the agentic commerce layer gets built on Google’s rails. Haynes’s prescription is unfashionable but practical: master Merchant Center data, experiment with agentic discovery endpoints, and optimise for the machines that actually send customers. It is a direct challenge to the AEO-versus-SEO framing that has dominated brand marketing this year.
The conversion data also complicates the dismissal. A 13.8% conversion rate means LLM-referred visitors, however few, are disproportionately valuable — the kind of traffic worth understanding even at small scale. The question is whether chasing ChatGPT citations is the way to get more of it, or whether, as Haynes argues, that effort is better spent on Google’s AI surfaces where the volume lives.
Background
Marie Haynes is a long-standing search-industry analyst known for research on Google’s algorithm updates and quality systems. Her study arrives as the search industry debates how to measure success in an era of AI answers: rank tracking means little when no two searches return the same results, and Search Console offers limited visibility into AI Mode and AI Overviews performance. She notes that 45 of the 100 sites she sampled were not tracking any key events at all.
Source: Search Engine Journal
0 Comments