Generative engine optimization has been marketed as the next great SEO frontier. A fresh look at 12 months of Google Analytics 4 data across 100 websites and more than 145 million sessions suggests that bet is badly mispriced.
In the dataset, ChatGPT referred just 0.62% as many sessions as Google Organic. Add every other large language model and the number crawls to 0.73%. Those are not small-sample anomalies; they are the whole story across sites of different sizes.
The numbers that deflate GEO
Conversion data makes the picture sharper, not better for GEO. ChatGPT visitors converted at 13.8%, more than 4.5 times the 3.02% rate of Google Organic. But in the same sample, Google still produced 96.5% of total conversions across all channels analyzed, and Google Organic delivered 27 times more conversions than all LLMs combined.
Yes, ChatGPT visitors are high-intent. There are just not many of them. The analysis also found ChatGPT session growth stopped around August 2025, alongside the GPT-5 release, and the platform has been sending fewer visitors since.
Why a high conversion rate doesn’t save ChatGPT
A 13.8% conversion rate sounds compelling. But a small, high-converting referral source cannot anchor a performance marketing plan when it is 0.62% the size of Google Organic. Chasing LLM citations as a primary channel means optimizing for the smallest measurable slice of an already-changing search pie.
The real shift is agentic search
Google’s CEO Sundar Pichai recently said, “Search will become an agent manager.” That reframing matters more than ChatGPT mentions. Search is moving from returning blue links to executing tasks: comparing products, filling carts, booking slots and completing purchases without forcing a visit to your site.
Three concrete developments point to where SEO work should move.
- Google Universal Cart bundles products from multiple retailers into a unified checkout in Google Wallet, watching prices, promotions and restocks.
- Universal Commerce Protocol, governed by Google, Shopify and Stripe, gives AI agents an open discovery layer to compare inventory and buy natively.
- WebMCP lets websites expose interactive functions such as forms, calculators and booking schedulers directly to agents.
Intelligent eyewear is the wildcard. Google’s first devices are arriving this fall, and they could expand search from screens to real-world moments: “look at that shirt, find me one in my size.”
Where to put the effort
First, fix measurement. In the sample, 45 of 100 websites were not tracking any key event in GA4. You cannot optimize for what you are not measuring, so choose one conversion or key event that matches revenue and track it relentlessly.
Then move from ChatGPT mentions to Google’s agentic surfaces. Master Merchant Center product data so offers can be bundled into Universal Carts. Learn how products and options are presented from Merchant Center and how agentic discovery endpoints work. For WebMCP, brainstorm interactive tools an AI agent would want to use on a customer’s behalf, then run small experiments.
This does not mean abandoning content quality. Google’s ranking systems are increasingly driven by AI and predict helpfulness. Improving user satisfaction is still the strongest form of machine optimization.
GEO is not worthless, but it is a side quest. The businesses that win the next search cycle will measure conversions, feed Google’s commerce surfaces cleanly, and test agent-ready functionality before the interface becomes mainstream.
Source: Search Engine Journal
