ChatGPT Drives Just 0.62% of Google's Organic Traffic Across 100 Sites

By Amara Osei

ChatGPT sessions amounted to just 0.62% of the volume driven by Google Organic across 100 websites tracked over 12 months, according to an analysis by search consultant Marie Haynes. The dataset covered more than 145 million sessions in GA4 (Google Analytics 4). Include every other LLM (large language model) and the figure rises only to 0.73%.

Haynes published the findings in an essay titled "GEO Is A Distraction," arguing that the industry's obsession with generative engine optimization misses where return on investment actually sits.

"That's not a typo. 0.62%," she wrote.

Do ChatGPT visitors convert better?

Yes — but the base is tiny. Of the 100 sites, 55 tracked conversions in GA4. ChatGPT visitors converted at 13.8%, against 3.02% for Google organic visitors. That is more than 4.5 times the rate. Yet Google still drove 96.5% of total conversions across all channels analyzed, and Google organic produced 27 times more conversions than all LLMs combined.

The takeaway: high intent exists in LLM referrals, but the volume does not.

Did the growth curve hold?

No. Haynes writes that when ChatGPT launched, many practitioners extrapolated exponential traffic growth and predicted LLMs would eventually outrank Google as a referral source. "We were wrong," she said.

Using Antigravity, she calculated collective ChatGPT visits across the 100 properties. In August 2025, coinciding with the release of GPT-5, the exponential growth of ChatGPT sessions ended. ChatGPT has sent fewer and fewer visitors to websites since then.

One sentence captures the caveat: the trend is measured over a small, self-selected dataset, and past plateaus do not guarantee future ones.

Why does Haynes call GEO a distraction?

Because the AI search actually sending people to businesses today runs inside Google, not ChatGPT, she argues. AI Overviews, AI Mode, and a core ranking system now largely driven by AI mean millions of people use AI search daily without leaving Google.

Haynes questions the money flowing to companies that track brand mentions in LLM tools. "It boggles my mind how much money has been raised and earned by companies that track your mentions in ChatGPT and other LLM tools," she wrote.

Her practical concern is measurement. Rank tracking means little, she says, and Search Console offers little that is immediately actionable. In her sample, 45 of the 100 sites were not tracking any key events at all. Her recommendation: define what a conversion looks like for your business, measure it, and focus on improving user satisfaction rather than optimizing for machines, since Google's ranking systems use deep learning to predict what searchers find most helpful.

What did Google's leadership say about the agentic web?

Google CEO Sundar Pichai recently said that "Search will become an agent manager" — a shift from passive link-finding to active work coordination.

At Google I/O, Sergey Brin was asked what the web will look like in 10 years. "I don't think we really know what the world will look like in 10 years," he said. Demis Hassabis added: "The web is going to change quite a lot if you think about an agent-first web, like it doesn't necessarily need to see renders and things like we do as humans using the web. So I think things will be pretty different in a few years."

Where does Haynes see the real opportunity?

She names four developments:

• Universal Cart. Aggregates products from multiple retailers into a unified checkout within Google Wallet, monitoring price changes, promotions and restocks. Shoppers can buy across vendors without visiting their sites. The play: master Merchant Center data, where a new AI assistant is available.
• Universal Commerce Protocol (UCP). Governed by Google, Shopify and Stripe, it gives AI agents an open discovery layer to compare inventory and buy natively. The play: technical SEO expands to agentic discovery endpoints, including /.well-known/ucp. Reporting will be hard, she warns, because no two searches will produce the same results.
• WebMCP (Model Context Protocol). Lets websites expose forms, calculators and booking schedulers to AI agents via declarative markup and APIs. Haynes calls it too early for most businesses to build production tools, but not too early to experiment.
• Intelligent Eyewear. Google's first Gemini-powered devices running Android XR — audio frames with Warby Parker and Gentle Monster, then micro-LED display glasses — arrive this fall. Skeptics say they will not take off; Haynes thinks they will expand rather than replace search.

She also shares a reusable prompt for businesses to map these opportunities against their own products, conversion goals and ambient "See it, Solve it, or Buy it" use cases.

Her closing argument is blunt: chasing ChatGPT citations optimizes for the wrong target, and the businesses that study how Google's AI surfaces represent their brand — while preparing for UCP, WebMCP and eyewear — will hold the advantage as the agentic web arrives.