Ethan Smith, CEO of Graphite.io — a company that sells SEO and content marketing services — published a study this week putting AI assistants at 56% of global search engine volume by session count. The finding is headline-worthy, though readers should factor in that Smith has a commercial stake in the story he's telling.
Smith's methodology covers five LLM platforms — ChatGPT, Gemini, Perplexity, Grok, and Claude — benchmarked against the six largest search engines. He counts 45 billion monthly AI sessions globally and 5.4 billion in the United States, the latter equivalent to roughly 34% of domestic search volume.
The real argument in the study isn't the headline number — it's about what past analyses got wrong. Web-only comparisons, the standard approach of measuring ChatGPT.com visits against Google.com traffic, miss the vast majority of AI usage because 83% of global AI sessions and 75% of U.S. sessions happen inside mobile apps. On Smith's numbers, web-only figures undercount true AI activity by a factor of four to five. That's either a long-overdue methodological correction or a convenient framing that flatters the trend — possibly both.
The search side of the ledger tells a more nuanced story than 'AI is eating Google.' Google's share of discovery-related activity has fallen from 89% in 2023 to 71% in Q4 2025, a real decline. But total discovery usage — search and AI combined — has grown 26% globally over the same period. The market is expanding, not just reshuffling.
The U.S. number is the one that should give people pause: American AI usage grew roughly 300% year-over-year through December 2025 — by far the most arresting figure in the study, and currently buried near the end. Globally, AI session counts have been roughly flat since July 2025. Whether the U.S. leads the world into another growth leg or reflects a different cultural and demographic profile is an open question, but the divergence is stark and the study doesn't dwell on it.
Within the AI segment itself, ChatGPT's concentration is extreme. Smith puts it at 89% of all global AI sessions across the platforms he measured — a dominance that rivals Google's grip on search and the kind of number that tends to attract regulatory attention eventually, whatever the broader market does.
For publishers and brands, the debate over visibility in LLM outputs — often pitched now as 'generative engine optimisation' — has become harder to wave away at these session volumes. Smith sells exactly that kind of service, which makes him a motivated narrator. But if the underlying data holds up to scrutiny, the channel is real. What remains genuinely unclear is how much of that 300% U.S. growth translates into the commercial intent that made search advertising worth $200 billion a year.