Six AI assistants reached 76 billion sessions a month worldwide in September 2026, up 37% year over year, according to Graphite's updated AI usage report based on Similarweb data. The United States accounted for 6.6 billion sessions, up 45% year over year. Just as notable as the growth is what the report no longer does: unlike its March edition, it does not compare AI usage with search engines.
Worldwide, the six AI assistants in the report reached 76 billion sessions a month in September 2026, up 37% year over year. That is a large increase on an already large base, and it suggests AI assistant use is still expanding rapidly rather than settling into a mature, slow-growth pattern.
In the United States, the figure was 6.6 billion sessions, up 45% year over year — faster growth than the worldwide average. The US is a large share of the total, but the faster US growth rate suggests the American market is still catching up in relative terms, or that US users are adopting AI assistants more quickly than users elsewhere.
ChatGPT remains the leader within that total, holding 81.2% of traffic share among the top six assistants — ChatGPT, Gemini, Claude, Grok, Perplexity and DeepSeek — and hitting a record 61.35 billion sessions in September. The other five assistants share the remaining share between them, which reinforces the picture of a market dominated by one product.
The March edition of the report compared AI usage with search engines. The updated edition does not, and Graphite says it can no longer reliably compare them. The reason given is that the March edition undercounted Google search use from the Google Android app, which is estimated at 11.8 billion monthly sessions in the US alone.
That undercount matters because it distorted the comparison. If Google search use from the Android app was missing from the earlier figures, then the earlier comparison understated search and overstated AI's share relative to search. Correcting that error would change the comparison substantially, and Graphite's response was to drop the comparison rather than publish a corrected version it could not stand behind.
There is a second problem, also stated by Graphite: there is no reliable data on what share of those Android app sessions is search. The Google Android app does more than search, so counting all of its sessions as search would overstate search, while counting only part of them requires a share figure that does not exist in reliable form. Without that share, a fair comparison cannot be made.
The report's data suggest that AI accounts for less than 1% of traffic on average to websites. That is a striking contrast with the session figures: AI assistants are used enormously, but the traffic they send to external websites is, on average, a tiny fraction of total website traffic. The gap exists because many AI sessions end inside the AI assistant, with an answer, rather than with a click to a website.
A related figure puts Google's AI presence in perspective: AI Overviews appeared in 62% of US searches. That is a high share, and it means most US searches now involve an AI-generated summary at least some of the time. The combination of high AI Overview presence and low AI referral traffic suggests AI is reshaping how people get answers, even while it sends relatively few clicks onward.
Both figures should be read as averages, not as universal rules. A site in a topic where AI answers are common may see a very different pattern from a site in a topic where AI answers are rare, and a site that is frequently cited may receive far more AI traffic than the average. The average is a starting point, not a prediction for any single site.
For SEOs, the decision to drop the AI-versus-search comparison is a useful lesson in measurement honesty. When the underlying data cannot support a comparison, the right move is to stop making it, even if the comparison was popular. SEOs should apply the same discipline to their own reporting: if a metric cannot be measured reliably, say so, rather than publishing a figure that looks precise but is not.
For link builders, the less-than-1% traffic figure is a reminder that AI visibility and AI traffic are different things. Being cited in an AI answer may build brand awareness and influence, even if it sends few clicks, while a link that drives traffic may matter more for direct visits. Both have value, but they should be measured and reported separately.
The practical takeaway is to track AI sessions, AI citations and AI referral traffic as three separate numbers, and to be cautious about any claim that compares AI use directly with search use. Graphite's own retreat from that comparison is a signal that the data is not yet good enough to support it.
Sources: Search Engine Journal, November 9, 2026, reporting Graphite's updated AI usage report based on Similarweb data for September 2026.
Related reading on this site: ChatGPT Search Is Now About One Fifth of Google's Volume, OpenAI Says, AI Traffic Is Growing 45% a Year — but Still Just 1.25 Visits per 100 From Google, ChatGPT Sends 0.62% of Google's Organic Traffic — but Converts at 13.8%, Google Organic Click Share Hits a 15-Month High as Zero-Click Searches Fall, Email and Paid Search Are Growing as Organic Search Shrinks, and Weekly AI Use Jumped From 34% to 51% in a Year, Reuters Institute Finds.
Linkbaits helps you create content that earns citations and backlinks. Start with the toolkit and build your first linkable asset this week.
Start free trial