I recently came across an intriguing Semrush study that revealed some fascinating insights into ChatGPT’s traffic patterns. Despite a whopping 206% increase in referrals, surprisingly few sites actually see significant traffic. This is largely because many queries are backed by pre-trained knowledge rather than live web searches.
According to the study, over 30% of outbound clicks go to just 10 domains. Google alone claims more than 20% of these clicks. It’s intriguing to see how much weight the tech giant holds in this landscape.
ChatGPT is gradually leaning less towards live web searches. It only triggers search functions in 34.5% of queries now, a decline from 46% in late 2024. This shift indicates a change in how the platform’s role is evolving in navigating the web.
Let me break it down further. Although ChatGPT’s referral traffic saw a significant rise, the traffic mainly flows towards a limited number of sites. In fact, about 21.6% of this traffic heads straight to Google, followed by nine other domains that make up a total of just over 30%.
Many other websites are left with a small fraction of residual traffic. The number of domains receiving any referrals peaked at around 260,000 in 2025 but has since settled near 170,000.
Why is this important for us? The visibility on ChatGPT doesn’t always translate directly into traffic. Often, the impact of referrals may seem marginal. Plus, the decline in search-triggered queries makes securing citations and traffic even more challenging.
While ChatGPT defaults to pre-trained knowledge, it resorts to web searches in certain scenarios, like when users request sources, inquire about current events, or when the model shows uncertainty.
I’ve noticed a shift in user behavior—most ChatGPT prompts don’t mirror typical search queries. Instead, between 65% and 85% reflect complex, conversational inputs, indicating a transformation in engagement. Interestingly, the number of queries per session jumped 50% in late 2025.
Looking into the data, Semrush analyzed over a billion lines of U.S. clickstream data between October 2024 and February 2026. This analysis tracked prompts, referral destinations, and patterns in search usage.
For those interested, more detailed insights can be found in the ChatGPT traffic analysis. The study, titled “ChatGPT traffic analysis: Insights from 17 months of clickstream data,” is an enlightening read.
Inspired by this post on Search Engine Land.
Leave a Reply