A new study published in Cyberpsychology, Behavior, and Social Networking has found that while artificial intelligence receives significant public attention, it is used far less frequently in day-to-day online activity than many might assume. Drawing from more than 14 million website visits, researchers found that AI-related browsing made up less than one percent of online activity for most people. The study also indicates that individuals who use AI more often tend to exhibit certain aversive personality traits.
There is growing public and academic interest in artificial intelligence, particularly its expanding role in education, work, and entertainment. While many surveys have asked people whether they use AI and how they feel about it, these self-reports are often unreliable. People tend to misjudge or misremember how often they use technology, especially newer forms like AI.
To move beyond self-reported data, the researchers set out to directly measure AI use through people’s actual web browsing histories. Their goal was to determine how often AI tools are used in everyday life, identify who uses them the most, and examine what types of online activities are typically associated with AI use. Understanding the psychological traits linked to higher levels of AI interaction could help researchers and policymakers anticipate how different groups may adopt or resist AI technologies in the future.
“There’s been enormous public discussion about AI and its societal impact, but surprisingly little objective data on how people actually use it in their everyday browsing,” said study author Emily McKinley, a PhD candidate at the University of California, Davis.
“Despite widespread concerns and excitement about tools like ChatGPT, we had almost no baseline understanding of actual usage patterns. We wanted to measure what’s really happening, examining not just usage frequency, but also the psychological profiles of AI adopters and how AI integrates into their broader digital behaviors.”
The project included two separate studies. The first involved 499 university students from two institutions, while the second focused on 455 members of the general public. In both cases, participants shared their web browsing history over a period of up to 90 days. Only those who used Google Chrome were included, as this browser allowed for the necessary data export. Participants also completed surveys measuring their personality traits, attitudes toward AI, and demographics.
Using a list of well-known AI websites, such as ChatGPT and Microsoft Copilot, the researchers identified which browser visits were AI-related. Other websites were categorized using a content classification system powered by a large language model. The researchers then analyzed the data to understand the proportion of AI visits relative to total browsing, what kinds of websites were visited immediately before and after using AI, and which psychological traits correlated with AI use.
In the student sample, AI use made up just 1 percent of all website visits on average. Most participants rarely used AI at all, and only a small number accounted for the majority of the traffic. The most commonly visited AI site by far was ChatGPT, representing over 85 percent of all AI-related visits. While this use rate was higher than visits to web versions of some social media platforms like Instagram, it was far below the rate of search engine use.
“We were genuinely surprised by how infrequent AI use was, even among students who typically serve as early adopters of emerging technologies,” McKinley told PsyPost.
Students who used AI more often were slightly more likely to score high on personality traits associated with narcissism and psychopathy. These individuals also tended to have more positive attitudes toward AI in general. There were some weak associations between AI use and demographics, such as income and gender, but age and ethnicity were not significantly related to AI use in this group.
In contrast, the general public sample showed even lower rates of AI use, with an average of just 0.44 percent of website visits being AI-related. Once again, ChatGPT was the most visited AI platform. Fewer significant correlations between personality and AI use were observed in this group, but there was a modest relationship between Machiavellianism and AI browsing. Those who expressed more favorable views of AI were also somewhat more likely to use it.
The researchers paid particular attention to individuals they called “prolific users,” defined as those whose AI browsing accounted for more than 4 percent of their total website visits. Among the student group, these prolific users scored much higher on measures of Machiavellianism, narcissism, and psychopathy compared to their peers. These patterns were less clear in the general public sample, possibly because AI use was less frequent overall, reducing the ability to detect meaningful differences.
“Interestingly, people who use AI more tend to score higher on aversive personality traits, particularly Machiavellianism, narcissism, and psychopathy, though these patterns were stronger among students,” McKinley said.
In both groups, the researchers also analyzed what participants were doing online in the seconds before and after visiting an AI website. Before using AI, many were on internet and telecom sites, such as search engines and login pages. After AI use, participants were more likely to visit websites related to education, computers, or professional tasks. These patterns suggest that AI tools are often used as part of a workflow, especially in academic or job-related contexts. The researchers argue that this may point to AI being seen more as a productivity tool than a source of entertainment.
Another key finding involved the gap between self-reported and actual AI use. In the general public sample, participants were asked how often they believed they used AI. The correlation between their estimates and their actual browsing data was moderate, suggesting that self-reports alone are not a reliable way to assess AI use. This result aligns with previous research on other forms of media, where people often underreport or overreport their habits.
While this study provides rare objective data on AI use, there are still some limitations. First, the analysis was limited to web-based interactions. Mobile app use, which may be more common for some users, was not included. Similarly, only Google Chrome users could participate, which may have influenced the sample.
The research also focused primarily on a narrow set of individual traits, such as personality and basic demographics. Other factors, such as emotional well-being, motivations, or social environments, could play an important role in AI adoption but were not examined here. The team suggests that future studies should explore how these additional psychological and social variables relate to AI use.
The researchers also note the importance of understanding what people are doing during their time on AI platforms. While this study captured how often users visited AI websites, it could not track whether they were writing essays, solving problems, or simply exploring out of curiosity. Capturing the content of interactions could shed light on the goals and intentions behind AI use.
Finally, as AI becomes more integrated into daily life, usage patterns may change. The researchers plan to continue this line of inquiry by examining whether AI use is linked to specific outcomes.
“We want to understand not just how often people use AI, but what they’re using it for and how that content relates to their individual characteristics,” McKinley explained. “We’re also interested in examining the downstream consequences of AI use: for example, does actual usage predict outcomes like academic integrity, information-seeking behaviors, or work performance?”
“This research represents one of the first attempts to objectively quantify AI use in naturalistic settings using months of actual browsing data. While AI currently dominates many of our cultural conversations, the data suggest most people are still using it quite sparingly. The methodological approach we developed, combining passive digital trace data with psychological measures, offers a valuable template for future technology research, helping us move beyond the limitations of self-report and understand how individual differences shape technology adoption in the real world.”
The study, “Evaluating Artificial Intelligence Use and Its Psychological Correlates via Months of Web-Browsing Data,” was authored by Emily McKinley, David M. Markowitz, Rui Zhu, and Brandon Van Der Heide.