Perplexity AI Agent Adoption Reaches 57%, Driven by Digital Sector and Higher GDP, Study Reveals

The increasing availability of AI agents raises critical questions about how people integrate these tools into their daily lives, and a new study sheds light on early adoption patterns. Jeremy Yang, Noah Yonack, and Kate Zyskowski from Perplexity, alongside Denis Yarats, Johnny Ho, and Jerry Ma, present the first large-scale analysis of AI agent usage, focusing on Perplexity’s Comet Assistant. Their work examines hundreds of millions of anonymized user interactions to determine who adopts these agents, how frequently they use them, and for what purposes. The results reveal a diverse user base with strong correlations between adoption and factors like socioeconomic status and professional background, while a detailed taxonomy of use cases highlights a preference for productivity, learning, and personal tasks, suggesting significant implications for the future of work, education, and information access. Comet AI Agent Capabilities and Early Usage This report details the development, validation, and early adoption of Comet, an AI agent designed to assist users with a variety of tasks. It examines the agent’s capabilities, how users are employing it, the characteristics of early adopters, and technical validation of its classification system. The research combines quantitative data, such as survey results and classifier accuracy, with qualitative observations of the agent’s behaviour to provide a comprehensive understanding of its performance. Comet is designed as an agentic AI, proactively breaking down tasks into manageable steps, searching for information, and executing actions on behalf of the user. Common use cases include assistance with shopping, social media interaction, entertainment, research, document creation, task management, job searching, and educational courses. For example, Comet successfully completed a complex task involving finding a round-trip flight for a conference, considering preferences for late-night, direct flights, and price, demonstrating its ability to create to-do lists, search for information, interact with websites, apply filters, and adapt to constraints. The typical early adopter is male, over 35 years old, works full-time in the technology industry, and currently uses or is interested in Perplexity. These users employ Comet for both personal and professional tasks. Researchers anticipate the user base will evolve as the product matures and reaches a wider audience. The report also validates a classifier designed to categorize agentic queries into topics, subtopics, tasks, and usage contexts, crucial for understanding how the agent is used and improving its performance. A dataset of 1,000 agentic queries, manually labeled and reviewed by annotators, achieved high agreement rates with the golden dataset, demonstrating the classifier’s accuracy across topic, subtopic, task, and usage context. Key findings demonstrate the potential of AI agents to proactively assist users with complex tasks, highlighting a tech-savvy user base and an accurate agent use case classifier. The research emphasizes the importance of manual labeling for creating high-quality datasets crucial for training and validating AI models, and demonstrates that task accuracy depends on accurate topic and subtopic classification. Real-World AI Agent Usage Analysis This research presents a large-scale analysis of how users interact with AI agents in real-world web environments, focusing on Comet, an AI-powered browser and its integrated assistant. Researchers meticulously collected and analyzed data from millions of users and hundreds of millions of queries spanning July 9 to October 22, 2025, to understand adoption, usage intensity, and specific use cases. They defined a Comet user as someone making at least one query during the study period, then refined the dataset by excluding enterprise users, those participating in specific programs, and users who deleted accounts or opted out of data retention. To isolate genuine user intent, the team defined “agentic queries” as those requiring the agent to actively control the browser or take actions on external applications via the Model Context Protocol or direct API calls, distinguishing them from simple information exchange. They removed demonstration queries shown during user onboarding and excluded instances where a single query triggered multiple browser actions, ensuring a clear understanding of user-initiated tasks. This detailed methodology enabled the team to build a hierarchical agentic taxonomy, organizing use cases across topics, subtopics, and tasks, and ultimately reveal substantial heterogeneity in how different user segments engage with AI agents. AI Agent Adoption Mirrors Global Digital Divide This research presents the first large-scale study of how people are adopting and using AI agents within open-world web environments, focusing on Comet, an AI-powered browser developed by Perplexity and its integrated assistant. Analysis of hundreds of millions of anonymized user interactions reveals substantial differences in adoption and usage across various user segments. The study demonstrates that early adopters, users in countries with higher GDP per capita and educational attainment, and individuals in digital or knowledge-intensive sectors, including digital technology, academia, finance, marketing, and entrepreneurship, are significantly more likely to adopt and actively use the agent. Researchers measured sustained growth in both agent adoption and usage intensity, with the period following general availability accounting for 60% of all agent adopters and 50% of agentic queries throughout the study period. Notably, early Comet adopters generated a disproportionately large share of both adoption and queries compared to their overall user representation. At the country level, adoption and usage intensity exhibited strong positive correlations with both GDP per capita and average years of education, confirming a link between socioeconomic factors and AI agent engagement. To systematically categorize how users interact with the agent, the team developed a hierarchical agentic taxonomy, organizing use cases across three levels: topic, subtopic, and task. Results show that Productivity and Workflow and Learning and Research account for 57% of all agentic queries, while Courses and Shopping for Goods comprise 22%. The top 10 tasks, out of a total of 90, represent 55% of all queries. Personal use constitutes the majority of queries at 55%, with professional and educational contexts accounting for 30% and 16% respectively. Analysis of query patterns reveals that while use cases demonstrate strong initial stickiness, users tend to shift toward more cognitively oriented topics over time. These findings provide valuable insights into the evolving relationship between users and increasingly capable AI agents. AI Agent Adoption And User Patterns This research presents the first large-scale analysis of how people are adopting and using AI agents within everyday web browsing. By examining hundreds of millions of interactions with the Comet assistant, scientists have identified key patterns in user behaviour and the types of tasks these agents are employed for. The study demonstrates sustained growth in both the number of users and the intensity of their engagement with AI assistance, particularly following general availability of the technology. The findings reveal that early adopters, and those in countries with stronger economies and higher levels of education, are driving much of this growth. Furthermore, individuals working in digital fields, academia, finance, marketing, and entrepreneurship represent a disproportionately large share of agent users. Analysis of user queries reveals a clear hierarchy of use cases, with productivity and learning tasks dominating, specifically courses and shopping. Personal use currently accounts for the majority of queries, though professional and educational applications are also significant. Importantly, the research indicates a tendency for users to shift towards more complex, cognitively demanding tasks over time.

The team acknowledges that the study focuses on a single AI agent, Comet, and therefore may not fully represent the landscape of all available tools. Future work could explore the use of AI agents across different platforms and applications to broaden understanding of their potential.

This research provides a valuable baseline for understanding the evolving relationship between people and AI, and offers insights for developers, businesses, and educators as they navigate this rapidly developing technology. 👉 More information 🗞 The Adoption and Usage of AI Agents: Early Evidence from Perplexity 🧠 ArXiv: https://arxiv.org/abs/2512.07828 Tags: