AI Usage Revealed: Real Conversations Show a More Complex Picture

In an era where artificial intelligence (AI) is becoming increasingly integrated into our daily lives, understanding how people actually use these tools is crucial. Major AI companies like Anthropic and OpenAI regularly publish reports on the usage of their products, such as Claude and ChatGPT. However, these reports are often selective, focusing primarily on work-related uses and filtering out personal interactions.

Anka Reuel, a computer science PhD candidate at Stanford’s Trustworthy AI Research (STAIR) Lab, highlights this issue: “There is no independent source to corroborate it.” To address this gap, Reuel co-leads the AI Observatory, a public platform that aggregates and analyzes real AI conversations from popular models like Claude and Gemini. These conversations are collected with users' consent through seven existing datasets, providing a more comprehensive and independent view of AI usage.

The AI Observatory’s research reveals significant differences in how people use various AI models and how these uses have evolved over time. According to Reuel, highly consequential decisions about AI’s benefits and risks are currently being made based on very limited data. The observatory aims to change this by offering a broader perspective that includes both work-related and personal interactions.

One of the most well-known sources of AI usage data is the Anthropic Economic Index, which focuses heavily on work- and productivity-related uses of Claude AI. However, when the AI Observatory researchers applied Anthropic’s methods to their dataset, they found that nearly half-48%-of the conversations would have been filtered out. These non-work-related conversations were more likely to involve health and relationships (44.2% versus 31.2% in Anthropic’s analysis), adult or illicit topics (7.9% versus 2.1%), harassment and hate (27.5% versus 5.66%), and sexual content (16.7% versus 2.4%).

David Widder, an assistant professor at the University of Texas at Austin who researches human-AI interactions, emphasizes the value of this independent analysis: “Having the AI Observatory’s bird’s-eye-view analysis helps researchers understand the different uses more consistently.” This comprehensive view is crucial for policymakers and researchers to make informed decisions about the ethical implications and regulatory needs of AI.

The Range of Real-World Interactions

The datasets analyzed by the AI Observatory include a wide range of conversations, from mundane tasks to deeply personal interactions. For instance, users often seek emotional support or companionship from AI models like Claude. Anthropic has published separate reports on these uses, but having them integrated into a broader analysis provides a more cohesive understanding.

Shayne Longpre’s observations from 2023 to 2026 show early indicators that social interactions with AI are on the rise, while sensitive use numbers are dropping. This trend suggests that as people become more comfortable with AI, they may be using it in more diverse and less controversial ways. However, the data from the AI Observatory still highlights significant instances of sensitive and potentially harmful uses.

Small businesses are also increasingly adopting AI for practical tasks. For example, many are using ChatGPT for drafting emails and Copilot for managing spreadsheets. Despite this growing reliance, few small businesses have formal policies in place to guide their use of AI. This lack of structured guidelines could lead to inconsistent or potentially risky practices.

What Comes Next

The insights provided by the AI Observatory underscore the need for a more nuanced understanding of how people interact with AI. As these tools become more ubiquitous, it is essential to consider both the positive and negative aspects of their usage. Policymakers and researchers must use this data to develop guidelines that ensure AI is used ethically and responsibly.

Reuel emphasizes that the AI Observatory’s goal is not to demonize AI but to provide a balanced and transparent view of its real-world applications. By doing so, they hope to foster a more informed and responsible approach to AI development and deployment. As we continue to integrate AI into our lives, understanding its true impact will be key to harnessing its potential while mitigating its risks.