🔍 In one sentence

An analysis of real-world Bing Copilot conversations shows that generative AI most often supports tasks in information and communication roles.

📌 Why This Matters

As tools like Bing Copilot become more widely used, there is growing discussion around which jobs might be automated or supported by AI. However, most of this debate is based on expert opinion rather than actual usage data.

Concerns remain that AI could broadly impact the workforce. But its influence varies depending on how people use it and what tasks it performs well—similar to how a Swiss Army knife is useful for some tasks but not others.

🧠 The Core Idea

Researchers examined 200,000 anonymized Bing Copilot conversations from US users, identifying both the users’ goals and the AI’s responses. These were then matched to a US government database of work activities to calculate an “AI applicability score” for different occupations.

This method provides insight into how AI is used in practice, whether as a tool for gathering and organizing information, or as a more active assistant that explains, advises, or teaches.

📊 Noteworthy Results

• Information and writing tasks dominate: The most frequent AI-supported tasks involved information retrieval, writing, and communication—key elements of knowledge-based work.
• Service and explanation roles for AI: AI mainly acted as an explainer, advisor, or coach, rather than directly completing users’ intended goals. In 40% of cases, the user’s goal and the AI’s action involved different types of work.
• Top-affected jobs: Roles such as interpreters, writers, sales reps, customer service agents, and tech-related occupations showed the highest AI applicability. Jobs involving physical tasks—like nursing assistants and machine operators—had the lowest.
• Modest link to education and wage: Jobs requiring a bachelor’s degree had slightly higher AI applicability scores. The link to wage levels was weak.
• Real-world use matches predictions: AI applicability scores had a strong correlation (r = 0.73) with previous expert predictions about AI’s labor market impact.

💡 What are the potential applications?

• Identifying roles that can currently be augmented with AI tools.
• Helping workers and policymakers understand which skills are more or less likely to be influenced by AI.
• Establishing a baseline for tracking AI’s evolving impact on different occupations.

⚠️ Limitations & Considerations

This study only includes data from Bing Copilot and reflects how users chose to interact with it, which may not represent all sectors or job types. It measures task-level alignment with AI capabilities, not actual employment changes or wage outcomes.

Source: Microsoft Research, Microsoft | Full Paper: http://arxiv.org/abs/2507.07935v1 | Authors: Kiran Tomlinson, Sonia Jaffe, Will Wang, Scott Counts, Siddharth Suri