Generative AI's Real-World Impact: Study Reveals Which Jobs Are Most Transformed by Tools Like Copilot

As generative AI tools flood workplaces, a critical question remains unanswered: How is this technology actually transforming work? A new study from Microsoft researchers provides unprecedented empirical evidence, analyzing 200,000 anonymized Bing Copilot conversations to map AI's real occupational footprint.

The AI Assistant in Action: What Are We Asking Copilot?

The research team, including Kiran Tomlinson and Siddharth Suri, classified user requests into work activities, discovering two dominant patterns:

• Top user requests: Gathering information (32%) and writing tasks (28%)
• Top AI outputs: Providing information (41%), writing assistance (22%), teaching (17%), and advising (12%)

"The most common work activities people seek AI assistance for involve gathering information and writing, while AI primarily provides information, writing support, teaching, and advising," the authors note, highlighting the assistant-like nature of current interactions.

Occupational Impact: The Jobs AI is Transforming First

By cross-referencing activities with occupational databases, the team developed an AI Applicability Score quantifying exposure:

Surprisingly, sales roles ranked highly due to their information-communication demands, while creative professions showed lower-than-expected scores. The findings challenge theoretical models predicting uniform disruption across knowledge work.

Beyond the Hype: Real-World Success and Limitations

The study measured task success rates through user engagement patterns, finding AI most effective for:

1. Factual information retrieval (78% success rate)
2. Technical documentation (72%)
3. Email drafting (68%)
4. Learning concepts (65%)

Complex problem-solving and nuanced creative tasks had significantly lower success rates. Notably, higher-wage occupations showed greater AI applicability, contradicting assumptions that AI would primarily impact lower-skilled roles first.

The Education Paradox

A counterintuitive finding emerged: Workers with graduate degrees used AI for 40% more work activities than those with only high school diplomas. This suggests AI is currently augmenting rather than replacing advanced expertise, creating a potential "augmentation divide" where highly educated professionals benefit disproportionately.

From Predictions to Reality

The team compared their empirical data with previous theoretical studies of AI exposure, finding:
- Overestimation of manufacturing/transportation AI impact
- Underestimation of administrative/sales role transformations
- Significant variation within occupational categories based on task composition

As generative AI evolves from novelty to infrastructure, this research provides the first large-scale evidence of how it's actually reshaping work. The full implications will unfold as these assistant-style interactions gradually restructure job designs and skill requirements across the knowledge economy.

Source: Working with AI: Measuring the Occupational Implications of Generative AI (Tomlinson et al., arXiv:2507.07935)