Microsoft study shows AI impacts 75% of major occupations in marketing communications

Microsoft researchers have documented widespread artificial intelligence capabilities across major occupation categories, with sales representatives, customer service workers, and writers showing the highest potential for AI collaboration. The comprehensive analysis, published July 22, 2025, examined 200,000 anonymized conversations between users and Microsoft Bing Copilot to understand which work activities people are using AI to assist with or perform.

According to the study titled Working with AI: Measuring the Occupational Implications of Generative AI, interpreters and translators face the highest AI applicability, with 98% of their work activities overlapping with frequent AI usage patterns. Customer service representatives, employing 2.86 million people nationwide, ranked among the top occupations for AI applicability alongside sales representatives, who represent over 1.14 million workers.

The research team from Microsoft Research, led by Kiran Tomlinson, Sonia Jaffe, Will Wang, Scott Counts, and Siddharth Suri, analyzed conversations from January 1 through September 30, 2024. Their methodology classified each conversation into work activities as defined by the O*NET database, which decomposes occupations into specific tasks and activities performed by workers.

“Given the rapid adoption of generative AI and its potential to impact a wide range of tasks, understanding the effects of AI on the economy is one of society’s most important questions,” the researchers stated in their findings.

The study reveals critical distinctions between AI assistance and AI performance. In 40% of conversations analyzed, users sought help with entirely different activities than what the AI actually performed. Users primarily sought AI assistance for information gathering, writing, and communication tasks, while AI most commonly acted in service roles as coaches, advisors, or teachers.

Information gathering emerged as the most successful AI application, receiving the highest positive feedback from users at 78% satisfaction rates. Writing activities followed closely at 76% satisfaction, while visual design and data analysis tasks showed the lowest success rates at approximately 55%.

Major occupation groups showed varying levels of AI applicability. Sales and related occupations scored highest at 0.32 on the researchers’ applicability scale, followed by computer and mathematical occupations at 0.30, and office and administrative support at 0.29. These three categories represent over 36 million workers combined.

The analysis found minimal correlation between AI applicability and wage levels, contrary to predictions that higher-wage occupations would face greater AI impact. The employment-weighted correlation between AI applicability and wages measured only 0.07, suggesting AI capabilities span across income levels rather than concentrating in high-wage positions.

Educational requirements showed stronger patterns. Occupations requiring bachelor’s degrees demonstrated higher AI applicability scores averaging 0.27, compared to 0.19 for positions with lower educational requirements. However, substantial overlap exists across all education levels.

Physical occupations showed the lowest AI applicability scores. Occupations involving manual labor, machinery operation, or direct physical interaction with people consistently ranked at the bottom of the applicability scale. Dredge operators, bridge tenders, and water treatment plant operators showed AI applicability scores near zero.

The research methodology incorporated three key measures: task completion rates, user satisfaction through thumbs-up feedback, and scope of impact. Activities receiving moderate or higher scope ratings indicated AI could assist with substantial portions of the work involved.

Writing and editing activities demonstrated the highest completion rates at over 85%, while visual design and scientific data analysis showed completion rates below 65%. The study found strong correlation between user satisfaction and task completion across different work activities.

When compared to predictions from a 2024 study by Eloundou et al., the Microsoft research showed remarkable alignment. The correlation between predicted AI impact and measured AI applicability reached 0.73 at the occupation level and 0.91 when aggregated to major occupation groups.

The researchers emphasized their findings represent only AI capabilities, not actual workplace implementation or economic outcomes. “Our data do not indicate that AI is performing all of the work activities of any one occupation,” they noted. Task completion rates rarely reached 100%, and scope of impact typically remained at moderate levels.

For the marketing community, these findings carry significant implications. Sales and advertising roles dominate the highest AI applicability categories, suggesting fundamental changes in how marketing professionals interact with prospects and customers. The study’s focus on communication and information provision aligns with emerging trends where AI agents may replace human attention as advertising targets.

Customer service representatives, a crucial component of marketing operations, ranked fourth among all occupations for AI applicability with scores of 0.44. This finding supports recent industry research showing marketers save an average of 114 minutes weekly through AI integration, translating to $3,520 in annual labor cost reductions per employee.

The timing coincides with broader industry recognition of AI’s transformative potential. Recent surveys show 68% of marketers plan to increase social media spending while incorporating AI tools, with automation emerging as the fastest-growing investment area.

Writers and authors, critical to content marketing strategies, showed AI applicability scores of 0.45, ranking fifth among all occupations studied. Technical writers, public relations specialists, and editors all appeared in the top 25 occupations for AI applicability, suggesting content creation workflows will experience significant transformation.

The research identified specific work activities most suitable for AI collaboration. “Provide information to customers” and “respond to customer problems or inquiries” ranked among the highest for both user satisfaction and task completion. These activities represent core functions in marketing customer engagement strategies.

Notably, the study found AI performs different activities than users request assistance with. While users seek help with research and analysis, AI typically provides explanation and communication services. This asymmetry suggests successful AI implementation requires understanding both user needs and AI capabilities.

The researchers acknowledged limitations in their analysis. The study examined only one AI platform and cannot determine what conversations occurred in work contexts versus personal use. Additionally, the decomposition of occupations into work activities, while standard practice, may not capture the full complexity of professional roles.

Future research directions include tracking how AI capabilities evolve over time and understanding how occupations restructure their responsibilities in response to AI advancement. The researchers noted that entirely new occupations may emerge, as has occurred with previous technological transformations.

The Microsoft study represents the first large-scale analysis of actual AI usage patterns across occupations, providing empirical evidence for understanding AI’s workplace impact. As artificial intelligence capabilities continue expanding, these findings offer crucial insights for workforce planning and professional development strategies.

Timeline

Summary

Who: Microsoft Research team led by Kiran Tomlinson, Sonia Jaffe, Will Wang, Scott Counts, and Siddharth Suri analyzed 200,000 conversations between users and Microsoft Bing Copilot, affecting millions of workers across sales, customer service, writing, and marketing occupations.

What: Comprehensive study measuring AI’s occupational implications across 844 work activities and 104 professions, revealing that 75% of major occupation groups show potential for AI collaboration, with sales representatives, customer service workers, and writers facing the highest automation potential.

When: Research analyzed conversations from January 1 through September 30, 2024, with the final study published July 22, 2025, coinciding with broader industry recognition of AI’s transformative impact on marketing and customer service operations.

Where: United States workforce analysis using O*NET occupational database and Bureau of Labor Statistics employment data, covering 164 million workers across computer-compatible occupations from administrative roles to creative professions.

Why: Understanding AI’s economic impact has become critical as 40% of Americans report using generative AI, with marketing professionals increasingly adopting AI tools that save an average of 114 minutes weekly and $3,520 annually per employee, while 68% plan to increase AI-enhanced digital advertising spending.

Key Terms Explained

AI Applicability Score: The comprehensive metric developed by Microsoft researchers to measure potential occupational impact by combining three factors: frequency of AI usage for specific work activities, task completion rates, and scope of impact. This score ranges from 0 to 1, with higher values indicating greater potential for AI collaboration. The methodology provides a standardized way to compare AI impact across different occupations, moving beyond subjective predictions to empirical measurement based on actual usage patterns.

Work Activities: The fundamental building blocks used to analyze occupational tasks, derived from the O*NET database’s hierarchical classification system. These activities represent specific functions performed across multiple occupations, such as “gather information from sources” or “provide customer assistance.” The study focused on 332 intermediate work activities (IWAs) rather than occupation-specific tasks, allowing researchers to identify how AI capabilities demonstrated in one context translate across all occupations performing similar activities.

Task Completion Rate: A critical success metric measuring how effectively AI systems finish user-requested work, determined through automated classification by GPT-4o-mini analyzing conversation outcomes. The researchers validated this metric against user thumbs-up feedback, finding strong correlation (r > 0.75) between completion rates and user satisfaction. Writing and information gathering tasks achieved the highest completion rates above 85%, while visual design and data analysis showed significantly lower success rates.

Customer Service Representatives: The occupation category employing 2.86 million workers that ranked among the highest for AI applicability with scores of 0.44. These roles involve core activities highly suitable for AI assistance, including responding to customer inquiries, providing information, and resolving problems. The finding suggests fundamental changes ahead for customer service operations, as AI demonstrates strong capabilities in communication and information provision tasks that constitute primary job functions.

Sales Representatives: A broad occupational category encompassing over 1.14 million workers across various industries, showing high AI applicability scores averaging 0.46 for service sales roles. The research identified sales activities like providing product information, explaining technical details, and responding to customer inquiries as particularly well-suited for AI collaboration. This finding aligns with industry trends toward AI-powered customer relationship management and automated lead qualification processes.

Information Gathering: The most commonly requested user activity in AI conversations, encompassing research, data collection, and knowledge acquisition tasks. This activity category received the highest user satisfaction ratings at 78% and demonstrated strong task completion rates above 85%. The prevalence of information gathering reflects AI’s core strength in processing and synthesizing large amounts of data quickly, making it particularly valuable for research-intensive occupations like journalism, analysis, and consulting.

O*NET Database: The comprehensive occupational information system developed under U.S. Department of Labor sponsorship that provided the foundational framework for this research. O*NET’s hierarchical structure breaks down 874 occupations into specific tasks, which map to detailed work activities, intermediate work activities, and generalized work activities. This standardized taxonomy enabled researchers to systematically analyze AI impact across the entire U.S. workforce structure.

Microsoft Bing Copilot: The publicly available generative AI system that served as the data source for this research, providing 200,000 anonymized conversations from January through September 2024. The choice of Copilot offered insights into mainstream AI usage patterns, as users access this system without specialized training or premium subscriptions. The researchers noted that different AI platforms may show varying usage patterns, with Copilot demonstrating less focus on programming tasks compared to other AI systems.

Occupational Impact: The broader concept of how artificial intelligence affects job roles, encompassing both direct automation of tasks and augmentation of human capabilities. The research distinguished between AI assistance (helping users complete tasks) and AI performance (directly executing work activities), finding these often involve different activities within the same conversation. Understanding occupational impact requires analyzing not just technological capabilities but also user adoption patterns and task success rates.

Scope of Impact: A qualitative measure rating how comprehensively AI capabilities address the full range of work involved in specific activities, classified on a six-point scale from “none” to “complete.” This metric helps distinguish between narrow AI applications that handle small portions of complex work activities and broad capabilities that address substantial portions of occupational tasks. The research found that AI typically demonstrates moderate scope impact, suggesting complementary rather than replacement relationships with human workers.