A recent study by researchers at Microsoft examined how artificial intelligence is being used in real work conversations. By analyzing 200,000 anonymized chats between people and Microsoft Copilot, the team created a detailed picture of where generative AI fits into modern work, and where it does not.
They tracked how often users asked AI for help with work tasks, how well AI completed those tasks, and how broadly those tasks applied to each job. Using that data, they created a numerical score for over 900 jobs. A high score meant AI was frequently used for important parts of the job and performed those tasks well. A low score meant little to no overlap.
The results showed a sharp divide. Some occupations matched closely with the way people already use AI. Others showed almost no connection. This report focuses on both sides by listing the 40 most and least AI-compatible jobs based on actual user behavior.
Where AI Works Well
The top scoring roles mostly involve language, information, or communication. These are jobs that depend on gathering details, answering questions, drafting content, or presenting knowledge to others. In many cases, the AI served as a digital assistant that helped users write, explain, translate, or summarize.
At the top of the list were interpreters and translators. Their work involves transforming written or spoken language across contexts, and AI has already shown strength in performing these tasks quickly and accurately. Writers, editors, and proofreaders also scored high, as many people are already using AI tools to generate, revise, or polish documents.
Other top-ranked roles include customer service agents, sales representatives, journalists, PR specialists, and educators. These jobs often require giving people information, preparing written materials, or guiding others through a process. These are areas where AI responses are more likely to be useful and well received.
The AI was not replacing these workers. Instead, the study showed that users were using AI to assist with parts of their tasks. This distinction was key to the way the research was designed. It separated what the user was trying to achieve from what the AI actually did during the conversation.
The 40 Most AI-Compatible Occupations
Each of these roles scored high in three areas: the share of job tasks that overlapped with AI usage, how well AI completed those tasks, and how much of the occupation those tasks covered.
- Interpreters and Translators
- Historians
- Passenger Attendants
- Sales Representatives (Services)
- Writers and Authors
- Customer Service Representatives
- CNC Tool Programmers
- Telephone Operators
- Ticket Agents and Travel Clerks
- Broadcast Announcers and Radio DJs
- Brokerage Clerks
- Farm and Home Management Educators
- Telemarketers
- Concierges
- Political Scientists
- News Analysts, Reporters, and Journalists
- Mathematicians
- Technical Writers
- Proofreaders and Copy Markers
- Hosts and Hostesses
- Editors
- Business Teachers (Postsecondary)
- Public Relations Specialists
- Demonstrators and Product Promoters
- Advertising Sales Agents
- New Accounts Clerks
- Statistical Assistants
- Counter and Rental Clerks
- Data Scientists
- Personal Financial Advisors
- Archivists
- Economics Teachers (Postsecondary)
- Web Developers
- Management Analysts
- Geographers
- Models
- Market Research Analysts
- Public Safety Telecommunicators
- Switchboard Operators
- Library Science Teachers (Postsecondary)
Most of these jobs involve structured knowledge work. Some include writing technical guides, while others involve answering questions or responding to common customer issues. The overlap with AI in these jobs was not just frequent, but successful. Conversations where AI helped with these tasks often ended with a completed goal or positive user feedback.
Where AI Has No Role So Far
On the other end, the researchers found dozens of jobs where AI showed no real connection to the work being done. These occupations had an AI applicability score of zero. That meant no significant overlap between their daily tasks and what AI was used for in the dataset.
In nearly every case, these jobs required physical skills, specialized equipment, or real-world handling. Many involved cleaning, operating machinery, preparing food, or providing in-person care. Even if AI could offer instructions, the actual task still had to be done by a person, on site, using physical tools or touch.
These occupations also tended to be hands-on in a way that language models are not designed for. They required moving, lifting, installing, or interacting with the environment in ways that AI cannot simulate. Some jobs required high precision, others involved safety risks or regulatory requirements. In all cases, the study found no practical use of AI for their work.
The 40 Least AI-Compatible Occupations
These jobs showed no measurable overlap with AI use in the study. They had zero coverage, meaning none of their key work activities appeared in AI-assisted conversations with users.
- Water Treatment Plant and System Operators
- Pile Driver Operators
- Dredge Operators
- Bridge and Lock Tenders
- Foundry Mold and Coremakers
- Rail-Track Laying and Maintenance Equipment Operators
- Floor Sanders and Finishers
- Orderlies
- Motorboat Operators
- Logging Equipment Operators
- Paving, Surfacing, and Tamping Equipment Operators
- Maids and Housekeeping Cleaners
- Roustabouts, Oil and Gas
- Roofers
- Helpers, Roofers
- Tire Builders
- Surgical Assistants
- Massage Therapists
- Gas Compressor and Pumping Station Operators
- Cement Masons and Concrete Finishers
- Dishwashers
- Machine Feeders and Offbearers
- Packaging and Filling Machine Operators
- Medical Equipment Preparers
- Highway Maintenance Workers
- Helpers, Production Workers
- Prosthodontists
- Tire Repairers and Changers
- Ship Engineers
- Automotive Glass Installers and Repairers
- Oral and Maxillofacial Surgeons
- Plant and System Operators (All Other)
- Embalmers
- Helpers, Painters and Plasterers
- Hazardous Materials Removal Workers
- Nursing Assistants
- Phlebotomists
- Ophthalmic Medical Technicians
- Floor Sanders
- Bridge and Lock Tenders
These occupations span fields like healthcare, heavy industry, transportation, construction, and cleaning. Many involve specialized tools, patient care, or site-specific duties. For these jobs, AI was neither asked to help nor observed completing any relevant work activity.
A Divide Shaped by Task Type, Not Income or Industry
The researchers also examined whether salary or education level influenced AI applicability. They found only weak patterns. Some lower-wage jobs scored high, while some high-wage roles showed little AI overlap. There was a slight trend where jobs requiring a bachelor’s degree showed more applicability, but even that effect was modest.
The key factor was the type of task. If the job involved writing, explaining, organizing knowledge, or communicating, it was more likely to match how AI is currently being used. If the job involved physical motion, hands-on problem-solving, or direct care, it was unlikely to match.
Study Focused on Measured Use, Not Predictions
This study looked only at actual use. It did not attempt to forecast future changes to job markets or make claims about automation risk. It did not track how employers use AI internally, nor did it consider how jobs might evolve over time. The scores only reflect current patterns in how people used Copilot to help with tasks that align to occupations listed in federal labor data.
Still, the data offers a real-world snapshot of how AI is beginning to fit into everyday work. Some jobs already show clear patterns of use, while others remain disconnected. As AI tools grow and change, those patterns may shift. For now, the gap between roles where AI helps and those it doesn’t remains wide.
Read next: Facebook Most Cited in Online Abuse Reports from Environmental Activists