Microsoft study names 20 “AI‑proof” jobs — health care and blue‑collar roles among the safest as communication work faces highest risk
A Microsoft Research analysis of ~200,000 Bing Copilot queries identifies occupations with the lowest overlap with generative AI, finding manual, in‑person, healthcare and many blue‑collar roles least exposed while communication work faces the highest risk.
Key takeaways
- Microsoft created an “AI applicability score” by mapping ~200,000 anonymized Bing Copilot interactions to U.S. occupational tasks — higher = more overlap with generative AI. See the Microsoft research paper and methodology and the ArXiv preprint.
- Jobs requiring physical presence or hands‑on care score lowest — roofers, construction laborers, home health aides and many healthcare and manufacturing roles show the least overlap with current LLM capabilities.
- Communication and knowledge work show highest exposure — interpreters, writers, PR and customer service roles map strongly to Copilot tasks like drafting and summarizing.
- Applicability ≠ displacement — Microsoft notes the score measures task overlap, not inevitable job loss; many roles may be augmented or redesigned. Read Microsoft’s note on applicability vs. displacement.
Main story
How Microsoft measured risk
Method: Microsoft linked real user questions and tasks sent to Bing Copilot in 2024 to O*NET job task categories to produce an AI applicability score. A higher score indicates alignment with tasks LLMs handle well — writing, summarizing, coding, planning and other information work; a lower score indicates physical, on‑site, or relational tasks outside current LLM capabilities. See the Microsoft methodology and the ArXiv preprint for full details.
Twenty jobs least likely to be impacted by AI
The following occupations appear in Microsoft’s lowest‑applicability group or align with the paper’s bottom decile of scores. Many of these roles were highlighted in press summaries; where indicated, items are directly named by coverage or inferred from low scores and O*NET mapping. Sources include a GeekWire summary, Microsoft’s research page, and the ArXiv preprint.
- Roofers — work at height, use tools, solve on‑site problems.
- Helpers — Roofers — assist with setup, materials, cleanup.
- Tire Builders — operate tire manufacturing equipment; machinery‑based tasks.
- Gas Compressor and Pumping Station Operators — manage complex equipment on site.
- Industrial Truck and Tractor Operators — forklifts and material handling require situational awareness.
- Cement Masons and Concrete Finishers — hands‑on craft work at construction sites.
- Dishwashers — manual cleaning tasks in food service.
- Machine Feeders and Offbearers — load and remove parts from machines.
- Packaging and Filling Machine Operators — run and adjust production lines.
- Medical Equipment Preparers — sterilize and ready instruments under strict rules.
- Surgical Assistants — hands‑on intraoperative support and real‑time judgment.
- Massage Therapists — physical, personal therapy and patient trust.
- Ophthalmic Medical Technicians — diagnostic tests and patient handling with specialized tools.
- Supervisors of Firefighters — field leadership and rapid, life‑saving decisions.
- Tire Repairers and Changers (inferred) — roadside and shop work requiring hands‑on skill.
- Construction Laborers — varied on‑site physical tasks with low AI overlap.
- Home Health and Personal Care Aides — in‑home care, mobility help and personal support.
- Nursing Assistants — bedside care tasks like bathing, feeding and moving patients.
- Dental Assistants — chairside procedures, instrument handling and x‑ray positioning.
- Phlebotomists — draw blood and handle specimens; requires fine motor skill and trust.
Why health care and blue‑collar work score low
Microsoft’s analysis shows occupations without heavy text, data or research tasks have less overlap with current generative AI. Physical roles, face‑to‑face care and specialized tool use do not match the kinds of tasks people sent to Copilot in 2024. The study also found that occupations requiring a bachelor’s degree tend to show higher applicability. See coverage for context in Fortune and the Microsoft research.
Which jobs are most at risk — a quick contrast
Contrast: Microsoft’s top‑applicability occupations include interpreters, translators, writers, reporters, PR professionals and customer service representatives — roles that align closely with Copilot tasks like drafting, summarizing and editing. For reporting and analysis, see GeekWire and Fortune.
Important caveats from Microsoft
- Applicability is not displacement: the score shows task overlap, not inevitable job loss — many roles may be augmented or redesigned. Read Microsoft’s commentary on applicability vs. displacement.
- Scope limits: the study focuses on LLMs and phase‑one generative tools and does not fully model robotics, autonomous vehicles or specialized industrial AI that could alter risk for equipment‑heavy roles. See contextual coverage at Fortune.
- Usage dependency: the analysis depends on how people used Bing Copilot in 2024; results could shift as tools and adoption evolve. Full methodology: Microsoft research and the ArXiv preprint.
Implications for Utah
Economic impact: Utah’s economy blends construction, manufacturing, health care and a growing tech sector. The Microsoft findings suggest Utah workers in trades, construction and hands‑on health care will face lower direct exposure to current generative AI, which matters for local housing, energy and mining projects. See regional framing in GeekWire and Microsoft’s study.
“For conservative voters and state leaders who emphasize local jobs, family wages, and self‑reliance, the study offers data to support investment in trade schools and apprenticeships.”
Social, cultural and practical effects
Social effects: Many Utah families depend on health aides, nursing assistants and skilled trades for steady work. Those occupations offer entry points without four‑year degrees and remain rooted in human interaction and manual skill — a key local resilience factor. (See Fortune.)
Cultural relevance: Utah’s culture values practical skills, service and community care. The low‑applicability list — roofers, construction laborers, home health aides, surgical assistants — reflects work that fits local values and supports arguments for emphasizing trades and vocational training.
Practical applications: Employers and workers in Utah should use this research to plan: prioritize on‑the‑job training and safety programs, expand apprenticeships, boost community college capacity, and fund certification programs in nursing assistance, phlebotomy and equipment operation. Workers in communication and office roles should expect tools like Copilot to change daily tasks and consider upskilling in areas where human judgment and on‑site skills remain essential. See Microsoft’s workplace discussion at Microsoft WorkLab and the applicability blog.
Sources and further reading
- Microsoft research paper and methodology
- ArXiv preprint of the study
- GeekWire summary and occupation lists
- Fortune coverage and context
- Microsoft blog on applicability vs. displacement
- Microsoft WorkLab discussion of AI at work
Worldwide (Times Media Service) — Sources: Microsoft Research, GeekWire, Fortune, ArXiv.
