A Stanford Digital Economy Lab working paper says interviews with workers found 79 tasks missing from O*NET across six occupations, showing how incomplete job lists can alter exposure estimates.
The paper, dated Oct. 26, describes TaskNet, a process that turns workers’ accounts of daily duties into occupation-level task lists. Researchers gathered 2,726 reports from 132 workers and distilled them into 202 statements accepted by O*NET Center analysts. A separate group of 318 workers submitted 7,923 ratings on the tasks’ relevance and importance.
For the six occupations, TaskNet matched 75% of the 118 task entries in ONET 31.0. It added 79 duties without a database equivalent and revised 58 descriptions. In a check using reports from workers outside the group that built TaskNet, the inventory covered 89% of reported tasks, against 79% for ONET. In employer job postings, TaskNet matched 72.5% of listed activities, compared with 70.2% for O*NET.
The revised inventories changed AI exposure estimates in both directions, the paper said. Mercor’s research post attributed the creation or reshaping of five tasks to AI adoption, citing prompt design and AI-system evaluation in software engineering and review of AI-generated legal material.
The Stanford paper says more complete task lists could help direct workforce policy and training toward workers most likely to be affected.
O*NET is the U.S. Department of Labor’s primary source of occupational information and is used for career exploration, workforce planning and research. The study reports changes in task coverage and AI exposure estimates, not employment outcomes.
