Automation probability figures for specific occupations range from under 5% to over 90%, and where any given job lands on that scale depends heavily on how the research was conducted. The original 47% figure from the Oxford Martin School's 2013 Frey and Osborne paper is still widely cited but is now considered a significant overestimate by most researchers in the field. More recent work using task-level data from the OECD and O*NET produces lower but more precise figures. The shift from occupation-level to task-level analysis is the single most important methodological development in this field, because it reveals that within almost every profession, some task configurations are highly exposed while others are well-protected.
What the research says
92M
jobs displaced by 2030
WEF Future of Jobs 2025
170M
new roles created by 2030
WEF Future of Jobs 2025
41%
of employers plan AI-driven headcount reductions
WEF 2025
55K
job cuts explicitly attributed to AI in 2024
Challenger, Gray and Christmas
The standard methodology involves two steps. First, each task within an occupation is rated on how automatable it is given current AI capabilities. Second, those task-level scores are weighted by how much time workers actually spend on each task and aggregated into an occupation-level figure. The weighting step is critical: an occupation where 80% of worker time is spent on highly automatable tasks gets a very different score from one where automatable tasks consume only 20% of time, even if both occupations contain some automatable work.
The 2013 Frey and Osborne paper from Oxford was groundbreaking but pre-dated the transformer and large language model era. It treated social intelligence, creativity, and perception as hard limits on automation that have since been substantially eroded by AI advances. The paper also rated entire occupations rather than tasks within occupations, which inflated risk estimates for jobs that contain a mix of automatable and non-automatable work. Later OECD analysis using task-level data arrived at a much lower 14% figure for highly automatable occupations.
The OECD's task-based methodology estimates 14% of jobs are highly automatable (greater than 70% of tasks). Another 32% face significant change as AI takes over some but not all of their tasks. Goldman Sachs research puts 300 million full-time jobs at risk of disruption globally. These figures represent different things: high automatability versus disruption versus full displacement, and it is important not to conflate them.
Standard occupation-level scores do not, which is their main limitation. Two software engineers can have vastly different automation exposures depending on whether their work is mostly writing boilerplate CRUD applications or designing novel distributed systems. O*NET task-level data provides more granularity, but even that does not capture how an individual's specific daily work is structured. That gap is exactly what task-level personal assessments are designed to address.
Across multiple research sources, the consistently highest-probability occupations are: telemarketers (greater than 95% task automatability by most estimates), data entry and transcription workers, loan underwriting clerks, insurance adjusters, standard bookkeepers, and basic legal researchers. The lowest-probability occupations include surgeons, mental health counsellors, elementary school teachers, plumbers and electricians, and roles requiring physical presence in unpredictable environments.
Free Assessment
Free Fossil Score assessment. No account required.
Get My Fossil Score