AI exposure Legal
Will AI replace paralegals?
Paralegals and Legal Assistants. US occupation 23-2011.
The reading
29.3%
of the time paralegals spend on their work goes on tasks where AI is already being used to do that work.of paralegals' working hours go on tasks AI has already started taking off their plate.
Ranked 70 of 756 US occupations. More exposed than 90% of them.Further along than 90% of them.
Per cent of working time Square-law scale
Average US job 7.7%
One of the most exposed jobs in the country.Front of the queue for an AI assistant.
What AI already does for paralegalsWhat paralegals no longer do alone
O*NET lists 7 tasks for paralegals. AI use at work shows up in 1 of them. Each score runs from 0.5, where every observed use helps a person with the task, to 1, where every use does the task outright.
- Gather and analyze research data, such as statutes, decisions, and legal articles, codes, and documents.Core task mostly automated 0.94
What it does not touch, yetWhat still needs a human
6 of the 7 tasks show no AI use at work in the data.
- Investigate facts and law of cases and search pertinent sources, such as public records and internet sources, to determine causes of action and to prepare cases.Core task
- Prepare affidavits or other documents, such as legal correspondence, and organize and maintain documents in paper or electronic filing system.Core task
- Prepare for trial by performing tasks such as organizing exhibits.Core task
- Appraise and inventory real and personal property for estate planning.
- Arbitrate disputes between parties and assist in the real estate closing process, such as by reviewing title searches.
- Direct and coordinate law office activity, including delivery of subpoenas.
How that compares
Among the 7 legal occupations, paralegals rank 3. The group averages 21.6%, and its most exposed job is Judges, Magistrate Judges, and Magistrates at 31.1%. The average US occupation sits at 7.7%.
Just above in the ranking
- 67Instructional Coordinators 30.4%
- 68Bill and Account Collectors 29.9%
- 69Middle School Teachers, Except Special and Career/Technical Education 29.7%
Just below
What this does and does not mean
Exposure is observed use, not a forecast. Anthropic maps work conversations with Claude onto the tasks O*NET lists for each job and asks, of the tasks AI could speed up, which are actually being used for at work. A use that does the task outright counts in full and a use that assists a person counts half. The task scores are averaged into the occupation figure by the share of time spent on each task.
A high figure means AI is already part of how this work gets done. It does not mean anyone has been replaced. Anthropic's own reading of the labour market finds no systematic rise in unemployment among exposed workers so far, and about 14% fewer 22 to 25 year olds starting exposed jobs.No measurable rise in unemployment among exposed workers so far, and about 14% fewer 22 to 25 year olds starting these jobs. Gap years for everyone. Labor market impacts, 5 March 2026.
Task list and titles from O*NET 31.0 by the US Department of Labor, Employment and Training Administration, used under the CC BY 4.0 licence. Scores from the Anthropic Economic Index (CC BY). Checked 1 November 2026. How the ranking works.