Robots and employment
9%, 23%, or 47% of jobs at risk? The government says nobody knows
9%, 23%, or 47% of US jobs are at risk from automation, depending on which study you open. The US Government Accountability Office looked at all of them and reached a blunter conclusion: no one has the data to know.
The three numbers are not rounding differences. They come from different research groups using different methods, and the spread from 9% to 47% is a signal that these are model outputs resting on assumptions, not measurements of a knowable fact.
This page traces the estimates and the GAO's verdict to the report itself. It is a page about the absence of a reliable number, which is the most honest thing that can be said about the question.
Data covers GAO review of automation-and-employment research (2019). Last reviewed by a human editor before publication.
The figures and where they come from
Each figure is rated for how safely you can cite it today. Ratings judge current usability, not whether a number was ever correct.
| Figure | What it is | Source | Citation Confidence | Notes |
|---|---|---|---|---|
| 9% of workers | Low estimate (Arntz et al.) | [A] | Medium | The lowest of the three estimates GAO assembled, from researchers who counted tasks within occupations rather than whole occupations. |
| 23% of work hours | Middle estimate (McKinsey) | [A] | Medium | The McKinsey Global Institute estimate of US work hours automatable by 2030. A different unit again: hours, not jobs. |
| 47% of employment | High estimate (Frey and Osborne) | [A] | Medium | The widely quoted high-end figure. GAO presents it as one estimate among several, not a settled number. |
| 'Better Data Needed' | GAO's assessment | [A] | High | The report's own title. GAO's conclusion is that the data available cannot support confident planning for automation's effects on jobs. |
| 7.3 million | Job openings, December 2018 | [A] | Medium | Context GAO cites: a strong labor market alongside the automation-risk debate, a reminder the projections are not describing the present. |
Why the numbers disagree
The estimates disagree because they count different things. One approach scores whole occupations for their susceptibility to automation and sums the employment, producing the 47% figure. Another looks inside occupations at individual tasks, and finds only 9% of workers at high risk. A third counts automatable work hours rather than jobs, and lands at 23%. Different units and assumptions, different answers.
The GAO's contribution is not a fourth number but a verdict on the first three. Its report is titled 'Better Data Needed,' and its finding is that the available data cannot support confident assessment or planning. When the government's own auditor says the inputs are inadequate, a precise output should be read with suspicion.
The spread also warns against treating any single figure as fact. A quantity that ranges from 9% to 47% across credible studies is not a measurement with error bars; it is a set of models disagreeing. The honest summary is a range and a caveat, not a point estimate.
How to cite these figures
Present the range, not a single number: credible estimates of jobs at risk from automation run from about 9% to 47%, because they count differently.
Cite the GAO's verdict directly: its 2019 report concluded that better data is needed to assess and plan for automation's effects, meaning no confident figure exists.
If you must give one number, attach its method and its source, and never imply it is the consensus. There is no consensus figure to give.
Where people go wrong
Quoting 47%, or any single figure, as the established share of jobs at risk. It is the high end of one method; others are far lower.
Treating the estimates as measurements. They are model outputs whose spread reflects differing assumptions, which is exactly why GAO called for better data.
Reading the projections as descriptions of now. GAO notes a strong labor market with 7.3 million openings alongside the debate; the figures are about a modeled future, not the present.
How we checked
This page is built on a single authoritative source: the GAO's 2019 report on workforce automation, which assembles the competing estimates and delivers its own assessment. We opened the report and confirmed the 9%, 23%, and 47% figures, the 7.3 million openings, and the 'Better Data Needed' finding in its text.
The value here is the compilation and the verdict, not a new number. Rather than pick a favorite estimate, we present the GAO's own conclusion that the data cannot support one, which is the defensible position.
We deliberately keep this separate from a page explaining what the 47% figure specifically means. This page is about the disagreement and the data gap; the point is that the government looked and found the numbers wanting.
Full source list
Primary sources, with live links. Every figure above traces to one of these.
- [A]U.S. Government Accountability OfficeMarch 2019
U.S. Government Accountability Office, GAO-19-257, "Workforce Automation: Better Data Needed to Assess and Plan for Effects of Advanced Technologies on Jobs"
https://www.gao.gov/assets/gao-19-257.pdf
Common questions
- How many jobs will automation eliminate?
- No one can say with confidence. Credible estimates range from about 9% to 47% of jobs or work hours, and the US Government Accountability Office concluded in 2019 that the available data is not good enough to know.
- Why do the estimates disagree so much?
- They count different things: whole occupations, tasks within occupations, or work hours, each with different assumptions. That is why the answers span from 9% to 47%.
- What did the GAO actually conclude?
- Its report is titled 'Better Data Needed.' It found that the data available cannot support confident assessment or planning for automation's effects on employment.
- Is there a reliable single number?
- No. A figure that ranges from 9% to 47% across credible studies is a set of models disagreeing, not a measurement. The honest answer is a range with a caveat.
More data, traced to source
- '47% of jobs at risk': what Frey and Osborne actually measured
The famous claim that nearly half of jobs will be automated traces to one 2013 study. It estimated the probability that occupations are susceptible to computerisation, not a prediction that 47% of jobs will vanish. Here is what it really said.
- 'One robot destroys 5.6 jobs': what the study really measured
The viral statistic that one robot destroys 5.6 jobs comes from a real, careful study. But the 5.6 figure is a local-labor-market estimate, not a claim that each robot nationwide eliminates 5.6 jobs. Here is what the researchers actually measured.
- How many US firms actually use robots? Far fewer than you would think
The story is that everyone is automating. The US Census Bureau's own firm survey found only about 2% of US firms use robotics. A newer Census release puts it far higher. The gap is the whole story.