AI economic impact studies for governments

What the major AI economic impact studies say

Published AI economic impact studies measure different quantities, so a single “AI will add X” headline hides which one is meant. The comparison below gives each study’s own measure, method and stated limit, using official and research sources. It does not rank them, and estimates published by consulting firms are outside its scope.

Four things AI impact studies measure

• Exposure. The share of jobs or tasks where AI could do part of the work. The IMF, the International Labour Organization (ILO) and the World Bank publish exposure measures. Exposure is not job loss.
• Productivity growth. How much faster output per worker or per unit of input could grow. Goldman Sachs, the OECD and Acemoglu estimate this, as annual rates or cumulative gains.
• GDP level. How much larger the economy is after a period. This follows from the productivity estimates and can look large or small depending on the horizon.
• Adoption, investment and sector size. Counts of users, spending and firms, from surveys and company data. The St. Louis Fed, the Stanford AI Index and the UK Department for Science, Innovation and Technology (DSIT) publish these.

These are different quantities, so a single “AI will add X” headline hides which one is meant. Estimates published by consulting firms are outside the scope of this comparison.

The major studies compared

The table gives each study’s own measure, method and stated limit. It does not rank them.

Study and date	What it measures	Headline figure	Method	Stated caveat
IMF, Gen-AI and the Future of Work, January 14, 2024	Share of employment exposed	Almost 40 percent globally; about 60 percent in advanced economies; 40 percent in emerging markets; 26 percent in low-income countries	Overlap between AI applications and the abilities each occupation requires, with a complementarity index	Results depend on calibration and the uncertain pace of adoption; exposure is not displacement
Goldman Sachs Research, March 26, 2023	GDP level and labor productivity growth	7 percent (almost $7 trillion) rise in annual global GDP over 10 years; about 1.5 points a year of US labor productivity growth	Task exposure from US and European occupational data, with past productivity booms as analogues	“Highly uncertain”; US scenarios from 0.3 to 3.0 points; not built into baseline forecasts
Acemoglu, NBER Working Paper 32487, May 2024	US TFP and GDP, 10 years	No more than 0.66 percent TFP gain; GDP 0.93 to 1.16 percent	Task-based model: 4.6 percent of tasks affected, with cost savings taken from experimental studies	Not peer reviewed; early evidence from easy-to-learn tasks
OECD, Miracle or Myth?, November 2024	Annual TFP and labor productivity growth	TFP 0.25 to 0.6 points a year (labor productivity 0.4 to 0.9) over 10 years	Micro-level gains, exposure and adoption, in a multi-sector general-equilibrium model	Price effects in general equilibrium could reduce the gain
ILO Working Paper 140, May 20, 2025	Share of workers exposed to generative AI	One in four workers in an occupation with some exposure; 3.3 percent in the highest category	Task scoring with a worker survey, expert rounds and an AI assistant	Perceived automation potential, not measured job loss
World Bank, World Development Report 2026, August 4, 2026	Jobs at risk and jobs with productivity gains	4.5 percent of jobs at risk in low- and middle-income countries against 14.2 percent in high-income; 16.2 percent of jobs in developing economies could gain productivity against 18.7 percent	Not stated in the press release	None stated in the press release
St. Louis Fed, November 13, 2025	Adoption and implied productivity	54.6 percent of US adults used generative AI in August 2025; productivity up to 1.3 percent since ChatGPT’s release	Population survey with self-reported time savings	Described as not causal; 2024 adoption figure was revised
Stanford AI Index 2026, economy chapter	Investment, adoption, early macro data	$581.69 billion corporate AI investment in 2025; 88 percent of surveyed organizations use AI; US productivity growth 2.7 percent in 2025	Compilation of company, survey and national data	Macro evidence is early and mixed; adoption survey data are self-reported and directional
UK DSIT, AI Sector Study 2024, September 3, 2025	Size of the UK AI sector	5,862 firms; £23.9 billion revenue; £11.8 billion GVA; 86,139 employees	Secondary data, 298 survey responses and 52 interviews	Revenue growth includes product and service relabeling
UK government claims, January 2025	GDP and productivity	Up to £47 billion a year over a decade; £400 billion more by 2030	£47 billion derived from IMF figures; £400 billion from a cited outside report	Government says the £47 billion is not an official IMF estimate

Productivity and GDP

Three reports bracket the range. Goldman Sachs’s March 2023 analysis says widespread adoption could eventually raise annual global GDP by 7 percent over a 10-year period, and that US labor productivity growth could rise by around 1.5 points a year, which it calls admittedly large and highly uncertain. Daron Acemoglu’s NBER paper applies Hulten’s theorem to a task-based view and finds a total factor productivity gain of no more than 0.66 percent over 10 years, with GDP up 0.93 to 1.16 percent once investment effects are included. The OECD’s Miracle or Myth? puts annual total factor productivity growth at 0.25 to 0.6 percentage points, or 0.4 to 0.9 points for labor productivity.

The three are not like for like. Goldman Sachs states a level effect on global GDP and an annual rate for US labor productivity. Acemoglu states a cumulative 10-year gain for the United States. The OECD states annual rates from a multi-sector model. Total factor productivity and labor productivity are different measures, and a cumulative gain over 10 years is a different number from an annual rate. A later OECD paper, summarized in an October 2025 brief by its authors, gives 0.4 to 1.3 points of annual labor productivity growth for high-exposure countries such as the United States and the United Kingdom and 0.2 to 0.8 points for countries such as Italy and Japan. Quoting only one end of any of these ranges changes the message.

How regional and sector impact studies work

Studies of a region or sector usually add direct, indirect and induced effects with an input-output model. The US Bureau of Economic Analysis’s RIMS II user guide says Type I multipliers account for the direct and indirect impacts of a final-demand change, and Type II multipliers add the induced impact from the spending of workers whose earnings are affected.

The clearest government example for AI infrastructure is the Virginia legislature’s review commission, which estimated in Data Centers in Virginia (December 2024) that the data center industry contributes about 74,000 jobs, $5.5 billion in labor income and $9.1 billion of Virginia GDP annually, including indirect and induced impacts. It used IMPLAN with 2022 data and 546 sectors, and customized the data center sector, which it notes lowers the operations employment multiplier by about 20 percent. That study covers data centers, which are AI infrastructure but not the impact of AI use. Measurement itself is still developing: a February 2026 BEA working paper notes that the US national accounts have no line item that identifies and measures the economic impact of AI.

Questions to ask of any AI impact study

1. Is the number exposure, productivity growth, GDP level, adoption or sector size?
2. Is it an annual rate or a cumulative gain, and over what horizon?
3. Which country or region, and which currency and price year?
4. Is it total factor productivity or labor productivity?
5. What does the author say about uncertainty, and does the headline keep that qualifier?
6. Who wrote it, who paid, and can the underlying model be read?