How much will AI boost productivity across G7 nations?

An OECD policy brief by Francesco Filippucci, Peter Gal, Katharina Laengle and Matthias Schief, summarizing their working paper Filippucci et al. (2025), projects that AI will raise annual aggregate labor productivity growth across G7 economies by a range that depends on both sectoral composition and adoption speed. Across three adoption scenarios — slow, medium and fast — "the projected annual aggregate labour productivity growth ranges between 0.4-1.3 percentage points in countries with high AI exposure," such as the United States and United Kingdom, "due to stronger specialisation in highly AI-exposed knowledge intensive services such as finance and ICT and more widespread adoption." In countries "where these determinants of AI gains are less favourable," such as Italy and Japan, "the estimated range is 0.2 to 0.8 percentage points." Broken out by scenario alone, the brief reports gains of "0.2 to 0.4 percentage points" under slow adoption, "0.5 and 1 percentage point" under medium adoption with expanded capabilities, and "0.8 to 1.3 percentage points" under rapid adoption with expanded capabilities.

The authors build the estimate by "adapting the task-based framework of Acemoglu (2024) to a sectoral context," combining three elements: micro-level productivity gains from AI, the AI exposure of tasks within each sector (drawn from Eloundou et al.'s 2024 task-level estimates for the O*NET database), and forecasts of firms' future AI adoption. For the micro-level input, rather than taking the most favorable experimental results at face value, the brief says it "relies on the three most precisely estimated effects, yielding an average baseline impact of 30%," folding in figures such as "around 14% in customer service, up to 56% in coding." Because adoption data are not harmonized across G7 statistical offices, the authors impute adoption for countries like the United Kingdom and Japan from "regressions that link adoption to digital infrastructure, skills and the sectoral composition of the economy," and project adoption to 2034 using the S-shaped diffusion curves of past general-purpose technologies — electricity, computers and the internet, and mobile phones, for the slow, medium and fast scenarios respectively. Knowledge-intensive sectors such as finance, ICT, publishing/media and professional services have 50-80% of tasks exposed to AI, versus 10-30% for manual-task sectors like agriculture, mining and construction, and the value-added share of the five most AI-exposed sectors ranges from about 15% in Japan, Italy and Germany to roughly 25% in the United States and 23% in the United Kingdom.

The brief explicitly builds on How much will AI actually boost productivity over ten years?, extending Acemoglu's single aggregate task-based calculation into a sector-by-sector, country-by-country projection; where this brief cites Acemoglu (2024) as implying "0.1 percentage point annual growth over the next decade," the OECD's range sits above that figure even at its low end, because it layers in sectoral specialization and an explicit adoption-speed dimension that Acemoglu's framework holds fixed. The brief's own positioning of its range against "about 1 percentage point (Aghion and Bunel, 2024; Filippucci et al., 2024)" also places it near What is generative AI's actual aggregate productivity gain today?, a separate estimate reached through self-reported survey time savings rather than task-exposure modeling — different data and methods converging on a broadly similar order of magnitude. Because the brief's figures are decade-ahead projections rather than observed outcomes, they neither confirm nor dispute Is AI productivity finally showing up in economic data?, which reads current labor-market data as the productivity signal already appearing; the OECD framework would locate any such signal as still early relative to its adoption curves running out to 2034.

The 0.2-1.3 percentage point range is a projection built on 2023-24-vintage task-exposure estimates, three selected experimental productivity studies, and modeled or imputed adoption trajectories through 2034 — not a measurement of realized productivity growth in any G7 country — and the brief is a policy-brief summary of a working paper, Filippucci et al. (2025), not described here as peer-reviewed. The comparison the authors draw to the US ICT boom of the mid-90s, when ICT's contribution to annual labor productivity growth was "about 1-1.5 pp per year," is offered only as a historical scale benchmark, not as evidence that AI's effect will follow the same trajectory.