Dallas Fed: AI Lifts Productivity Only Where People Actually Use It

Your industry's AI "exposure score" predicts almost nothing about what happens next. What predicts it — according to new Federal Reserve Bank of Dallas research spanning the U.S. and 16 European economies — is whether the people around you actually _use_ the tools. Same occupations, same technology, wildly different results.

That is the quietly hopeful finding inside a July 7, 2026 analysis by Dallas Fed economist Scott Davis. And it flips the most common fear about AI on its head: your job's automation potential is not your destiny. Adoption is.

The 16% of workers producing 40% of the gains

Start with the headline number. U.S. labor productivity has grown at an annualized 2.4% since the beginning of 2024, compared with 1.6% average growth in the five years before the pandemic [Fact]. That is a real acceleration, and it is not spread evenly.

Three sectors are doing the heavy lifting: information, finance and insurance, and professional and technical services. Together they have posted average annualized productivity growth of 3.7% since the first quarter of 2024, versus 1.7% for the rest of the economy [Fact]. Those same three sectors happen to rank highest on the Felten–Raj–Seamans AI industry exposure index, which scores how well AI can replicate the tasks that make up a given job.

Here is the part that should reframe how you read every scary automation headline. Those three high-exposure sectors account for only 16% of total hours worked in the United States — yet they are responsible for 40% of all U.S. productivity gains since the start of 2024 [Fact]. A small slice of the workforce is generating a disproportionate share of the growth, and they are doing it in exactly the jobs that "AI exposure" rankings flag as most at risk.

Look at the industry-level numbers behind the Dallas Fed's chart data. Information sits at an AI exposure score of 1.15 and posted 7.8% productivity growth. Professional and technical services: exposure 1.71, growth 4.1%. Finance and insurance: exposure 2.04 — the highest of any sector — growth 2.7% [Fact]. Meanwhile construction, with an exposure score of -1.10, managed 1.2%, and leisure and hospitality, at -1.05, was essentially flat at -0.03% [Fact]. The industries AI can touch are the ones getting more productive. The ones it cannot touch are stuck.

Europe ran the same experiment and got nothing

If exposure alone drove productivity, the pattern would repeat everywhere. AI does not care about borders — an accountant's task list in Munich looks like an accountant's task list in Manhattan. By construction, the exposure index for a given sector is identical in the U.S. and the EU [Fact].

So Davis ran the same regression on European national accounts data. The result: in the EU, there is no correlation between a sector's AI exposure and its recent productivity gains [Fact]. Zero. The high-exposure European sectors are not pulling ahead the way their American counterparts are. European information services grew productivity 3.2%, but European finance and insurance actually _shrank_ -1.0%, and real estate fell -2.8% [Fact].

Two continents, same tools, same task lists, opposite outcomes. Something other than exposure is doing the work.

The missing variable is usage — and it is measurable

The Dallas Fed's answer comes from the Anthropic AI Usage Index, which normalizes Claude usage in each country by working-age population. A country using AI at exactly the global average scores 1.0.

The United States scores 3.69 — 3.69 times the global per-capita average [Fact]. The EU-27 average is 1.85, roughly half [Fact]. And the spread inside Europe is enormous: Estonia 3.05, France 2.66, the Netherlands 2.61, Ireland 2.39, Sweden 2.29, Portugal 2.23, Denmark 2.10, Germany 1.79, Italy and Spain both 1.62, Poland 1.41 [Fact].

Now line those usage numbers up against how strongly each country's productivity growth tracks AI exposure. Estonia — Europe's heaviest AI user — has the strongest relationship of any country in the sample, with a slope of 2.20. The U.S. sits at 1.18, France at 0.92, Portugal at 0.75 [Fact]. Move down the usage ladder and the relationship collapses or reverses: Germany -0.86, Spain -0.79, Poland -1.91 [Fact].

Davis's conclusion is that the U.S. is not an outlier. It sits right on the trendline. High productivity growth in AI-exposed American sectors is not American exceptionalism — it is what happens anywhere people actually adopt the technology [Claim].

Why this is better news than it sounds

The doom framing of AI exposure goes like this: high exposure means your tasks can be automated, which means you get replaced. The Dallas Fed data suggests a different chain of events. High exposure plus high usage produces higher output per hour — and in these sectors, that has so far shown up as _more valuable work_, not vanishing work. Information and professional services remain among the fastest-growing, best-paid parts of the U.S. economy even as their productivity surges.

And there is a second, more empowering implication. Exposure is assigned to you by the nature of your job. Usage is chosen — by you, by your employer, by your country's policy environment. Estonia does not have a structural advantage over Germany in the tasks its accountants perform. It has an adoption advantage. That gap is closable, and closing it is a decision, not a fate.

If you are a financial analyst, a management analyst, a software developer, a market research analyst, or an insurance underwriter, you are sitting in exactly the sectors where this productivity dividend is landing. The question the data asks you is not "will AI replace me?" It is "am I in the 3.69 group or the 1.41 group?"

The honest limits

Davis is careful, and so should we be. He states explicitly that this is correlation, not causation [Fact]. Productivity is shaped by dozens of factors — capital investment, labor market flexibility, energy costs, sector composition — and some of them correlate with AI use without being caused by it.

The Ireland data point makes this vivid. Ireland has a relatively high usage index of 2.39 but the most _negative_ slope in the sample at -3.39 [Fact]. Ireland's national accounts are famously distorted by multinational profit-shifting, which is a reminder that these are noisy macro series, not clean experiments.

Two more caveats worth naming. The usage index measures one AI product, Claude, as a proxy for national AI adoption [Estimate] — a reasonable proxy, but a proxy. And productivity growth is not the same thing as employment growth: an industry can produce more per hour while employing fewer people. Our reading of the ECB's 2026 labour market data and June's Challenger layoff report shows both effects running at once in different corners of the economy.

What to do with this on Monday

Stop treating your exposure score as a verdict. It describes what AI _could_ do with your task list, not what will happen to your paycheck. The European sectors with high exposure and low usage got neither the productivity gains nor a reprieve — they just got left behind.

Measure your own usage index. Roughly how many of your recurring weekly tasks currently touch an AI tool? If the honest answer is "almost none," you are living in a 1.41 country regardless of what passport you hold.

Push adoption where you have leverage. The Dallas Fed's cross-country evidence says the gains flow to places that use the tools intensively, not to places that merely could. Inside a company, that lever is often a single team's willingness to rebuild one workflow.

Watch the other 84%. The unresolved question Davis leaves open is whether AI becomes a genuine general-purpose technology and lifts the sectors that are currently low-exposure — construction, transportation, hospitality — where productivity growth has been sliding since 2023 [Fact]. That, not white-collar displacement, may be the bigger story of the next five years.

You can check where your own occupation sits on exposure and automation risk on our occupation pages — including financial analysts, software developers, management analysts, and insurance underwriters.

Sources

• Scott Davis, "International comparisons show AI effect on productivity," Federal Reserve Bank of Dallas, July 7, 2026 — https://www.dallasfed.org/research/economics/2026/0707
• Dallas Fed chart data (Charts 1–5, sector exposure and country usage series) — https://www.dallasfed.org/-/media/documents/research/economics/2026/0707data.xlsx
• Anthropic Economic Index (AI Usage Index by country) — https://www.anthropic.com/economic-index
• U.S. Bureau of Labor Statistics, Labor Productivity and Costs — https://www.bls.gov/productivity/

_AI-assisted analysis. Figures are drawn directly from the Dallas Fed article and its published chart data; O\*NET-based exposure and automation-risk scores on our occupation pages use our own methodology and may differ from the Felten–Raj–Seamans index cited here._