“Is AI productivity real?” got a widely shared answer in early 2026: a National Bureau of Economic Research survey of nearly 6,000 executives found that 89% saw no change in productivity from AI over the past three years. That statistic circulated everywhere. What circulated far less is that a second NBER paper, published weeks later using different survey data, found AI was already delivering measurable, positive productivity gains at many firms — gains the researchers say are strengthening heading into 2026. Both are real, peer-adjacent economic research, both are current, and they’re measuring different things. Understanding what each one actually asked is more useful than picking whichever headline confirms what you already believed about AI at work.
The Headline That Went Viral: “89% Saw No Impact”
The paper behind that number, “Firm Data on AI,” surveyed nearly 6,000 senior executives across the US, UK, Germany and Australia. It found 69% of firms actively use AI, and more than two-thirds of executives use it themselves — but only around 1.5 hours a week on average, a usage intensity worth noting on its own. Looking backward, more than 90% of managers reported no impact on employment at their firm over the past three years, and 89% reported no change in productivity, measured as sales volume per employee. That’s the number that made headlines, and it’s accurate to what the survey asked: a backward-looking, self-reported assessment of the last three years, not a forecast and not an independent measurement of actual output.
The Same Survey Also Found Positive Expectations
Here’s the detail that got dropped from most of the coverage: those same executives, looking forward, predicted AI would boost their firm’s productivity by an average of 1.4% and output by 0.8% over the next three years, alongside a modest expected reduction in employment. In other words, the “no impact yet” finding and “meaningful impact coming” finding sit inside the same paper, from the same respondents. That combination — limited impact reported so far, larger impact expected ahead — is a recognizable pattern economists call the productivity paradox: transformative technologies are widely believed to matter well before their effects show up cleanly in the numbers.
A Different NBER Paper Found Current, Measurable Gains
A separate NBER working paper, “Artificial Intelligence, Productivity, and the Workforce,” surveyed a different sample — nearly 750 corporate executives, weighted toward CFOs — and reached a notably different headline: labor productivity gains from AI are already positive, vary by sector, and are expected to strengthen further in 2026. The largest effects were concentrated in high-skill services and finance, and the researchers attribute the gains primarily to revenue-based productivity improvements — the kind associated with innovation and demand, not simply firms spending more on AI infrastructure.
| “Firm Data on AI” | “AI, Productivity, and the Workforce” | |
|---|---|---|
| Sample size | ~6,000 executives | ~750 executives, weighted toward CFOs |
| Geography | US, UK, Germany, Australia | Primarily US firms |
| What it measured, looking back | Self-reported change in sales-per-employee over 3 years | Revenue-based total factor productivity |
| Headline finding | 89% reported no productivity change | Positive gains already, concentrated in high-skill services and finance |
| Forward-looking view | +1.4% productivity, +0.8% output predicted over 3 years | Gains expected to strengthen further in 2026 |
These two findings aren’t necessarily contradictory once you separate what each one actually measured — one is a blunt, self-reported “did productivity change” question across a huge, broad sample; the other is a more granular, revenue-based measurement concentrated in a smaller set of firms weighted toward finance roles most likely to have the data to answer precisely. Both can be true: most firms, broadly surveyed, haven’t yet seen or reported a change, while a meaningful subset — particularly in high-skill services — already has.
The Perception Gap Your Business Case Shouldn’t Ignore
Across this research, one pattern shows up consistently and matters more than either headline number: executives report feeling bigger productivity gains from AI than what shows up in their own firm’s revenue data. Researchers attribute that gap to delayed output realization and quality improvements that haven’t yet made it into measured revenue — which is a real phenomenon, but it’s also exactly the gap that lets a business overstate AI’s ROI internally before the numbers actually back it up. If you’re building an investment case for AI tools around how much faster or easier work feels, treat that as a leading indicator worth testing, not as proof of a return you can bank on yet — the same caution this site applied when weighing whether a paid AI tool is actually worth the upgrade over a free one.
Where the Real Gains Actually Concentrate
Gains aren’t evenly spread. High-skill services and finance saw the largest measured effects — around 0.8% annual productivity growth in the more granular study — while low-skill services, manufacturing and construction saw smaller, still-positive gains closer to 0.4%. A broader review of task-level productivity research cited in the 2026 International AI Safety Report found a similar pattern of inflated expectations versus real-world results: controlled studies typically show productivity gains of 20–60% on specific tasks, but most real-world work settings see closer to 15–30% — still meaningful, but well below the headline figures controlled experiments produce. None of this means AI doesn’t help. It means the size of the help depends heavily on the specific task and sector, not on “AI” as a single, uniform intervention.
What This Means for a Small Business or Freelancer
The practical takeaway isn’t “AI works” or “AI doesn’t work” — it’s that broad, sector-wide averages tell you very little about your specific situation, and self-reported impressions of productivity are known, in this research, to run ahead of what shows up in actual output. Before assuming a tool is paying off, measure it directly: track a specific, countable outcome (completed deliverables, response time, revenue per hour worked) before and after adopting a tool, the same discipline this site laid out when covering how to measure automation results with a draft-mode baseline. Concentrate effort on the single highest-leverage function in your business rather than adopting AI broadly and hoping gains average out — the research consistently shows gains concentrating in specific functions, echoing what this site found when looking at where small businesses actually save time with AI and cutting costs with AI in small teams. Starting from a short, deliberately curated list like this site’s best AI tools for small business owners or a more complete small business AI toolkit beats adopting broadly and hoping the averages work out in your favor. And treat any AI agent’s reported time savings the way this site recommended when covering what AI agents can and can’t actually do for a business — verify the output, don’t just trust that it felt faster.
Frequently Asked Questions
Did 89% of businesses really see no benefit from AI?
According to one NBER survey of nearly 6,000 executives, 89% reported no change in productivity, measured as sales per employee, over the past three years. That’s a backward-looking, self-reported figure from one specific study — a separate NBER paper using different survey data found positive, measurable gains already happening at many firms.
Why do two NBER studies on the same topic disagree?
They used different sample sizes, different geographies, and different measurement approaches — one relied on a broad, self-reported yes-or-no impact question across nearly 6,000 firms, the other used revenue-based productivity analysis on a smaller, CFO-weighted sample of about 750. Both findings can be accurate for what they specifically measured without contradicting each other.
Which industries are seeing the biggest AI productivity gains?
High-skill services and finance show the largest measured gains, around 0.8% in annual productivity growth in the more detailed study, while manufacturing, construction, and low-skill services show smaller but still positive gains closer to 0.4%.
Why do executives feel more productive than the data shows?
Researchers attribute the gap to delayed output realization and quality improvements that haven’t yet been captured in measured revenue. Perceived productivity gains from AI consistently run ahead of what shows up in a firm’s actual financial results.
How much time do executives actually spend using AI at work?
Around 1.5 hours per week on average, according to the larger NBER survey, despite more than two-thirds of executives reporting they use AI regularly. Usage intensity remains low even among self-described adopters.
Should a small business trust controlled-study productivity claims like “AI boosts output by 40%”?
Treat them cautiously. A review of task-level research found controlled studies typically show gains of 20–60%, but most real-world work settings see closer to 15–30% — still meaningful, but consistently lower than headline figures from controlled experiments.
