When will the AI show up?
One question people sometimes ask is if AI is such a massive economic opportunity and showing already historically rapid adoption, how come we don’t already see the impact of it in the GDP statistics? Does this mean that most estimates of its potential impact are wrong and far too overhyped?
The reason that this line of thinking doesn’t follow is, to put it briefly, that the national accounts are very noisy. The margin of error is just too big. Between the wider swings to the macroeconomy, short term structural shocks from Ukraine to tariffs, and the ONS’s own habits of radically revising its numbers drastically later on, its hard to know how to interpret any short term quarterly movement in GDP growth.
On average, productivity growth in the UK has been around 0.5% since 2010 - but with a standard deviation in typical times of about 1.3 percentage points. Headline GDP is noisier still, and that is before you even get to measurement error and revisions. Statistically, that means even a sustained boost of a full percentage point a year would take well over a decade to separate cleanly from normal fluctuations in the numbers.
The chart shows the number of annual observations at which a single, prespecified statistical test has an 80% chance of detecting a sustained boost to UK labour-productivity growth, at 95% confidence (two-sided). Annual growth in output per hour is modelled as independent draws from a normal distribution around a known trend; a boost of b percentage points shifts the mean, and the years required are (kσ/b)² rounded up, where k ≈ 2.80 combines the confidence and power thresholds and σ is the annual volatility of growth. The three settings correspond to the standard deviation of ONS whole-economy output-per-hour growth (series LZVD, May 2026 vintage) over different windows: 0.85pp for 2010–2025, 1.3pp for 1997–2019 and 1.6pp for the full 1972–2025 record.
While feasible, that is a pretty significant boost: most estimates suggest AI’s impact is likely to be something like 10 percentage points over a decade - but crucially only ramping up more slowly in the first few years.
That doesn’t mean you don’t get strong hints before that by looking at cross sector or occupational data of which areas are seeing faster growth. For example, Melek and Miller (2026) find that labour productivity has risen above its pre Covid trend, with this accelerating post the release of ChatGPT - and as you might expect, much of this gain is concentrated in a few core sectors, largely ICT, retail and professional services. Similarly Teutloff (2025) found that demand for potentially AI substitutable skills such as writing and translation on an online freelancing platform declined by 20-50% relative to trend Looking at the bigger picture, Jason Furman argues that on the latest revisions peak-to-peak productivity growth in the US from 2019-2025 is the now the second highest of the last few decades, behind the dot com boom.
But you can always argue back against any of these as a single data point. Is US growth driven by a genuine structural improvement in productivity, or we mismeasuring the spillover impact of a data centre bubble? How much of a supposed post ChatGPT rise was really a rebound from Covid? Can you really extrapolate from freelance roles and what are almost by definition some of the most easily separable and structured tasks from the day to day of a company?
If we want really unambiguous evidence, we will have to wait.
