Every summary of the ONS release said AI is not costing British jobs. We used an earlier wave of the same survey in July to argue the calm was a lag, not an all-clear.

The Office for National Statistics released its analysis of AI in UK businesses on 20 July, drawing on the June wave of the Business Insights and Conditions Survey: 38,637 responding businesses, a 26.7 per cent response rate. It is the largest regular read on this question in the country.

Most of the coverage that followed reached the same conclusion: British firms are using AI to support staff rather than replace them, adoption is up and the jobs are holding. That reading is available in the data. This desk has been here before and did not take it.

On 8 July we used an earlier wave of the same series, which showed roughly a quarter of firms using AI and 5 per cent reporting a cut in headcount. Our Ground Truth desk argued the low figure was not evidence that jobs were safe. It was a count of how many employers had so far chosen to act on a capability nearly all of them now hold. A choice each firm makes separately is not a settled outcome. The calm was the lag.

The June wave does not change that argument, and we are not going to pretend it does. Around half of businesses report no effect on headcount, and just under 7 per cent of medium-sized firms report a decrease. Narrow it to firms using AI specifically to improve operations, which is where cuts would show first, and 63 per cent still report no change against 6 per cent reporting a fall. The Bank of England's Decision Maker Panel found much the same in January.

Read all of that as a count of decisions taken so far, not as a finding about what the technology can do.

What the new wave adds is a measure of depth, and it is the part the summaries left out.

The ONS tracks the average number of AI technologies used per adopting business as a rough proxy for how far the adoption goes. Since late 2023 that average has moved from about 1.4 to about 1.6, while the share of firms with ten or more employees using any AI at all has risen from around 12 per cent to around 35 per cent.

The statisticians draw the inference themselves, saying it implies "relatively limited transformative impacts to date" for most adopting firms.

Two other readings point the same way. Only one adopter in ten says it uses AI extensively, and 15 per cent say more than half their staff use it daily.

The distance between industries is the part that should give any national plan pause: information and communication at 58 per cent, construction at 13. And the technology most people picture when they picture automation, robotics, is used by 2 per cent of firms.

One comparison needs handling with care, because it is the one most likely to be quoted. The ONS sets 55 per cent of employees reporting AI use for work or education against 35 per cent of businesses reporting any use, and reads the difference as people using tools independently of their firm.

The direction is plausible. The size is not. The employee figure comes from a different survey, covers Great Britain rather than the UK, and counts study as well as work. The same release cites St. Louis Fed research putting UK workers nearer 36 per cent, on which the gap all but disappears.

Three limits belong with it. The survey is official statistics in development. It counts AI as a list of named technologies, so a light user and an embedded system answer alike. And it excludes agriculture, energy, finance, public administration and public education and health.

So the honest reading is narrower than either the alarm or the reassurance. Adoption is wide and shallow. The employment effects are small so far, and so far is doing a lot of work in that sentence.