The survey that could have settled the argument asked 38,637 British businesses a question almost nobody reads

The Office for National Statistics runs something called the Business Insights and Conditions Survey. Every fortnight it goes out to a large sample of British firms and asks what is actually happening: turnover, prices, staffing, and for the last three years, artificial intelligence. Wave 159 was in the field between 15 and 28 June this year. It got 38,637 responses, a response rate of 26.7 per cent, and the results were published on 20 July.

That is a larger and more recent sample than any of the vendor surveys that have shaped the past year of coverage, and it costs nothing to read.

What matters is how it asks. Most attempts to measure AI and work count jobs: how many people were employed before, how many after, and the difference is the answer. That method cannot see anything until the headcount moves. A job that changes completely while the same person keeps doing it registers as zero.

The ONS asks a second question alongside the first. It asks businesses which roles the technology has affected. Not how many people left, but which parts of the work the machine now touches. That is a question about the content of a job rather than the existence of one, and it can be answered while nobody is being made redundant.

The answers are specific. Of businesses using image processing tools, 53 per cent report an impact on administrative and clerical roles, and the same proportion on creative and design roles. Of those using machine learning for data processing, 41 per cent report an impact on administrative and clerical work and 39 per cent on data analysis. More than half of businesses using visual content tools report an effect on creative or design roles.

Set that against what the same survey finds on headcount, where around half of all businesses report no impact at all and just under 7 per cent of medium-sized firms report a decrease. Both are true in one dataset. The roles are moving and the payroll mostly is not.

Two limits any honest reading has to carry. The headline adoption figure covers businesses with ten or more employees, so the smallest firms sit partly outside it. And the survey excludes six sectors entirely: agriculture, oil and gas extraction, energy generation and supply, public administration and defence, publicly provided education and health, and finance and insurance.

That last exclusion matters more than it looks. Finance and insurance is where automated decision making has moved fastest in Britain. It is outside the frame. Anything this survey says about the effect of AI on British work is a statement about an economy with the banks taken out of it.