UK, who is paying for artificial intelligence
A preliminary study based on millions of UK bank accounts finds that 5.1% of adults pay for an AI subscription out of pocket, from 20.3% of IT professionals to 1.7% of social care workers. In my opinion, what employers pay for remains out of view.
In the United Kingdom, one adult in twenty pays out of their own pocket for a subscription to an AI assistant: 5.1% in June 2026. That is the claim of a study published on 5 November by Nikhil Datta, Peter John Lambert and Patrick Schneider. They did not conduct a survey. They analysed transactions from millions of current accounts and card accounts, covering the period from 2022 to mid-2026, made available through open banking, the system that lets authorised services see the data in your bank account with the account holder’s consent. Lambert, an economist, presented the results in a thread on X. It is a working paper, a preliminary study not yet reviewed by other scholars, and the numbers are best read that way.
The first thing that stands out is the speed: subscribers have been doubling every twelve months since 2024. In the chart, though, the band of uncertainty widens precisely in the last stretch, where the curve climbs most.
Then there is the question of who. Adoption among under-30s is five times higher than among over-60s, even after accounting for salary, gender and region. Salary alone explains little: between the poorest fifth and the richest there is a factor of 1.8, which shrinks sharply once occupation and demographics come into play. The authors say they are surprised, given that free versions exist, and suggest as an explanation students and young people trying to impress the boss. That is their hypothesis, not a measured fact.
Even the gap between men and women (adoption is 27% higher among men) tells a different story from the one it first seems to tell. According to the author, among today’s new subscribers men and women are equal in number: the gap arose, and persists, because the first people to adopt the tool were mostly men. A snapshot that mixes those who arrived first with those arriving now can say the wrong things about the present. The thread does not say whether the same holds for age and for jobs, and that is the first thing I would ask of the data.
The sharpest divide is between jobs. If I read the chart correctly (the axis says “share of adopters”, but the bars add up to far more than a hundred, so they must be percentages within each occupation), 20.3% of IT professionals pay, 16.0% of university and college lecturers, and 15.8% of those working in finance and insurance. At the bottom are social care workers (1.7%), nurses and midwives (2.5%), firefighters (2.5%), electricians and construction workers (2.9%). Between top and bottom there is a twelvefold gap. I suspect the difference reflects how much a job involves working with text, code and numbers on a screen: there, the tool is easier to justify as a personal expense. The data do not say so, and the bars show no margin of error, which matters in the smaller categories.
There is also something these accounts probably cannot see. If the study measures only subscriptions paid for personally, tools bought directly by companies for their employees will not appear in those personal accounts, though a subscription an employee pays for and later claims back would. That is my hypothesis about what the thread implies, and it needs confirming in the paper. If so, the picture shows who pays for themselves, not who uses artificial intelligence at work. And the question that interests me most stays out of frame: when the company pays, who knows what it has bought, to do what, and who answers for it?
Other cautions. A sample of open banking users is not necessarily representative of the British adult population. The authors say they took the challenges of reweighting and imputing missing data seriously, and acknowledge that much work remains to be done, but I have not seen the details. The thread’s headline speaks of a market worth $1.4 billion a year: in the charts I have seen the spending is in pounds, and I cannot see where that figure comes from or at what exchange rate.
For now, the picture is this: paid AI is most common in desk-based knowledge work and rarest in care and manual trades. One hypothesis the data cannot test: if people who pay are heavy users, and heavy users exercise certain skills less, the risk may be concentrated in the very occupations where paid AI is most common. The authors say the study’s website, updated monthly, provides additional results, maps and a downloadable data pack, allowing readers to check the figures for themselves. The monthly updates may show whether paid adoption is spreading beyond the groups that led it, or whether the gaps between jobs and ages persist.