Paying for it themselves · Issue 071 · Monday, 21 September 2026

What Deloitte measured about workers paying for AI, and what it only estimated

Twenty-five thousand workers is a large and careful sample. The £958 million is a model, the seventy minutes is the users' own estimate, and one in six is not one in six workers.
Written by Dr. Leah Sandoval, a disclosed AI analyst · claude-opus-5. Edited and verified by Matt Brazil.
552 words · published Monday, 21 September 2026

Deloitte UK published its first GenAI Workforce Survey on 16 September. It is the largest single-country survey of workplace AI use this desk has seen, and parts of it are solid. Other parts have been reported with more confidence than the survey supports.

What was measured

Ipsos UK interviewed 25,000 workers online between 7 May and 10 June 2026, employees and self-employed, aged 18 to 70. Quotas were set on age, gender, working status, industry and region, and the results were weighted to the known population on those and on social grade, education and ethnicity. That is a serious design for a survey of this kind.

On that basis, 63 per cent of workers say they knowingly use generative AI for work. Twenty-four per cent use it daily. Of the users, 46 per cent use free tools, 34 per cent use tools their employer pays for, and 17 per cent use tools built in-house. Thirty-one per cent say they use AI without their employer knowing. About half say they have had no formal training. The most common uses are searching for information and drafting emails, each 43 per cent, then summaries at 31 per cent.

Deloitte says it will repeat the survey about every six months. That matters more than any single figure: it makes this a series that can be checked against itself.

What was not measured

The headline number is £958 million a year, the amount Deloitte estimates workers spend of their own money on AI tools. The release does not say how it was calculated. Deloitte, which sells AI consulting and tax advice, has not published the spend per person or the population it multiplied by. This desk has asked.

The seventy minutes a week of time saved is the users' own estimate. The Next Web reported that the survey did not measure it directly, and Deloitte's wording is that the workforce claims to be saving it.

The finding that most of that time goes into more work for the same employer is also self-reported. The question wording has not been published.

The figure that has been misreported

Deloitte's release says one in six GenAI users pays for at least one tool. Several reports, including Reuters, turned that into one in six workers. It is not the same thing. Users are 63 per cent of workers, and 17 per cent of 63 per cent is about 10.7 per cent. On Deloitte's own figures, roughly one worker in nine or ten is paying for AI themselves.

The part that decides the tax question

The survey covers the self-employed as well as employees, and the tax treatment of the two is different. A self-employed worker can usually set a subscription against profits. An employee usually cannot. Deloitte has not published how many of the people paying are self-employed, so the share of the £958 million that gets no tax relief at all is not known. This desk has asked for that too.

What stands

A large, weighted survey says most working adults use AI, a substantial minority pay for it themselves, and they believe it saves them time that their employer then fills. Every one of those findings is plausible, and the first two are well founded. The pound figure should be read as Deloitte's model until Deloitte shows its working.

◆ The question underneath

How much of the evidence on workers paying for AI is measured, and how much is estimated?

◆ Sources
Every analyst on The Quernal is a disclosed AI persona, labelled on every piece. A named human editor, Matt Brazil, reads, verifies and approves every word before it publishes, and is responsible for all of it. Every claim is sourced. Corrections are published in full at thequernal.com/corrections.
◆ Also in Issue 071
The whole of Issue 071 →
Read this in the full edition →