AI in HR statistics: what the UK measures about AI at work

Ask how many people use AI at work and you will have a percentage inside a minute. Most of those percentages come from a supplier's own panel, state no sample, and cannot be checked. The United Kingdom does publish real numbers on this. They come from the Office for National Statistics, from the Department for Science, Innovation and Technology, and from the regulator that polices how AI is used on job applicants. They are less dramatic than the marketing figures, and considerably more useful in a board paper.

This article sets out what is genuinely published on AI adoption at work and on employee attitudes to it, what each figure actually covers, where the HR-specific gap sits, and what the law expects once AI touches a decision about a person.

Attitudes to a change at work are measurable rather than a mood in the corridor. See how New Dynamics feedback collects what people say about a change while it is still happening, so you are not guessing at the mood afterwards.

For the wider picture on business systems rather than AI, our post on HR software adoption statistics covers what the UK publishes about software, cloud and payroll. This article stays on AI and on the people doing the work.

Where the checkable UK figures come from

Three official releases carry almost everything worth quoting, and each one covers a different population. Reading the base is the whole skill here.

Businesses. The ONS article Artificial intelligence in UK businesses: 2023 to 2026, released on 20 July 2026, draws on the Business Insights and Conditions Survey. The ONS records that "In Wave 159, BICS achieved a response rate of 26.7%, with 38,637 businesses responding to the survey", with fieldwork from 5 June to 28 June 2026.

Individuals. The ONS bulletin Public opinions and social trends, Great Britain: June 2026, released on 17 July 2026, asked adults in Great Britain about AI. For the period 3 to 28 June 2026 the ONS "sampled 8,710 households", with "3,510 individuals in the responding sample for the latest period, representing a 40% response rate".

Business functions. The DSIT AI adoption survey report 2026, published on 29 October 2026, used "computer-assisted telephone interviewing (CATI) between 7 January and 21 April 2026 to achieve 3,503 interviews with businesses". This is the only one of the three that names HR.

Two of those surveys exclude whole parts of the economy. BICS leaves out agriculture, public administration and defence, public provision of education and health, and finance and insurance. The ONS also excludes the smallest businesses from some estimates, because "their relatively large survey weights can have a disproportionate influence on aggregate results". Quote the figure with the base attached or it will be wrong in the next meeting.

AI adoption at work: the business figures

The headline is a tripling in under three years. The ONS reports that "The self-reported use of artificial intelligence (AI) in UK businesses with 10 or more employees has increased from around 12% to around 35% since late 2023, with larger firms more likely to have adopted AI".

Size matters more than any other split. The same release finds "28% of businesses with 0 to 9 employees reporting use of at least one AI technology, compared with 49% of those with 250 or more employees". Sector matters nearly as much, with "over half of businesses in information and communication (58%) reporting using AI, compared with much lower levels in construction (13%)".

The second ONS finding is the one that rarely survives into a slide deck, and it is the important one. "AI adoption to date has been relatively shallow among UK businesses with 10 or more employees, with the average number of AI technologies used per adopting business rising only modestly, from around 1.4 to around 1.6 since late 2023." Among businesses that report using at least one AI technology, "only 10% report using AI extensively", and just "15% of businesses (with 10 or more employees) reporting that more than half of their employees use AI as part of their daily work".

By technology, the mix is unglamorous. "Large language models were the most widely used among businesses (with 10 or more employees) in June 2026 (18%); this is followed by visual content creation (16%), data processing using machine learning (12%), and image processing using machine learning (6%)."

DSIT puts the all-sizes number slightly lower, and the direction is the same: "In 2026, one-third of UK businesses were using at least one AI technology (32% compared to 16% in 2025)." Its reading of what AI is doing is blunt. "AI delivered measurable but largely incremental improvements", the most common benefits being "improvements in workforce productivity (71%)" and "the development of more efficient or flexible processes (60%)", with half of adopters reporting a productivity increase that was "generally described as modest".

Two adoption figures matter to HR specifically. The first is training. The ONS finds that "Only 11% of businesses (with 10 or more employees) reported that more than half of their workforce has received AI-related training", even though "training or retraining existing staff" is the most common way employers say they bring AI skills in. The second is employment. "Most businesses report that the use of AI has not resulted in a change to their overall workforce headcount so far", and of businesses using AI to improve business operations, "63% reported no change in worker headcount, while 1% reported increased headcount and 6% reported decreased headcount".

Where effects do show up, they land on exactly the kind of work HR teams recognise. Businesses "using machine learning data processing most commonly report impacts on administrative or clerical roles (41%) and data analysis roles (39%)".

Employee attitudes to AI at work

This is the part of the evidence base that gets quoted least and should be quoted most, because it tells you how a rollout will land.

Asked whether AI will benefit them, adults in Great Britain split three ways in June 2026. The ONS reports 36% "agreed or strongly agreed", 37% "neither agreed nor disagreed" and 27% "disagreed or strongly disagreed".

The movement is in the disagreement, not the agreement.

The ONS is explicit about what that means: the proportion who disagree "has increased from 20% in August 2025 when we last asked this question; this is now at its highest level since we first asked this question in November 2023 (14%)". People have not become more positive. They have stopped sitting in the middle.

On the balance of risk, "Nearly 4 in 10 adults thought AI brings more risks than benefits (38%), while 13% thought it has more benefits than risks, 43% thought there are equal benefits and risks, and 5% said they did not know". That 38% "has increased from 25% in August 2024 when we last asked this question".

Trust is narrow and task-specific. Around one in three adults, 32%, "reported that they would not trust AI to carry out any tasks". The tasks people are most willing to trust AI with are "customer service and support (27%)", "scientific research (25%)" and "creative tasks (24%)". The least trusted are "decision-making in government (4%)", "caregiving (5%)" and "decision-making in business (7%)".

Read that last line next to any plan to let a tool make a call about a person. Seven per cent of adults say they would trust AI with decision-making in business. That is the public mood an HR team is rolling into, and it explains why transparency about automation does more for acceptance than a better model would.

Work is still where people meet AI most. The most common area of knowing use in the past 12 months was "work or education (41%)", ahead of "home (27%)" and "hobbies or interests (23%)". Age dominates: "Around 6 in 10 adults aged 16 to 29 years (63%) and 30 to 49 years (60%) reported that they had used AI for work or education; this is compared with 27% of adults aged 50 to 69 years and 3% of adults aged 70 years and over."

Job anxiety follows the same age pattern, inverted from where you might expect it. "Almost half of adults aged 16 to 29 years (47%) and adults aged 30 to 49 years (49%) reported that they thought AI could put their job at risk, compared with 24% of adults aged 50 to 69 years and 4% of adults aged 70 years and over." The people using AI most at work are the people most worried about what it does to their jobs.

Two ONS employee figures look contradictory until you check the base. The business article states that "Over half of employees (55%) report using AI for work or education, compared with around a third (35%) of businesses reporting use of at least one AI technology", and that estimate covers "employed and self-employed individuals". The 41% in the public opinions bulletin is of all adults aged 16 and over, including people who are retired or not working. Same survey, different population, both correct.

The gap between 55% of employees and 35% of businesses is the single most useful number in this article for an HR audience. It says informal, unmanaged use is already normal, and that the policy conversation is late rather than early. The ONS offers the same reading: "individual employees may use AI tools independently, as part of their day-to-day work activities".

AI in HR statistics: the one official figure, and its limits

Most people searching for AI in HR statistics want a percentage for how many HR teams use AI. For a long time the honest answer was that no official source published one. That has changed slightly, and the change comes with conditions.

The DSIT artificial intelligence adoption research report 2026 asked "Which of the following business areas, if any, at your organisation are using AI today, or plan to at some point in the future?". Its answer is that "In 2026, businesses were most likely to mention administration (81%), marketing (71%) and IT (66%)". HR appears in that same chart at 33%, up from 26% in 2025, on an unweighted base of 1,460 businesses. The report notes that "The use of AI within the business areas of IT, HR and finance increases as business size does", and the HR figure runs from 27% of businesses with 5 to 9 employees to 54% of those with 250 or more.

Three caveats travel with that 33%, and all three come from the report itself.

1. It is use or planned use. The question allows "today, or plan to at some point in the future", so an intention counts the same as a live process.
2. The base is businesses already in the AI market. It is of those who use or plan to use AI, not of all UK employers, so it is not an HR adoption rate for the economy.
3. It names a department, not a process. Nothing in the release tells you whether AI is drafting a job advert, sifting applications, summarising review evidence or answering policy questions.

So the gap is narrower than it was, and it is still real. No official United Kingdom statistic measures AI use inside a specific HR process. There is no published figure for AI in performance reviews, in shortlisting, in absence casework or in employee listening. Percentages you find for those come from supplier research, usually on a panel of that supplier's customers, and a panel describes the panel. Our post on AI HR tools sets out what these tools actually do, and our post on AI recruiting tools covers the hiring end. For the figures the United Kingdom does publish on work, pay and the workforce, the HR statistics hub collects them with the publisher and the period attached.

What the law expects when AI touches a hiring decision

An adoption rate tells you what others are doing. It does not tell you what you owe the person on the other side of the decision. For an HR audience that duty is the part that bites.

This section describes the position in the United Kingdom only, under the UK GDPR and the Data Protection Act 2018, as updated by the Data (Use and Access) Act 2025. It is general information rather than legal advice. Take your own advice on your own process.

The Information Commissioner's Office is the UK regulator here, and it has been working on recruitment specifically. In a news release on 31 March 2026 it said that "Over the last year, we have engaged with more than 30 employers to better understand the use of ADM in recruitment", that it "wrote to 16 organisations likely to be using ADM to make decisions about candidates", and that in 2024 it audited providers of AI recruitment tools and "made almost 300 clear recommendations for providers and developers to improve compliance with the law". Its expectations of employers are to proactively monitor for bias, be transparent with jobseekers and explain rights to recourse.

The definition that matters is narrow. The ICO describes automated decision-making as a decision that "is based solely on automated processing (ie there is no meaningful human involvement in the decision)" and that "has a legal or similarly significant effect on a person". Its guidance is clear that a recruitment decision counts, including "a decision about whether to shortlist a candidate, recommend them for interview, reject them or promote them".

The ICO's draft guidance on automated decision-making and profiling for recruitment and selection then sets a test that most HR processes would currently fail. It says you must ensure "a human has the power to disagree with the AI recommendations or predictions, and can overturn them", and it states the consequence plainly: "If a human has no power to overturn the AI recommendations, the recruitment decision has been made by solely automated means, and there has been no meaningful human involvement. This is the case even if the human has reviewed the information."

Note the status of that document. The ICO says "Our consultation on this draft guidance is now closed. The final version will be published in due course." The direction is settled even if the wording is not final.

Three practical consequences follow for a people team.

• A rubber stamp is not human involvement. If the recruiter cannot realistically reject the tool's ranking, the tool made the decision. The ICO's bad-practice example is a company that uses test scores "to eliminate half of the candidates, without a human having considered their overall performance", which it calls unlawful absent an exception.
• Keep the override log. The guidance asks you to "keep a record of each time a human reviewer overrides an automated recruitment decision", because that record is how you evaluate both the system and the quality of the human review.
• Tell people before, not after. Candidates must be given "meaningful details about the logic involved and the significance and likely consequences", and a straightforward way to contest a decision and ask for human intervention.

Business practice is drifting the wrong way on oversight. DSIT reports that "Most businesses (81%) reported applying some level of human input or oversight to AI-generated outputs, although this had declined slightly since 2025", and that "56% of businesses who had adopted AI, and 60% of those planning to use AI" said they had concerns about bias and lack of accuracy. Concern is high and oversight is slipping at the same time.

The AI statistics worth producing yourself

None of these needs a benchmark to be worth having, and every one of them is available inside your own organisation.

1. Share of staff who use AI for work, asked the ONS way. Use the ONS wording about use for work in the past 12 months so your figure sits next to a published one. Compare it with what your systems show, because the gap between the two is your shadow usage.
2. Your own version of the attitude question. Ask whether people agree AI will benefit them, with the same five-point scale, and track it. Watch the middle category rather than the average, because that is where the ONS saw movement.
3. A list of every people decision AI currently touches. Adverts, sifting, scheduling, note taking, review summaries, policy answers, casework drafting. Name the process, the tool and who can overturn it.
4. Override rate on every assisted decision. How often a human disagreed with the recommendation, and what happened next. A zero override rate is a finding, not a clean bill of health.
5. AI training coverage. The proportion of your workforce that has had AI training, measured the way the ONS measures it, so you can say whether you are inside or outside the 11% of businesses where more than half the workforce has been trained.

Our guide on measuring AI adoption sets out how to collect these without turning the exercise into surveillance.

What usually goes wrong

• Quoting a business adoption rate as an employee figure. Around 35% of businesses report using an AI technology. Around 55% of employees report using AI for work or education. They answer different questions and both are published by the ONS.
• Reading rising adoption as rising depth. The average adopter uses around 1.6 AI technologies, and only 10% of adopters describe their use as extensive.
• Treating attitudes as warming. Agreement that AI will benefit people has barely moved in three years. Disagreement has grown, and the share saying risks outweigh benefits has risen from 25% to 38%.
• Presenting the DSIT HR figure as an HR adoption rate. It covers use or planned use among businesses that already use or plan to use AI, and it names a department rather than a process.
• Assuming a human in the loop is enough. Under the ICO's test, a reviewer who cannot overturn the recommendation is not meaningful human involvement at all.
• Rolling out tools before training anyone. Only 11% of businesses with 10 or more employees have trained more than half their workforce, which is the cheapest gap on this list to close.
• Skipping the override log. Without it you cannot show the human review worked, and that is the evidence a regulator or a tribunal would ask for.

Frequently asked questions

What percentage of UK businesses use AI?

The ONS reports that self-reported use of AI in UK businesses with 10 or more employees has increased from around 12% to around 35% since late 2023, with larger firms more likely to have adopted AI. The DSIT adoption survey, which includes smaller businesses, found that one-third of UK businesses were using at least one AI technology, at 32% compared with 16% a year earlier.

How many employees use AI at work in the UK?

Over half of employees, 55%, report using AI for work or education, against around a third of businesses, 35%, reporting use of at least one AI technology. The employee estimate covers employed and self-employed individuals in Great Britain and may include some use outside formal business activities. Among all adults aged 16 and over, 41% said work or education was an area where they had knowingly used AI in the past 12 months.

What are employee attitudes to AI at work?

In Great Britain in June 2026, 36% of adults agreed or strongly agreed that AI will benefit them, 37% neither agreed nor disagreed and 27% disagreed or strongly disagreed. The disagreement figure is the highest recorded since the question was first asked, up from 14% in late 2023. Nearly four in ten, 38%, think AI brings more risks than benefits.

Are there official AI in HR statistics for the UK?

Almost none. The DSIT adoption survey names HR as a business area, at 33% of businesses that use or plan to use AI, rising to 54% among those with 250 or more employees. That is use or planned use, among businesses already in the AI market, and it describes a department rather than a process. No official source measures AI inside shortlisting, performance reviews, absence casework or employee listening.

Does UK law allow AI to reject a job application?

In the United Kingdom, data protection law restricts decisions based solely on automated processing where they have a legal or similarly significant effect, and the ICO treats shortlisting, recommending for interview, rejecting and promoting as decisions of that kind. You can use AI to assist, provided there is meaningful human involvement, which the ICO says requires a human with the power to disagree with and overturn the recommendation. This is general information rather than legal advice.

How much AI training are employers actually giving?

Not much, by the ONS measure. Only 11% of businesses with 10 or more employees reported that more than half of their workforce has received AI-related training, even though training or retraining existing staff is the most common way employers say they integrate AI skills. Around 62% of businesses citing a lack of AI expertise as a barrier report training or retraining staff, against around 26% of those reporting no barriers.

Your next step: measure AI use where you can see it

1. Ask your own people the ONS question about using AI for work, and the ONS question about whether they think AI will benefit them. Two questions, repeatable every year.
2. Write down every people decision AI already touches, with the tool named and the person who can overturn it named too.
3. Start the override log this cycle, even on one process, and review it after a month.
4. Count your AI training coverage, then decide whether 11% of businesses training more than half their workforce is a bar you want to clear or beat.

When you are ready to turn that into a measurement plan, read and download our guide to measuring AI adoption. The guide is free to read, and the PDF uses our short download form.

Want to know what your people really think before the next tool lands? Book a New Dynamics demo and bring the questions you want asked. You can also email contact@new-dynamics.com.