Field note · What we found
One in three UK businesses now use AI. One in ten use it seriously.
The Office for National Statistics has been asking UK businesses the same question since 2023, and in July 2026 it published the run. Adoption has nearly tripled. Depth has barely moved. And more than half of employees say they use AI at work, against a third of the businesses they work for — which is the finding a compliance officer should read twice.
The short answer
In June 2026 the Office for National Statistics found that around 35% of UK businesses with ten or more employees use at least one AI technology, up from around 12% in September 2023. Adoption is shallow: adopters use about 1.6 technologies each and only 10% say they use AI extensively. Large language models are the most common technology (18%), then visual content creation (16%). And 55% of employees say they use AI for work — twenty points more than the businesses that employ them report.
Every vendor deck for the next twelve months will carry the same number: a third of UK businesses now use AI. It is true. The Office for National Statistics published it on 20 July 2026, from the Business Insights and Conditions Survey, wave 159, fieldwork 15 to 28 June, 38,637 businesses responding. That is a real sample from a real statistics office, which makes it rarer than it should be in this subject.
So this note does the thing the deck will not: it reads the rest of the release.
What does “using AI” mean in this survey?
Self-reported use of at least one AI technology. That is the whole test. A firm whose marketing assistant pays for a chatbot subscription counts the same as a firm that has rebuilt its case-handling around one.
The release is candid about this. Adopters use around 1.6 AI technologies each, up from around 1.4 in late 2023 — a rise the ONS itself calls modest. Among businesses with ten or more employees that use AI at all, only 10% report using it extensively. The other nine in ten are somewhere between a trial and a habit.
Which technologies, exactly?
Large language models at 18% and visual content creation at 16% lead, and both are the kind of technology that arrives through a browser tab rather than a project. Data processing with machine learning — the kind that reconciles a ledger or scores a case — is at 12%. Robotics is at 2%. Read together, the ranking says most “AI adoption” in Britain is somebody writing with it, not something running the business.
The number a compliance officer should read twice
The same ONS release sets its business survey beside its household one. In May and June 2026, 55% of employees said they use AI for work or education. The businesses they work for report 35%.
There are honest reasons for a gap — different surveys, different questions, and the household figure includes education. But a twenty-point difference in the same season has one obvious reading: a lot of people are using tools their employer has not adopted, on work their employer is responsible for. In an unregulated business that is a productivity story. In a mortgage brokerage or an advice firm it is a question about what client data has been pasted where, with no record of it.
Only 11% of businesses with ten or more employees say more than half their workforce has had AI-related training. So the usual shape is: staff using it, firm not counting it, nobody trained on it.
What is stopping the rest?
Less than you would think. 41% of businesses with ten or more employees report no barriers to adoption at all. Where a barrier is named, it is expertise rather than money: among businesses with 100 to 249 employees, around 18% say insufficient expertise has delayed adoption. Cost is cited less often.
Two more things the release says quietly. Improving business operations is the most common purpose, reported by over 60% of larger businesses. And most businesses report that using AI has not changed their overall headcount so far — and around 40% of medium and large businesses say they are integrating AI skills by training or retraining the staff they have.
What to do with this
- Treat “are we using AI?” as the wrong question. By the ONS definition you almost certainly are. The useful question is which one process runs on it, what it is allowed to touch, and what it writes down.
- Assume the 55% applies to your staff. Before buying anything, find out what is already in use on client work and write the rule for it. In a regulated firm the record is the product, and unrecorded use has no record.
- Pick depth over breadth. A tenth of adopters use AI extensively. The other nine tenths are paying for a licence and getting a demo. One process, wired in properly, with a named person and a log, beats six subscriptions. Six questions settle which process.
- Do not quote the 35% as a reason to buy. It is a description of the market, not evidence that the thing works. The evidence is in your own numbers, and the arithmetic takes five minutes.
The ONS will ask again. When it does, the number to watch is not the share using AI. It is whether 1.6 has moved.
Questions people actually ask
What percentage of UK businesses use AI in 2026?
Around 35% of UK businesses with ten or more employees reported using at least one AI technology in June 2026, according to the Office for National Statistics (Business Insights and Conditions Survey, wave 159, published 20 July 2026). Among businesses with nought to nine employees the figure was 28%; among those with 250 or more employees it was 49%.
How has UK business AI adoption changed since 2023?
It has almost tripled. The ONS puts self-reported use among businesses with ten or more employees at around 12% in September 2023 and around 35% in June 2026. Depth has barely moved: the average number of AI technologies per adopting business rose from around 1.4 to around 1.6 over the same period.
Which AI technologies do UK businesses use most?
Large language models (18% of businesses with ten or more employees), followed by visual content creation (16%), data processing using machine learning (12%) and image processing using machine learning (6%). Robotics is at 2%. The figures are from the ONS for June 2026 and businesses could report more than one.
Why do more employees say they use AI than businesses do?
In May and June 2026, 55% of employees told the ONS Opinions and Lifestyle Survey they use AI for work or education, while 35% of businesses with ten or more employees reported using it. The surveys differ, but the gap is consistent with staff using tools their employer has not formally adopted. For a regulated firm that is a data-handling and record-keeping question before it is a productivity one.
Sources
- Office for National Statistics, “Artificial intelligence in UK businesses: 2023 to 2026”, published 20 July 2026. Business Insights and Conditions Survey wave 159, fieldwork 15 to 28 June 2026, 38,637 responses, response rate 26.7%. Every business figure in this note — 35%, 12%, 28%, 49%, 58%, 13%, 1.4 to 1.6, 10%, 18%, 16%, 12%, 6%, 2%, 41%, 18%, 11% — is from this release. source ↗ — primary; official statistics from the national statistics office re-checked every 6 months
- The 55% employee figure is from the ONS Opinions and Lifestyle Survey for May to June 2026, as cited in the same release. It is a household survey with a different question and includes use for education, so the twenty-point gap against the business figure is indicative rather than exact. source ↗ — primary, but a different survey from the business figure it is set against re-checked every 6 months
- All figures are self-reported by businesses. The ONS does not audit whether a reported AI technology is in production use; “extensively” is the respondent’s own word. We have not adjusted anything. — a limitation of the source, stated
Checked 11 September 2026. Next scheduled check 10 March 2027. Numbers that move — leaderboards, live indices — are re-checked every 30 days; annual datasets and rules in force every six months; dated research once a year. If something here has gone stale before we got to it, tell us and we will correct it and say what changed.
The tenth that uses it seriously did one process properly.
We build one agent for one job — enquiries, chasing, a first draft, a pack — wired into the diary and the file you actually use, stopping at a named person, and writing down what it did. That is what “extensively” looks like from the inside, and it starts with a conversation about which process, not which product.
Talk to us about this