Spain buys AI but does not teach people to use it: 40% of workers admit they cannot

The new report from UGT, one of Spain's two largest trade unions, read alongside figures from Eurostat, MIT and McKinsey, paints an uncomfortable picture: adoption of artificial intelligence is running ahead of training. And the agentic phase is about to widen that gap.

The number that sums up the year

Four in ten people in work in Spain admit they do not know how to use a generative AI tool. That is almost exactly the share among the unemployed (41.7%), which tells you a good deal about how much AI training is actually happening inside companies.

The figure comes from the UGT report Brecha Digital e Inteligencia Artificial: la nueva inequidad ("The digital divide and artificial intelligence: the new inequality"), published on 7 September 2026. It is the first study to look at Spain's AI gap from a labour perspective: not how many companies use AI, but who really uses it, what for, and why millions of people are still outside.

The rest of the picture points the same way. Fourteen million people in Spain use generative AI, 37.9% of the population; more than 23 million never use it, and of those, at least five and a half million do not even know it exists. Six in ten people in work do not use it at all.

The technology is there. The budgets are there. So are the roadmaps, presented to the board.

What is missing is the cheapest and hardest part: training people.

The paradox: AI comes in through the front door at home, not the one at work

The main use of AI in Spain is private (30.2%). Professional use sits at 17.9% and educational use at 16.2%. As UGT points out, nearly 23 million people work in Spain and around ten million study: the fact that work and study use are almost level shows that AI has not yet entered working processes.

The company figures confirm it:

• Only 21% of Spanish companies have AI running in their operations. Among those with fewer than 10 employees it falls to 13.4%.
• 42% of large companies (more than 250 employees) had no such technology in 2025.
• One in three has no ICT specialist on the payroll.
• And on training: only 2 in 10 companies with more than 10 employees ran any ICT training for their staff in 2024. Among those with fewer than 10 employees, which are 95% of the country's business fabric, it was 3 in 100.

It is worth underlining what that last figure measures: training in information technology in general, not in artificial intelligence. The report itself finishes the thought: if it were narrowed to reskilling in AI, the numbers would be "sonrojantes", blush-making.

It is the digital version of buying a fleet of vans for a team with no driving licences. The vans show up on the balance sheet and can be seen from the street; the licence cannot be seen, takes time, and looks good on no slide.

A gym makes nobody thinner by having members. A company does not transform itself by having licences.

The three gaps stacking up

The age and education gap. Among 16 to 24 year olds, generative AI use reaches 64% across the EU; among 65 to 74 year olds it drops to 7%. In Spain, the distance between the youngest and the oldest bracket is 70 points. Educational level opens another gap of more than 50 points, and even so, 44% of people with a university degree never use a chatbot.

The regional gap. Catalonia leads with almost 45% of its population, followed by Madrid (43.4%) and the Balearic Islands (42.7%); at the bottom sit Extremadura, Castilla-La Mancha and the Basque Country. On use at work, only Catalonia and Madrid pass 20%, and seven regions do not reach 15%. In Andalusia, 16.8% of companies with ten or more employees used AI, against a national average of 21.1%.

The gender gap. Women are 55.5% of those who never use AI, ten points above men. In Andalusia, for example, 35.7% of men use it against 31.8% of women.

And one figure dismantles the technical excuse: 94% of the population owns a smartphone, and the initial learning curve of a chatbot is close to zero. The brake is neither access nor difficulty. It is the lack of perceived usefulness and, above all, the absence of any real integration of AI into how work is organised. When UGT asks why people do not use it, 67.1% answer that it "was not necessary", 43.4% that they did not know how, and 24.7% that they did not know it existed.

What about Europe? The problem is not where Spain sits, but who it is compared with

A cliché is worth dismantling here. On individual use, Spain is not at the back: at 37.9% it is above the EU average (32.7%) and ahead of Germany (32%) or Italy (19.9%).

Source: Eurostat, ICT survey of households 2025 (people aged 16 to 74, use in the previous three months).

The real distance is not from the European average, it is from the countries leading the table. And it shows up more clearly in the business fabric: 20% of EU companies used AI in 2025, with Denmark at 42% and Finland at 37.8%, almost double Spain.

The difference between those countries and this one is not technological: it is skills policy. Finland, Estonia, Singapore, the United Arab Emirates, India, Korea and the EU itself, through its national digital skills coalitions, are already running mass AI qualification programmes, with specific modules for the self-employed and small businesses. Spain still has no equivalent plan, and that is exactly what UGT calls for in the report: a national AI skills plan and a national digital divide observatory.

Why it matters to companies

This could look like a matter for education policy. The bill, however, is paid by the profit and loss account.

The MIT study The GenAI Divide: State of AI in Business 2025 concluded that only 5% of generative AI pilots in large companies achieve a measurable impact on revenue. The other 95% stay in the experimental phase. The cause the authors point to is not the models, which are the same in Helsinki as in Tarragona: it is poor integration into processes.

McKinsey's State of AI 2026 survey points the same way: although 80% of people using AI say it improves their personal productivity, only 37% of organisations attribute any EBIT impact to AI, and barely 6% clear the high-performer threshold. What sets that 6% apart is that they redesign entire workflows instead of slotting tools into processes that already existed.

Redesigning a workflow takes people who understand both the process and the tool. In other words: training. Which closes the equation by itself: without in-house competence there is no integration, and without integration there is no return. It is the same conclusion the Bank of Spain data pointed to in an earlier piece on this blog, where the barrier to AI in small companies turned out not to be money.

And the demand is already there. A survey published in December 2025 by RRHH Digital put at seven in ten the workers asking their companies for urgent AI training; in that same study, 63.7% said their company had not made these tools available to them over the previous year and only 16.9% had been given official access. The barometer by Planeta Formación y Universidades, with 3,710 employees surveyed, found that only one in ten Spanish workers receives AI training.

There is no resistance to change. There is unmet demand. And that margin does not last forever: someone who asks for training for two years and does not get it stops asking, or looks for it at another company.

The second phase is already here (and it will widen the crack)

The UGT report itself places the turning point at "generative and agentic" AI. The question is no longer how to write a good prompt, but how to integrate agents that carry out complete tasks: handling a customer, drafting a note, preparing a proposal.

That second phase multiplies the requirements of the first. It demands orderly data, and most Spanish small businesses keep theirs in loose spreadsheets. It forces a rethink of how work is shared out. And it needs qualified human supervision: not from someone who "knows ChatGPT", but from someone who understands the business process and can tell when the agent is getting it wrong.

McKinsey already sees agentic AI consolidating in 2026, alongside the finding that 20% of organisations see adoption limited by running costs and that 39% expect AI-related headcount reductions over the coming year, against 32% the year before.

If the first wave left 62% of the population out, the second will leave out the companies that do not take training seriously. Writing a prompt can be learned in an afternoon. Redesigning a process so an agent can run it, with its controls and its accountabilities, cannot.

What can be done: 5 steps for a small business (without dying in the attempt)

1. Audit who uses what

An anonymous five-question internal survey: do you use AI?, which one?, what for?, does the company pay for your subscription?, have you been trained? The real map always beats the imagined one. Remember the figure above: private use of AI is almost double professional use, so there is probably more of it going on than management thinks, paid for out of employees' own pockets and with no policy behind it.

2. Pick 3 candidate processes

Do not try to transform everything. Choose three where AI already has proven use cases: customer service, proposal writing, data analysis. Start there. Three processes done properly convince more than fifteen half-finished pilots, and they are the exact opposite of the pattern MIT identifies in the 95% that produces no return.

3. Train on the process, not on the tool

A generic ChatGPT course takes 2 weeks and is forgotten in 1 month. Training on "how to use AI to answer tenders in your sector" takes 8 hours and sticks. The difference is not the length: it is that the second one gets used the following Monday on the company's real documents.

4. Measure hours recovered, not vanity

What matters to the CFO is not how many prompts your team writes, but how many hours a month it saves. Quantify it from day 1: average time for the task before, average time after, number of times a month. It is a subtraction, not a consultancy project.

5. Name an internal "AI champion"

One person, not a department, who acts as the reference point, filters the noise and shares what they learn. Someone from operations or administration with some curiosity, not necessarily the most technical profile. Without that figure, the plan dies within three months: the knowledge stays with the four people who went on the course and nobody passes it on.

What does it cost, and what does it give back?

From our own experience training small companies, a plan like this runs between €2,000 and €5,000 in the first year, counting specific training, licences and the internal hours of whoever leads the process.

On the return, it pays to be honest, because the figures in circulation come from very different places. The study The Business Opportunity of AI, which IDC runs on behalf of Microsoft, calculated an average return of 3.5 dollars for every dollar invested, with 5% of organisations reaching 8 dollars; the following edition raised the average to 3.7 and put the leaders at 10.3. These are self-reported figures from a vendor-sponsored study, so they are best read with the accounts open next to them.

The counterweight is the independent data already seen above: 5% of pilots with measurable impact (MIT) and 37% of organisations attributing any effect to their EBIT (McKinsey). The practical conclusion is the same in both cases: the return does not come from the licence, it comes from the integration, and integration depends on someone in the house knowing how to do it.

Put in hours, which is how your team will understand it: if a €3,000 plan frees up four hours a week for three people, it has paid for itself before the first quarter is out.

The question is no longer whether your company will use AI

It is whether your people will know how to direct it.

The technology can be bought in an afternoon. The competence to use it is built over months, and only if someone decides to build it. Spain has spent two years accumulating the first and postponing the second, just as the wave arrives that demands more judgement, not less.

At IAescola we train small companies and teams on real processes, not on fashionable tools: you come in with your documents, your tenders and your emails, and you leave with something that works on Monday. If you would like us to look together at where to start in your company, let's talk.

Has your company already trained its team on AI? Or are we still at the PowerPoint stage?

Sources

• UGT, Brecha Digital e Inteligencia Artificial: la nueva inequidad (September 2026): full report and executive summary, in Spanish.
• UGT Andalucía, briefing note on the same report (September 2026).
• Eurostat, 32.7% of EU people used generative AI tools in 2025 and the Use of artificial intelligence by individuals series.
• MIT NANDA, The GenAI Divide: State of AI in Business 2025.
• McKinsey, The State of AI: Global Survey 2026.
• IDC for Microsoft, The Business Opportunity of AI (2023 and 2024 editions).
• RRHH Digital, survey on AI training in Spanish companies (December 2025).
• Planeta Formación y Universidades barometer on AI in Spanish business (2025).