In short
- In 2025, one in four people in work in France and Spain, and one in three in Belgium, used generative AI for their work (Eurostat). Among office employees, a Cegid/OpinionWay survey (2026) finds 70% in France and 82% in Spain.
- Most often, each with their own tool: in France, 67% of office employees who use AI use tools not provided by their company. And 63% of the office employees surveyed in France say their company has put no training on AI tools in place (same study).
- On the side of French companies, the Banque de France sees economic effects that are still limited. Our reading: there are decisions missing that only the company can make (choosing a process, plugging AI into it, supervising it, training the team).
A typical, fictional scenario: Monday morning, in an SME of 30 employees. The accountant rewords a delicate reminder with a free chatbot. The salesperson has an offer translated there for a Spanish client. An assistant pastes in a contract to get a summary. Nobody decided anything: each person found their own tool.
Recent surveys point the same way: AI enters the company through employees, often with tools the company did not provide. Our reading: many companies have tried it, few have organised it. The risk: gains that nobody measures, and data that can leave with no rule.
Your employees already use generative AI, each with their own tool
Generative AI means tools that write, summarise, translate or create images on request, such as chatbots. Eurostat, the statistical office of the European Union, measured in 2025 who uses them for work.
| Country | People in employment |
|---|---|
| Belgium | 32,7 % |
| Spain | 26,1 % |
| France | 24,6 % |
| European Union (27 countries) | 21,8 % |
Share of people in employment aged 16 to 74 (employees, self-employed and family workers) who used a generative AI tool for work in the last three months, in 2025. Source: Eurostat, table isoc_ai_iaiu, database updated on 5 June 2026.
This is the only official measure of individual use at work that we found. It does not break down SMEs and does not say who provides the tool. A survey commissioned by Cegid, a software publisher, from OpinionWay sheds light on this second point (500 office employees surveyed in France and 501 in Spain, in May and June 2026). It describes use that is often outside the company's framework, sometimes called “shadow AI”.
- In France, 70% of the office employees surveyed use AI. Of these, 67% say they use tools not provided by their company, “souvent sans en informer leur employeur” (“often without telling their employer”).
- In Spain, 82% use it. Of those who have used it, the same proportion (82%) admit to using tools not provided by their company, without telling it.
- Tools provided: only 39% of the office employees surveyed in France have professional AI tools provided by their company, against 49% in Spain.
- Training: in France, 63% of the office employees surveyed say their company has put no training on AI tools in place. In Spain, on a differently worded question, 56% have received none from their company.
For French SMEs alone, the study puts forward that 77% of users rely on their own solutions rather than the employer's tools. This figure comes from a sub-sample whose size is not published: a clue, not a measurement.
Data that can leave with no rule
What remains is what employees entrust to these tools. We found no solid public source measuring the share of employees who enter confidential data into them, but the authorities have made it a point of attention. In France, in a sheet for micro-businesses and SMEs, the CNIL and France Num warn that what you enter in a free online tool may be used as training data. Their instruction: “N'introduisez pas de données confidentielles, personnelles ou stratégiques.” (“Do not enter confidential, personal or strategic data.”) The ANSSI, the French cybersecurity agency, recommended in 2024 to “proscrire l'utilisation d'outils d'IA générative sur Internet pour un usage professionnel impliquant des données sensibles” (“ban the use of generative AI tools on the Internet for professional use involving sensitive data”), including contract data. The contract in our scenario is one of them.
First: look, then write the rule
Ask your teams which tools they use and what for, without making it a trial: the aim is to know. Then write the usage rule, as the CNIL and France Num recommend. One page is enough: authorised tools, data never to be entered, review before anything is sent, who to tell in case of doubt.
For the company: declared use, few measured effects
Company statistics count companies, not people: they cannot be compared term for term with the figures on employees.
- Eurostat: in 2025, 15.0% of French companies with 10 to 49 persons employed used at least one AI technology. That is 17.2% in Spain, 28.8% in Belgium and 17.0% in the European Union.
- Banque de France: at the start of 2026, 64% of French companies with 20 to 49 employees say they use generative AI. But it is a use “principalement limité ou expérimental” (“mainly limited or experimental”), and “à ce stade, les effets économiques observés demeurent limités et concernent principalement des gains de productivité” (“at this stage, the economic effects observed remain limited and mainly concern productivity gains”).
The two figures cannot be compared: the question, the threshold and the date differ (Eurostat: at least one AI technology, in 2025; Banque de France: generative AI, experimental uses included, at the start of 2026). Keep the Banque de France's finding: use is spreading fast, its effects remain limited. Other official figures: the 2025-2026 figures on AI in SMEs.
Our reading: when everyone works with their own tool, the time saved stays individual. Nobody has measured the starting point, nothing shows in lead times or margins, and what works for one person is not passed on to the others, especially when the use stays hidden.
What is missing: company decisions
The tools exist, and your employees already know how to use them. Once the rule is written, four decisions, which only the company can make, turn individual uses into results for it.
- Choose a process. Just one to begin with: a frequent task, whose rules you can write down and whose current cost you know (time, lead times, errors). Chasing overdue invoices, sorting a shared inbox, preparing quotes. Without a figure for the starting point, it is impossible to say later whether AI helps. See which process to start with.
- Plug AI into it. Instead of copy-pasting into everyone's own tool, a tool chosen by the company, linked to its software, with access limited to what is strictly necessary. One person has read the terms, which vary by plan and subscription: where does the data go, is it used to improve the model?
- Supervise it. At the start, a person reviews what the AI prepares before it goes out. You track a few indicators: share of proposals accepted, errors, time saved. This tracking is what makes the gain visible.
- Train the team. What must never be entered, how to check an answer, who to tell about a doubt. This is the first obstacle cited: among small companies in the European Union (10 to 49 persons employed) that considered AI without adopting it, about 70% cite a lack of expertise (Eurostat, 2025). On the EU AI Act, see our article on training.
None of this requires being an IT specialist. You need a person from the business, time reserved in their diary, and a decision from the business owner.
A note on Spain
The AEPD, the Spanish data protection authority, advises users of AI chatbots to limit the personal data they expose and to check whether the tool keeps learning from conversations. This advice is aimed at users in general, not employers. The INCIBE, the national cybersecurity institute, advised companies as early as 2023 to avoid giving ChatGPT personal, sensitive or banking data.
Sources (accessed 12 October 2026)
- Eurostat, use of generative AI by individuals, table isoc_ai_iaiu: table and people in employment, 2025
- Eurostat, use of AI in companies and obstacles, table isoc_eb_ai
- Cegid and OpinionWay, press release for France of 3 September 2026 and press release for Spain of 31 August 2026
- Banque de France, Bulletin n° 266, article 1, 3 September 2026
- CNIL and France Num, sheet 4 for micro-businesses and SMEs, November 2024
- ANSSI, Recommandations de sécurité pour un système d'IA générative, 29 April 2024
- AEPD, recommendations for users of AI chatbots, undated
- INCIBE, ChatGPT and the risks for the company, 11 April 2023
Going further
Kendu Intelligence offers to support SME business owners on these decisions: choosing the process, installing an AI agent there that will prepare the work and that your team will approve, and supervising it. Training for your employees is in preparation: training. We also develop custom software with Claude, for which you set the price.
A process in mind? Request an audit: a one-page report on your tools, your processes and the useful agents. It costs €490 excl. VAT; its price is deducted from the quote if you sign it within 30 days.