How many businesses use AI in France: what the surveys say

The answer: between 18% and 67%, depending on who counts and how

The question has several published answers, each accurate within its own scope. In 2025, 18% of enterprises with 10 or more employees reported using at least one AI technology (Insee, the French ICT usage in enterprises survey). Eurostat, which publishes the same survey for the whole EU, gives 18.16% for France and 19.95% for the EU.

In 2026, 40% of French small businesses and SMEs reported using at least one AI solution, rising to 53% among SMEs (France Num barometer, carried out by Crédoc for the Directorate General for Enterprise).

In early 2026, 67% of French firms with 20 or more employees reported using generative AI, mostly on a limited or experimental basis (Banque de France, survey of about 7,000 firms in manufacturing, construction and non-financial market services excluding wholesale and retail trade, carried out from 25 February to 3 April 2026). For non-generative AI, the share falls to about a third.

Three things separate these figures: who was surveyed (from zero, 10 or 20 employees, and not the same sectors), the question asked (organised use chosen from a list of technologies, tools used for work purposes, or generative AI alone, experimental use included) and when the data was collected (early 2025, or spring 2026). Setting them side by side as measurements of the same thing is misleading.

Four sources, four scopes

The table sums up what each source counts. The sources and their addresses are listed at the foot of the page.

What explains the gaps between the figures

Insee and Eurostat are not two measurements. They publish the same survey, whose concepts and questionnaires are harmonised across European countries, at two levels: 18% and 18.16% describe the same result.

The real gap lies between the official survey and the France Num barometer. Scope does not explain it: France Num adds businesses with fewer than 10 employees, which use AI less than SMEs (38% against 53%), and leaves out those with 250 or more, which use it most. In the band both surveys cover, 10 to 249 employees, the gap persists: 15% to 31% according to Insee, 48% to 62% according to France Num. What remains is the question and the date.

The question first. Insee asks about organised use within the enterprise’s departments; an employee who uses a chatbot on their own initiative is in principle not counted. France Num asks whether the business uses AI tools for work purposes, with no such restriction.

Then the date. A year separates the two collections, and use is rising fast in both series: from 10% to 18% between 2024 and 2025 for Insee, from 26% to 40% between 2025 and 2026 for France Num. These publications do not allow the effect of the question to be separated precisely from the effect of the year.

The Banque de France measures something else again. Its 67%, measured on firms with 20 or more employees, is comparable neither with the Insee figure nor with France Num: it covers a single family of technologies, generative AI, counts experimental use and excludes trade and finance. Below 20 employees, the Banque de France considers its sample not representative enough.

By company size: the gap between small and large

In every survey, use rises with size. According to Insee, enterprises with fewer than 50 employees, 85% of the scope, use AI half as often as those with 50 to 249 employees, and almost four times less often than those with 250 or more. Insee attributes this to the fixed costs of adoption. Between 2023 and 2025, the gap between enterprises with fewer than 50 employees and those with 250 or more widened from 16 to 43 points.

The Banque de France sees the same slope in early 2026: 64% of firms with 20 to 49 employees use generative AI, and 83% of those with 1,000 or more; for non-generative AI, 31% and 60%.

What holds businesses back: perceived usefulness, expertise, data

Among enterprises with 10 or more employees that do not use AI, the main reason is neither cost nor technology: 71% see no use for it. That reason weakens with size, while lack of expertise and data difficulties grow (Insee, 2025).

• Enterprises already using AI are held back too: 53% by a lack of expertise, 43% by data protection concerns, 42% by a lack of legal clarity, 33% by costs that are too high (Insee).
• Eurostat asks a different population: enterprises that considered AI but did not adopt it. Across the EU, lack of expertise then comes first (70.89%) and lack of usefulness last (20.68%). Same word, “barrier”, two populations.
• A perceived lack of usefulness is a matter of quantification more than technology: a use is judged on a specific process, with time spent, volume and cost. The gain simulator gives an order of magnitude from your own assumptions.

Declared use is not integrated use

All these surveys count businesses that report some use. None of them checks whether AI is built into a process, with an owner, human review and a measured gain.

The published details show it. According to the Banque de France, reported use of generative AI is mainly limited or experimental. In France Num, 52% of the businesses using AI rely on free solutions only, and the most common uses are content generation (34% of small businesses and SMEs) and chatbots or assistants (24%). In the Insee survey, a quarter of user enterprises concentrate on a single technology, most often generative AI, and a third on written-language analysis. At the other end, 10% use at least five technologies; nearly two thirds of them developed these in-house, and they employ 48% of the staff of all user enterprises.

Set against the tiers of AI maturity, these figures read as follows:

• Tier 1, ad hoc use: a few employees use consumer tools. That can be enough to answer “yes” to France Num; Insee in principle excludes it.
• Tier 2, fragmented initiatives: teams have their own tools, with no common framework. Organised use within one department is enough to be counted by Insee.
• Tier 3, integrated AI: AI supports core processes, with an owner for each use case and traced human review. None of these surveys measures this tier.

Frequently asked questions

Sources

Method: external figures checked against the source cited; SHIFTLABS figures are presented as estimates or examples.

• Insee, Insee Première no. 2120 (21 July 2026), “Les technologies de l’information et de la communication dans les entreprises en 2025” (in French): usage rate, size, reasons for non-use, scope and method (accessed 18 October 2026)
• Eurostat, Statistics Explained, “Use of artificial intelligence in enterprises” (data extracted in December 2025): EU average, reasons for non-use (accessed 18 October 2026)
• Eurostat, dataset isoc_eb_ai, “Artificial intelligence by size class of enterprise” (updated 15 June 2026): values for France and the EU, 2025, enterprises with 10 or more persons employed (accessed 18 October 2026)
• Directorate General for Enterprise, press release “Baromètre France Num 2026 : quatre TPE-PME sur dix utilisent désormais l’intelligence artificielle” (28 September 2026, in French; accessed 18 October 2026)
• France Num, “Baromètre France Num 2026 : le numérique et l’intelligence artificielle dans les TPE et PME” (published 17 September 2026, updated 12 October 2026, in French): methodology (accessed 18 October 2026)
• Crédoc for the DGE, France Num barometer, 2026 survey results, full report (2 October 2026, in French): question asked, size, types of use (p. 54), paid solutions (accessed 18 October 2026)
• Banque de France, Bulletin de la Banque de France no. 266, article 1, Véronique Genre and Léo Parpais, “L’IA s’installe dans les entreprises françaises : une diffusion rapide, mais des gains de productivité encore limités” (online 3 September 2026, in French): 67% for generative AI, about a third for non-generative AI, survey (accessed 18 October 2026)
• Banque de France, same article, PDF version (Bulletin for September-October 2026, in French): scope and sectors (note 2), definitions of generative and non-generative AI (note 3), 20-employee threshold (note 4), share by size, questionnaire in the annex (accessed 18 October 2026)