Study Reveals: How the French Use AI Daily in 2025

Since its launch in October 2024, the Compar:IA platform, developed by the Ministry of Culture and DINUM, has allowed internet users to anonymously compare responses from two generative AI models. In just a few months, it has amassed a vast corpus of French-speaking conversations, providing a unique database on the real uses of artificial intelligence in France.

Using this data, Bunka.ai, a startup from CNRS specializing in human-machine interaction analysis, conducted a study published in July 2025. By analyzing over 175,000 questions and answers from 59,475 distinct users covering more than 30 different models (both proprietary and open source), researchers identified major trends in everyday AI use in France.

A Thought Partner Beyond Just a Tool

One of the key findings of the study is how the French perceive AI. It is primarily seen as a conversational partner for learning, understanding, and refining ideas. According to Bunka.ai, 67% of uses involve a collaborative logic that enhances the user’s capabilities, as opposed to 32% for strictly automated and directive use.

These “augmentative uses” include tasks such as learning (64% of cases), iteration (2%), or validation (1%). The “automated uses” correspond to interactions of the type “a command, a response”, which are more common in professional and creative contexts.

This trend confirms that conversational AI is not only seen as a tool for execution but as a “collaborative learning partner”. For instance, the study shows that users engage AI to explore a scientific concept (79% of interactions in science involve learning) or to develop thoughts in humanities (78%). Conversely, in creative fields, half of the exchanges (53%) take the form of direct content production requests, like writing a story.

Varied Uses of AI in France

Beyond this collaborative relationship with AI, the study highlights the diversity of tasks assigned to the models. Four main types of usage dominate:

• Learning (27% of interactions),
• Seeking advice (19%),
• Creating content (16%),
• Searching for information (16%).

There are also more marginal uses, such as data analysis (6%) or planning (3%).

The topics discussed are equally varied. Conversations related to technology (programming, hardware configuration, cybersecurity) account for 11% of the corpus, closely followed by education and learning (10%). Three other areas each make up 9%: humanities and social sciences, the workplace, and the legal and political sphere. Health and wellness, arts, economy, or nutrition also appear in exchanges, confirming that AI permeates many aspects of daily life.

This diversity is accompanied by strong correlations between subjects and types of tasks. Conversations about health primarily seek advice (50% of cases), while exchanges related to natural sciences are overwhelmingly oriented towards learning (64%). In arts and creation, content production dominates (45%). Users discussing cooking mainly request recipes, thus seeking advice (40%) and information (38%), whereas discussions related to hobbies or social organization often involve planning (22%).

Variations According to Context

The study also highlights differences according to the usage environment. In professional settings, interactions tend to be more directive and oriented towards immediate execution. For example, nearly 46% of work-related exchanges rely on automated requests (writing an email, preparing a summary, providing a document template). In artistic creation, this directive approach is even more pronounced, reaching 53%.

Conversely, in fields related to health, natural sciences, or humanities, the approach is much more collaborative. Users ask multiple successive questions, refine answers, and use AI as a thinking partner, rather than just a tool. In health, for instance, advice requests often come with reformulations and clarifications, showing a desire for dialogue.

The study also notes that the complexity of language varies significantly depending on the topics. Technical domains (technology, science, law, economy) use specialized vocabulary and expert formulations, while exchanges about cooking, arts, or leisure adopt a simpler, more conversational register. This adaptation reflects a natural appropriation of AI by users, who adjust their way of discussing based on the context.