AI Survey France

The study includes an individual-level, PCS-specific information treatment. Respondents are randomly assigned to receive or not receive an infographic providing information on the impact of AI on their occupation, along three dimensions: automation, augmentation, and simplification.

Treatment variable:
• Name: AI impact information
• Description: Personalized information on the impact of AI on the respondent’s occupation, based on their Profession et Catégorie Socioprofessionnelle (PCS). The infographic describes three dimensions of AI exposure, following Althoff and Reichardt (2026): automation (the extent to which AI can substitute for human labor), augmentation (the extent to which AI can enhance productivity), and simplification (the extent to which tasks can be simplified for workers). Each dimension is presented on a 0 to 100 scale, accompanied by qualitative labels
and interpretative descriptions. The infographic also includes a summary measure of overall AI impact, examples of tasks affected by AI, and information on the relative ranking of the occupation’s exposure to AI.
• Levels: Treatment (exposure to the infographic) vs. control (no exposure)
• Assignment: Random assignment at the individual level

2026-04-02

2026-04-30

• Perceived impact of AI on own job (FS1): “Overall, do you consider AI to be more of a threat
or an opportunity for your job?”
• Relative exposure beliefs (FS2): “Compared to other jobs in France, how exposed do you
think your job is to AI?”

• Expectations about AI:
– Probability that AI transforms how tasks are performed (AI4a)
– Probability that AI replaces some job tasks (AI4b)
– Perceived effects of AI on employment and productivity in France (AI5b)
• Future adaptation intentions:
– Likelihood of training or skill investment (SK4)
– Likelihood of learning AI tools (SK4)
– Likelihood of job search (SK4)
• Policy preferences:
– Support for redistribution policies (PV1)
– Support for AI regulation (PV2)
– Support for social insurance and compensation for displaced workers (PV3)
• Economic anxiety and job insecurity:
– Probability of job loss, wage decline, or adverse job changes (EA3)
– Anxiety about future employment (EA4)
• Job satisfaction:
– Overall satisfaction with current (or last) job (SAT1)
• Workplace attitudes:
– Trust in management (TR2)
– Job autonomy (AUT1)
– Occupational identity (OI1)

Survey experiment

The study is a randomized survey experiment. After answering baseline questions on demographics, employment, AI use, job tasks, and skills, respondents are randomly assigned at the individual level to either receive a personalized information treatment or to a control group that does not receive this information.
Respondents in the treatment group are shown a personalized information treatment based on their Profession et Catégorie Socioprofessionnelle (PCS), a classification of occupations into approximately 400 categories. Respondents in the control group do not receive this information.
The treatment consists of an infographic describing the impact of AI on the respondent’s occupation along three dimensions following Althoff and Reichardt (2026): automation (the extent to which AI can substitute for human labor), augmentation (the extent to which AI can enhance
productivity), and simplification (the extent to which tasks can be made easier to perform).
Each dimension is presented on a 0 to 100 scale, accompanied by qualitative labels and short interpretative descriptions. The infographic also includes a summary measure of overall AI impact, illustrative examples of tasks affected by AI, and information on the relative ranking of the occupation’s exposure to AI, both across all occupations and within similar occupational groups.
The exposure measures underlying the treatment are generated using a large language model based on detailed occupational descriptions, following recent approaches (Felten et al., 2018, 2023; Eloundou et al., 2024; Althoff and Reichardt, 2026).
Following the treatment, all respondents are asked a common set of questions measuring beliefs, expectations, and policy preferences.

Qualtrics's built-in randomization.

Individual

No

approximately 5,000 individuals

approximately 5,000 individuals

approximately 2,500 individuals in the control group and 2,500 individuals in the treatment group