AI skills demand in the EU

However, there are large regional differences in AI usage rates between EU Member States. Denmark, Finland, Sweden, Belgium and Luxembourg are the leading countries, with AI usage rates of 34% to 42%, whereas only 5% of companies in Romania used AI in 2025. Several aspects could explain these regional differences. For example, tech-savvy sectors like IT could be over represented in some EU countries and therefore drive AI usage rates. Conversely, a lack of AI skills could be an obstacle for companies if they intend to implement AI, but the local labour market does not have enough people with AI skills. Therefore, it is important to analyse not only the AI usage rate of companies, but also the demand for AI skills to better understand companies’ readiness to adopt AI comprehensively. 

Measuring AI skills demand in the EU

The aim of this fact sheet is to analyse the in-depth AI skills demand in all 27 EU Member States and to draw conclusions about the number of highly AI-exposed employees in these countries. There could be a significant gap between in-depth knowledge of AI technologies and general AI knowledge. While general AI knowledge is a basic prerequisite for using AI applications effectively, more in-depth AI knowledge enables employees to develop their own AI applications, for example. In-depth AI skills demand can be measured precisely via data on online job advertisements that reflect the demand side of the labour market. However, the demand in job postings should be viewed only as a lower bound of actual demand, as skill needs can also be met through corporate training or by consulting external IT service providers, for example. Our assumption is that AI skills explicitly mentioned in job postings indicate that AI plays a central role for the tasks to be performed in that job, beyond simply using easily accessible AI applications that do not require in-depth knowledge. For the analysis, we used a dataset from Lightcast (2026) that collects the full texts of online job advertisements for all 27 EU Member States from 2022 to 2025. In total, the dataset includes approximately 182 million unique job postings. To analyse the data, a multilingual rule-based model was developed that classifies job postings based on whether they require in-depth AI skills. The model performs very well across all EU countries (see Büchel et al., 2026). 

However, the Lightcast data systematically overrepresents management, professional and ICT occupations relative to their actual employment share, while underrepresenting elementary, agricultural and construction occupations – a pattern documented across multiple national and EU-wide benchmarking studies (Napierala et al., 2022; Tsvetkova et al., 2024). This bias arises because employers typically rely more heavily on formal online advertising for hard-to-fill, skill-intensive positions – reflecting both recruitment-channel preferences and persistent skill shortages in occupations such as ICT – whereas occupations with high internal promotion rates or informal hiring practices are advertised online far less frequently (Napierala, 2023). Overall, this results in a biased, non representative sample. Analysing the data without correcting for this bias would be likely to overestimate AI job posting shares (Green/Lamby, 2023). To correct for this, post stratification weights were applied. These weights were derived from actual country- and year-specific employment shares per occupation (Eurostat, 2026b). For each occupation, country and year, we calculated a weight equal to that occupation’s true employment share divided by its observed share in the Lightcast sample. Consequently, occupations overrepresented in job postings received a lower weight and underrepresented ones a higher weight – bringing the sample’s occupational structure back in line with actual employment.