Based on an online petition website and over 15,000 petitions that received close to 3.8 million signatures, this article shows that the use of this type of source can avoid certain aporia of classical studies of mobilizations. This is achieved by shifting from the analysis of stated behaviours to that of real behaviours, and working on a variety of instances of collective actions. The authors also show that this type of study encounters a difficulty common to work on big data from the Internet: the absence of supervised collection for research purposes, which entails a set of difficulties in processing the data. They argue that these difficulties can be overcome by cross-comparing new data from the database (timestamping, presence of comments, etc.) and by illuminating them by means of other methods traditionally used in social science (interviews, questionnaires, etc.).