Progress regarding the theory and methodology of public opinion polls is now allowing us to envisage experimental survey devices which give more room for open-ended questions, or for sequences which combine open and closed questions. They thus make it possible to build veritable corpuses of "expressed public opinion". This article attempts to present the collections formed by the responses to these devices, by examining their pretension of constituting political corpuses. We do not set out the tools or the methods - in particular the analysis of textual data applied to this type of corpus - but instead we identify the socio-political and linguistic conditions which govern the creation of such corpuses and which also partly determine the strategies for analysis and interpretation. Such an enterprise means looking first of all at the notion of "the political discourses of ordinary citizens", and then examining the survey from a socio-technical angle by analysing, in particular, its pretensions of constituting an enunciation device. Can an opinion poll constitute a "stage" for public debate? This initial question involves, on the one hand, thinking about the survey's interactions and the skills that it mobilises, along with the types of gain in generality created by the survey "stage", which would enable us to liken a response to a survey's open-ended question to a type of political discourse; and, on the other hand, to think about the statistical representativeness, linked to a conception of political representativeness, of these collection and processing devices. After clarifying these conditions, I will suggest we define corpuses "enunciated public opinion" as sequences of open and closed questions, composed of a chain of interactions created by the succession of question-answer couplets ; I will also provide an example of a device.