Social media has instituted new parameters for the political conversation in the digital public sphere. Previous research had identified several of these new phenomena: political polarisation, hate speech discourses, and fake news, among others. However, little attention has been paid to the users’ geographical location, specifically to the role location plays in political discussion on social media, and to its further implications in the digital public sphere. A priori, we might think that on the digital landscape geographical restrictions no longer condition political debate, allowing increasingly diverse users to participate in, and influence, the discussion. To analyse this, machine learning techniques were used to study Twitter’s political conversation about the negotiation process for the formation of the government in Spain that took place between 2015 and 2016. A big data sample of 127,3 million tweets associated with three Spanish cities (Madrid, Barcelona, and Valencia) was used. The results show that the geographical location of the users directly affects the political conversation on Twitter, despite the dissolution of the physical restrictions that the online environment favours. Demographics, cultural factors, and proximity to the centres of political power are factors conditioning the structure of digital political debate. These findings are a novel contribution to the design of more effective political campaigns and strategies, and provide a better understanding of the dynamics of the digital public sphere provided by Twitter