The integration of artificial intelligence into the internal governance of political parties constitutes one of the most consequential and underexplored emerging phenomena of the digital transformation of Western democracies. While the use of AI in electoral campaigns is advancing rapidly, its penetration into internal party organization remains at incipient stages: Sandri et al. (2024) confirm that most European parties are still at proto-AI stages with regard to their internal governance. Nevertheless, recent cases such as the use of Pol.is by the German party Aufstehen to build its political platform with more than 33,000 participants, the vTaiwan experience in algorithmic deliberation of public policies, or the announcement by the Japanese party Saisei no Michi to delegate internal resource management decisions to an AI, signal that this transition is beginning to produce real cases whose democratic implications have not yet been systematically evaluated. It is precisely for this reason that this is the critical moment to build the analytical frameworks and regulatory safeguards that will shape how this adoption unfolds when it becomes generalised.

This paper responds to that imperative by developing an analytical framework that evaluates five categories of AI tools — natural language processing systems, liquid democracy and quadratic voting platforms, preference-aggregation chatbots, algorithmic transparency dashboards, and large language models — against five dimensions of intra-party democratic governance: participation, deliberation, representation, accountability, and inclusion. Unlike normative approaches focused exclusively on AI ethics, the framework incorporates cartel party theory (Katz & Mair, 1995) and the concept of digital plebiscitarianism (Gerbaudo, 2019) to account for the structural organisational incentives shaping technological adoption. The central question is not whether AI can strengthen internal democracy, but under what conditions parties have structural incentives to use it democratically.

The central thesis is that AI is likely to function as a multiplier of pre-existing political conditions rather than as an autonomous democratising force. The analysis is organised around a bidimensional matrix integrating democratising potential and risk of capture for each tool-dimension combination, operationalised through eight falsifiable hypotheses with specific validation designs. Three findings emerge: no single tool simultaneously enhances all five governance dimensions; generative AI presents the highest risk profiles in accountability and inclusion; and digital inclusion is the most transversally vulnerable dimension across all tool categories analysed.

The results carry direct implications for the ongoing regulatory debate on the European AI Act and the Digital Services Act, positioning intra-party AI deployment as a paradigmatic case for studying how algorithmic systems interact with power structures in democratic institutions at a moment when regulatory frameworks are still being negotiated.