Drawing on theories of deliberative democracy and Systemic-Functional Linguistics, this study explores how institutional language constitutes a barrier to democracy. Complex texts exclude individuals, including those experiencing intellectual and cognitive disabilities (ICD), older persons, and migrants, from participation in deliberative democratic processes. The inclusion of traditionally underrepresented individuals in deliberation poses different challenges to deliberative democratic theory. There is an alleged tension between quality deliberation and autonomy concerns with deliberative innovations for inclusion. We aim at clarifying this tension and argue that the joint efforts of AI and deliberative theory offer a promising avenue for inclusive deliberation. The use of AI to enhance democracy provides tools to overcome barriers for deliberation, making participatory and deliberative processes more inclusive. We argue that AI could contribute to deliberative innovations increasing opportunities for inclusion by providing accessible information. However, these innovations are not value-neutral, the use of Large Language Models (LLMs) for deliberation raises ethical concerns. We focus on algorithmic biases, disinformation, and manipulation threats and claim that our target groups are especially vulnerable to ethical concerns due to their position against a background of structural injustice. Then, based on quality-controlled human annotated datasets, we present a typology of simplification strategies and develop a classifier to detect linguistic complexity and an LLM-based text simplification system which enable a more inclusive participation. We conclude that participatory AI systems must be transparent, interpretable, and co-developed with communities to uphold democratic values and promote social justice.