Abstract: Data-driven synthesis planning is a crucial step in the discovery of novel materials with desirable properties. Zeolites are crystalline nanoporous materials with applications in catalysis, adsorption, and ion exchange. The synthesis of zeolitic materials remains a significant challenge due to its high-dimensional synthesis space and intricate structure-synthesis relationships. Considering the $\textit{one-to-many}$ relationship between structure and synthesis, we propose a generative modeling approach using a chemically-guided diffusion model for materials synthesis planning. Given a target zeolite structure and organic structure-directing agent (OSDA) as inputs, the diffusion model generates probable synthesis routes and achieves state-of-the-art performance compared to regression and deep generative models. The model learns chemically meaningful relationships, generating realistic synthesis routes that closely follow the distribution of literature-reported synthesis routes. As such, this approach could enable the discovery of zeolitic materials beyond domain-specific heuristics and trial-and-error experimentation.