Background: Autoimmune diseases are highly heterogeneous, with unpredictable treatment outcomes that often result in prolonged morbidity. Conventional bulk transcriptomic approaches obscure cellular diversity and fail to capture the spatial microenvironment that drives drug responses. Aim: To identify spatial transcriptomic biomarkers that predict patient-specific therapeutic responses in autoimmune diseases. Methods: We applied single-cell spatial transcriptomics (scST) to patient-derived synovial tissue from rheumatoid arthritis (n = 12) and systemic lupus erythematosus (n = 8) to construct a high-resolution atlas of immune and stromal interactions during therapy. Results: By integrating scST with machine learning–based predictive modeling, we identified cell-state signatures that stratify patients into responders and non-responders before treatment initiation. Spatial colocalization of interferon gamma–responsive macrophages and C-X-C motif chemokine ligand 13-positive T follicular helper cells predicted resistance to Janus kinase inhibitors (AUC = 0.89). In contrast, enrichment of programmed cell death protein-1 in highly exhausted T cells adjacent to fibroblastic reticular cells improved response to tumor necrosis factor-alpha blockade (AUC = 0.92). Notably, extracellular matrix (ECM)-associated remodeling genes, including COL6A3 and FN1, emerged as critical determinants of microenvironmental drug sensitivity, highlighting the ECM as a therapeutic co-driver in autoimmunity. Validation in an independent cohort (n = 20) confirmed the predictive robustness of these spatial biomarkers. Conclusion: Our findings demonstrate that scST can resolve patient-specific immune niches and provide actionable biomarkers for precision immunotherapy. Relevance for patients: Beyond its immediate implications for rheumatology, this framework establishes spatial single-cell mapping as a predictive diagnostic platform for diverse autoimmune diseases, transforming treatment from trial-and-error to individualized therapeutic guidance.