Ischemic stroke, caused by the occlusion of cerebral blood vessels, requires prompt intervention to restore perfusion and improve patient prognosis. Accurate prediction of functional outcomes is essential for guiding treatment decisions and optimizing patient management. However, such predictions often rely on clinical and imaging data, which may not be available early in the care pathway. This study investigated the potential of free-text clinical notes recorded at admission to predict functional outcomes. Admission documents were encoded using lexical (TF-IDF) and semantic (BERT-based) features, and XGBoost classifiers were trained to predict whether a patient would have a favorable outcome 90 days post-stroke. Using a proprietary dataset of 284 patients, the best text-only performance was achieved with ClinicalBERT applied to original records augmented with synthetic content generated by Llama (AUROC = 0.744, balanced accuracy = 0.662). Compared to structured-data baselines incorporating clinical and non-contrast CT information, the text-based model demonstrated higher recall but lower specificity. These findings highlight admission text as a viable, low-resource alternative for early stroke outcome prediction, supporting future integration with multimodal data for real-time decision-making.