The integration of artificial intelligence (AI) into oncology is transforming diagnostic and treatment-planning approaches, making the evaluation of its clinical effectiveness increasingly important. This study aimed to systematise current clinical applications of AI in oncology treatment planning. The research was based on a systematic review of scientific publications conducted in accordance with the PRISMA protocol. Validated deep learning and classical machine learning models, their architectures, and clinical applications were analysed. Across studies using different cancer types, datasets, endpoints, and validation designs, reported tumour- segmentation accuracy for convolutional neural networks ranged from 85% to 99%, reported survival-prediction accuracy for extreme-gradient-boosting models ranged from 80% to 94%, and reported performance for long short-term memory networks ranged from 78% to 89%. These ranges are descriptive and should not be interpreted as direct comparisons between algorithms. Automated radiotherapy planning systems reduced planning time by approximately 60% and achieved Dice coefficients of 0.74–0.82 for lung tumour segmentation. However, external validation revealed important limitations in model generalisability, as 87.5% of studies relied on retrospective data and lacked sufficient international validation. The findings indicate that AI has substantial potential to improve diagnostic precision and treatment-planning efficiency in oncology. Nevertheless, wider clinical implementation requires standardised data collection, prospective multicentre studies, and rigorous validation to ensure the reliability, fairness, and generalisability of AI-based decision support systems.