Question generation for educational assessments is an expanding field within artificial intelligence, particularly in educational technology domains such as intelligent tutoring systems and dialogue-based platforms. These tools are crucial for creating assessment questions with clear-cut answers, typically relying on syntactical and semantic cues within declarative sentences to transform them into questions. Recent efforts have focused on generating educational questions in Arabic, but performance has been hindered by issues like sentence parsing inaccuracies, named entity recognition problems, and errors from rule-based question transformation, exacerbated by the complexity of lengthy Arabic sentences. This research introduces an innovative Arabic question-generation system built on a three-stage process: extraction of keywords and key phrases, question generation, and subsequent ranking. The proposed framework leverages an existing pre-trained model without fine-tuning, achieving results comparable to fine-tuned models. Notably, it achieved impressive metrics with a BLEU score of 17.31, ROUGE-L of 51.6, and METEOR of 32.74. These results underscore the framework’s effectiveness and potential in generating high-quality assessment questions in Arabic. Human evaluation further validated the model’s performance, highlighting its significant contribution to the field of automatic question generation for educational assessments.