Intra-utterance code-switching (CS) is common in spoken language, posing significant challenges for automatic speech recognition (ASR) systems that need to handle mixed languages effectively. A primary obstacle in developing a CS-ASR system is the scarcity of suitable data. The complexity of CS grammatical structures further complicates the task, especially with Arabic, which has numerous dialects differing significantly in vocabulary, pronunciation, and syntax. To address CS, ASR systems are typically trained with available transcribed CS speech. This paper leverages advancements in large language models (LLMs) to enhance CS-ASR systems by generating Arabic-English code-switched textual data. Additionally, we introduce the Saudilang Code-switch Corpus (SCC), an evaluation dataset of Saudi CS with English. Our results show a relative reduction in perplexity by over 8% and a 5.5% average relative decrease in WER on two ecologically valid CS evaluation datasets. We plan to release the generated CS data and the new Arabic CS evaluation set to the research community.