Abstract: With the development of pre-training technology, the capability of generative artificial intelligence has been significantly improved, enabling the autonomous generation of texts that closely mimic natural language. However, their performance in quoting Chinese classical poetry has not been thoroughly explored, which is crucial for enhancing the literary expressiveness of the generated content. To address this, we construct a large-scale Chinese Poetry Quotation Dataset (CPQD), and propose a set of automatic and manual evaluation metrics from multiple perspectives including poeticness, coherence, and fluency. Subsequently, we analyze three open-sourced Chinese LLMs, three commercial Chinese LLMs, one open-sourced and one commercial English LLMs, in both zero-shot and few-shot setting. The results demonstrate that only one of the commercial Chinese language models outperforms the fine-tuned LSTM model. We fine-tune the open-source models using the proposed poetry quotation dataset, and the BLEU increases nearly tenfold, significantly surpassing that of the commercial LLMs.