This paper reports on a mixed-methods pilot study examining the potential of a large language model (LLM) for translating tourism promotional texts (TPTs), in comparison with a conventional neural machine translation (NMT) system, from English into Arabic. Four professional translators participated in a post-editing experiment followed by semi-structured interviews involving cue-based retrospection. The post-editing task aimed to provide empirical evidence of the effort involved in working with TPTs, while the interviews sought to capture participants’ judgments and subjective evaluations of the outputs. Overall, most participants exerted less effort post-editing LLM-generated outputs to a publishable standard compared to NMT outputs. They perceived the LLM outputs to be more creative, with creativity manifested through non-literal translations and aesthetic augmentation, while also noting that the outputs were unpredictable and far from perfect; in contrast, the NMT outputs were generally viewed as more informative yet lacking the promotional appeal needed for TPTs. The paper concludes by outlining the implications of the findings and suggesting several avenues for future research.