Metastasis is the primary cause of treatment failure in nasopharyngeal carcinoma (NPC); however, the current tumour-node-metastasis staging system has limitations in predicting distant metastasis and guiding induction chemotherapy (IC) application. Here, we established a transcriptomics-based gene signature to assess the risk of distant metastasis and guide IC in locoregionally advanced NPC. Transcriptome sequencing was performed on NPC biopsy samples from 12 pairs of patients with different metastasis risks. Bioinformatics and qPCR were used to identify differentially expressed genes (DEGs), while univariate and multivariate analyses were used to select prognostic indicators for the gene signature. A signature-based nomogram was established in a training cohort (n = 191) and validated in an external cohort (n = 263). Eleven DEGs were identified between metastatic and non-metastatic NPC. Four of these (AK4, CPAMD8, DDAH1 and CRTR1) were used to create a gene signature that effectively categorised patients into low- and high-risk metastasis groups (training: 91.1 versus 70.4%, p < 0.0001, C-index = 0.752; validation: 88.4 versus 73.9%, p = 0.00057, C-index = 0.741). IC with concurrent chemoradiotherapy (CCRT) improved distant metastasis-free survival in low-risk patients (94.4 versus 85.0%, p = 0.043), whereas patients in the high-risk group did not benefit from IC (72.6 versus 74.9%, p = 0.946). Our transcriptomics-based gene signature was able to reliably predict metastasis in locoregionally advanced NPC and could be used to identify candidates that could benefit from IC + CCRT. •Transcriptomics identified 11 DEGs between metastatic and non-metastatic NPC.•A four-DEG signature categorised patients into low and high-risk metastasis groups.•IC improved distant metastasis-free survival in patients in the low-risk group.•The signature can reliably predict metastasis in locoregionally advanced NPC.•The signature could be used to identify candidates that could benefit from IC + CCRT.