Accurate species-level identification is essential in harmful algal research, as closely related harmful algal species may differ in toxicity, bloom-forming potential and ecological impact. Illumina short-read sequencing is widely used for harmful algae detection, but partial markers such as SSU/18S, ITS or LSU/28S may require multi-marker confirmation and can produce inconsistent or incomplete taxonomic signals. PacBio HiFi long-read sequencing provides a useful alternative by generating highly accurate reads that can span multiple informative rDNA regions within a single sequence. In this study, Pseudo-nitzschia was used as a representative harmful algal genus to evaluate the ability of PacBio HiFi metabarcoding to improve species-level identification. A total of 600 seawater samples collected across the coastal waters of China between 2019 and 2024 were analysed using PacBio HiFi long-read metabarcoding of the ITS1–5.8S–ITS2 and LSU(D1–D2) rDNA regions. Taxonomic assignments were evaluated using a BLAST-based framework based on covered marker regions, percentage identity, query coverage and alignment length. After excluding potential reverse-complement duplicate matches, most unique top-hit Pseudo-nitzschia OTUs covered the full ITS1–5.8S–ITS2–LSU(D1–D2) region, while smaller proportions were represented by ITS-only, LSU-only or other partial-region matches. Full ITS–LSU hits provided the strongest species-level evidence because this region combines the highly variable ITS regions with LSU phylogenetic information. However, LSU-only and partial-region hits were also informative, supporting the detection of additional Pseudo-nitzschia lineages, particularly when full-region reference sequences were limited. Lower-coverage assignments were further evaluated and distinguished using ITS1–ITS2 secondary-structure information, including compensatory base changes (CBC), hemi-CBC and sequence–structure phylogenetic analyses, to refine species or species-complex boundaries. Using this confidence-based framework, multiple Pseudo-nitzschia species were strongly supported, while lower-coverage matches were retained as species-complex, candidate lineage or nearest-match assignments. Overall, this study demonstrates that PacBio HiFi long-read metabarcoding improves harmful algal species identification by integrating information from multiple rDNA regions in a single sequencing approach. This framework provides stronger species-level resolution from full ITS–LSU reads while retaining useful taxonomic signals from LSU-only and partial ITS-region matches, offering a more reliable approach for resolving closely related harmful algal taxa than conventional short-marker approaches alone.