Research publications addressing the Sustainable Development Goals (SDGs) have grown exponentially, reflecting an increasing global focus on sustainability challenges. However, linking these publications to relevant SDGs remains a time-consuming and non-trivial task due to the broad scope and interconnected nature of the goals. This study aims to improve the efficiency and accuracy of mapping publications to SDGs using automated methods. Specifically, we investigate the performance of a domain-adapted similarity measure compared to OpenAI's GPT-3.5 Turbo and GPT-4 models. Using a dataset of over 82,000 research publications from an Australian university, we apply the similarity measure to assign SDG tags and benchmark the results against outputs from the two GPT models. Our findings show that the similarity-based method achieves comparable performance, with successful classification rates of 82.89% (GPT-3.5) and 89.34% (GPT-4), respectively. The proposed approach provides a reliable, transparent, and cost-effective solution for large-scale SDG classification, particularly valuable for institutions handling sensitive data or lacking access to commercial AI tools. This work provides a practical and reliable approach to help institutions track how their research contributes to the United Nations Sustainable Development Goals.