SQL compiler error messages are the primary way users receive feedback when they encounter syntax errors or other issues in their SQL queries. Effective error messages can enhance the user experience by providing clear, informative, and actionable feedback. Despite the age of SQL compilers, it still remains largely unclear what contributes to an effective SQL error message. With 2,052 answers yielded by 165 participants for qualitative analysis, this study is an attempt to understand what novices perceive as effective elements in SQL error messages. The results uniformly indicate that communicating the precise error position, articulating what is wrong in the query with clear natural language, and showing hints on how to fix the error are perceived as the most effective elements for error recovery. These insights have potential to be utilized in providing more effective error messages in SQL compilers and SQL learning environments, and for guiding generative AI for enhanced error messages in order to minimize frustration caused by cryptic error messages, improving learning and adoption, and reducing debugging time.