<![CDATA[Large language models (LLMs) are increasingly used for qualitative data analysis; however, questions remain regarding their reliability compared to human coders. Following PRISMA 2020 guidelines, this systematic review synthesizes empirical evidence on the use of generative artificial intelligence for coding interview and focus group data. Of the 1,085 records retrieved from six academic databases between 2020 and 2026, 30 studies met the inclusion criteria. The findings indicate that LLMs, predominantly GPT-4, achieve moderate to substantial thematic agreement with human coders, with Cohen’s kappa values ranging from 0.40 to 0.91 (median 0.72) and accuracy rates between 77% and 96%. Reliability significantly improves with optimized prompting strategies and multi-run ensemble methods. Although LLMs demonstrate exceptional efficiency, reducing analysis time by 80% to 95%, they still face limitations in capturing cultural nuance, interpretive depth, and context-dependent coding. Therefore, current evidence supports the use of LLMs as an augmentation tool rather than a replacement for human researchers. Hybrid human-AI workflows, combining computational efficiency with human interpretive rigor, represent the most promising approach for robust qualitative analysis. For educational researchers, these findings highlight the potential of LLMs to advance qualitative learning analytics by enabling rapid processing of large-scale student data. Ultimately, this hybrid approach allows for deeper insights into technology-enhanced learning environments without sacrificing pedagogical nuance.]]>
Copyrights © 2026
