LLMs in Qualitative Research: Opportunities, Limitations, and Practical Considerations

LLMs require deliberate technical configuration for research integrity: Researchers using LLMs in qualitative work must actively manage specific technical settings—including context window size, temperature settings, and prompt design—rather than relying on default configurations, as these choices directly affect research outcomes.

Transparency about model limitations is essential: The paper emphasizes that LLMs are fundamentally more opaque than earlier AI tools like topic models, making it critical for researchers to consult model documentation (system cards) and openly acknowledge these "black box" limitations when reporting findings to maintain research credibility.

Researcher judgment cannot be automated away: Qualitative research's core commitments to reflexivity (examining one's own biases), positionality (acknowledging one's perspective), and interpretive judgment mean that LLMs should augment rather than replace human researchers—these distinctly human analytical skills remain irreplaceable.

Prompt design shapes research outcomes: How researchers write instructions to LLMs (both system prompts and user prompts) significantly influences results, meaning researchers must treat prompt engineering as a methodological choice that should be documented and justified, similar to other research decisions.

Responsible integration requires multidisciplinary expertise: The paper argues that using LLMs responsibly in qualitative research demands researchers understand both qualitative methodology and AI technical considerations—a combination that may require collaboration across disciplines.