Research Spotlight: Using AI Responsibly in Qualitative Interview Analysis

The article by Cheng-Ta Wu, Chiung-Wen Chang, and Han-Yao Chang examines how AI can be integrated into qualitative interview analysis, with Evidano used as the central demonstration tool. Rather than treating AI as a replacement for scholarly judgment, the authors frame it as a research assistant that can help organize, summarize, code, surface quotes, and make patterns easier to inspect.

We spoke with lead author Cheng-Ta Wu, Distinguished Professor in the College of Education at National Chengchi University, about what the demonstration revealed, where the platform saved the most effort, and where the researcher’s judgment has to stay in charge.

Qualitative research often depends on close reading, context, judgment, and patience. These strengths also make it difficult to scale when researchers face large volumes of interviews, documents, survey responses, or multilingual text. The article matters because it does not present AI as a shortcut around interpretive work. It presents AI as a way to make that work more manageable, reviewable, and focused.

For education researchers and institutional leaders, the study also makes the AI question concrete. Instead of asking whether AI will replace researchers, it asks how researchers can design workflows where AI assists with structure and retrieval while humans remain accountable for meaning.

Cheng-Ta Wu described the implication he considers most important: “The most crucial implication is that AI should neither be feared as an existential threat to interpretive scholarship nor embraced uncritically as an automated shortcut. By intentionally designing an interactive, human-AI collaborative workflow, educational researchers can overcome cognitive overload when navigating voluminous narrative data while keeping epistemic control, contextual nuance, and reflexivity firmly in human hands.”

Three contributions stand out in the article because they speak both to qualitative method and to the realities of applied educational research.

AI can make the early stages of qualitative analysis more inspectable. The paper shows Evidano generating summaries, themes, a codebook, frequency analysis, and answers to researcher-supplied questions from eight interview transcript files, then presenting links from the project outline to individual analyses and key quotes.

Researcher control remains central. The workflow described in the article allows researchers to choose analysis modes, upload their own questions or codebook, add or revise viewpoints, edit content, and export results after review.

The case example surfaces the pressures and strategies of middle-level leadership. In the sample analysis, Evidano summarized the multiple roles of elementary school middle leaders, their communication demands, trust building, team collaboration, training needs, and peer support networks.

Asked what surprised him most, Cheng-Ta Wu pointed to how much the framing of the request shaped the quality of the output: “What stood out most clearly is that AI functions at its highest potential when prompted not just for open-ended extraction, but within structured analytic frameworks—such as researcher-guided inquiry and inductive-deductive cross-verification. In our demonstration with middle-level school leadership data, Evidano did not just summarize text; it effectively laid bare the underlying structural tensions between policy implementation and emotional labor, providing inspectable pathways that dramatically expedited the researchers’ interpretive synthesis.”

The article is a methodological and applied demonstration rather than a conventional empirical interview study. The authors introduce Evidano’s core workflow, then develop a case example on the challenges and coping strategies of middle-level leadership in public elementary schools, including six research questions and six semi-structured interview questions.

For the demonstration, the authors created eight interview transcript files with ChatGPT, uploaded them to Evidano Pro, and examined two analysis modes: AI-generated themes, and answers to researcher-supplied questions. The analysis covered summaries, themes, codebook-style outputs, frequency analysis, key quotes by viewpoint or topic, editable results, export options, and chat-based exploration. The article is therefore primarily qualitative, interpretive, and workflow-focused, with frequency analysis used as a supporting quantitative component.

Because the article explicitly examines Evidano, then known as AILYZE, this spotlight describes the platform as part of the authors’ demonstrated workflow, not as the source of the study’s interpretation. In the workflow, Evidano helped generate summaries, themes, a codebook, frequency analysis, and answers to researcher-supplied questions. The article also shows researchers reviewing linked sections, adding viewpoints to key quote searches, editing generated content, downloading outputs, and using chat to explore results. This matters because it places AI inside a broader research process where the researcher still decides what to ask, what to review, what to revise, and what to interpret.

Cheng-Ta Wu was specific about which parts of that process the platform actually carried, and which it did not: “Evidano provided transparent traceability between extracted themes, viewpoint-driven queries, and source transcripts, which greatly streamlined the early, laborious stages of coding and quote mapping. However, researchers must always remember that AI proposes possibilities—it does not validate truth. Scholarly rigor still demands rigorous human auditing: verifying quote contexts, critically probing for latent subtleties, and ensuring that the final theoretical synthesis reflects the lived reality of the participants.”

The central message is responsible assistance. As the authors write in the article, 「未來的質性研究應採納人機協作的模式，將AI視為一個強大的助手，而非替代品。」 (future qualitative research should adopt a human–AI collaborative model, treating AI as a powerful assistant rather than a substitute).

For researchers, the broader implication is that AI-assisted qualitative analysis should be evaluated as a workflow design problem, not only as a technical capability. The article encourages researchers to ask where AI can reduce repetitive work, where human review must be strongest, and how outputs can remain traceable to the underlying data.

For educators and institutional leaders, the study points to a future in which qualitative evidence can be explored more efficiently without losing sight of lived context.

Cheng-Ta Wu’s advice to researchers adopting these tools is procedural as much as intellectual: “Start by viewing AI as a tireless junior research partner rather than a replacement for critical thinking. Maintain full procedural transparency by explicitly documenting prompt designs, iterative analytical modes, and audit trails. Most importantly, never outsource interpretive accountability: the ultimate meaning, contextual depth, and ethical stewardship of the narrative must always belong to the researcher.”

The most valuable contribution of the article is its balance. It recognizes the speed and structure that AI can bring to qualitative data analysis, while insisting that meaning, caution, and accountability remain human responsibilities. For a field built on careful attention to language, context, and experience, that balance is exactly where the conversation should continue.