From Stories to Themes: AI Qualitative Analysis

According to the 2026 PLOS Neglected Tropical Diseases study, participatory storytelling in Karachi was associated with movement from shame and concealment toward greater self-acceptance, confidence, and openness among people affected by leprosy. The study also found that personal empowerment was stronger than wider community change, showing why AI qualitative analysis must distinguish individual, interpersonal, community, institutional, and structural themes.

• The study recruited 17 participants in Pakistan, with 16 taking part in photovoice and 14 completing in-depth interviews during June 2025.
• The researchers reported that confidence was the most common change and that all participants described increased confidence after the intervention.
• The study found that gender norms, visible disability, community misconceptions, travel costs, and economic constraints continued to limit broader social change in 2025.
• The researchers concluded that sustained community contact, supported peer educators, inclusive outreach, and socioeconomic opportunities were needed beyond storytelling sessions.
• AI qualitative analysis can help researchers code large collections of interviews and compare themes, but sensitive health research still requires human review, reflexivity, consent safeguards, and contextual interpretation.

The study found that participatory storytelling was most strongly associated with individual-level changes in stigma and empowerment. According to Ibrahim and colleagues, participants described a shift from shame, fear, and concealment toward self-acceptance, confidence, openness, and pride.

The research team conducted a qualitative study after a 2024 storytelling pilot at the Marie Adelaide Leprosy Centre in Karachi. According to the study methods, the intervention combined peer support groups, creative activities, health education, advocacy, contact-based activities, public engagement, and a filmed component.

The researchers used 60 to 90 minute in-depth interviews conducted in June 2025, live interpretation from Urdu into English, photovoice, field notes, and thematic analysis supported by NVivo 14. The study combined deductive codes from stigma, empowerment, and the Social Ecological Model with inductive codes that emerged from participant accounts.

One participant described the emotional effect of disclosure by saying, "I have no fear, no shame, not feeling any guilt. I’m feeling like... I feel lighter and so relaxed." The study attributes this quotation to participant P06, a male government worker in his 40s.

Another participant described improved communication and listening skills by saying, "My communication skills are better. I can relate to different stories, and I can convince people better." The study attributes this quotation to participant P07, a female health worker in her 50s.

The study authors cautioned that the intervention did not eliminate enacted stigma. According to the researchers, community myths, gender expectations, visible impairments, and structural barriers continued to shape participants' experiences.

Qualitative health researchers should use AI to make patterns visible while retaining human responsibility for meaning, ethics, and interpretation. The Pakistan study demonstrates that a single theme such as empowerment can operate differently at individual, interpersonal, community, and institutional levels.

• Separate levels of change: The study used the Social Ecological Model across individual, interpersonal, community, institutional, and structural levels, so an AI workflow should preserve those distinctions instead of producing one undifferentiated theme list.
• Track mechanisms, not only outcomes: The study connected storytelling with psychological safety, peer solidarity, communication skills, public visibility, and institutional endorsement. Researchers can use AI to map how these mechanisms appear across interviews.
• Compare participant segments carefully: The study identified different experiences related to gender, visible disability, employment, age, and social circumstances. AI-assisted cross-segment analysis can reveal contrasts, but small purposive samples require cautious interpretation.
• Preserve contradictions: The study reported increased confidence alongside persistent community stigma. A useful AI analysis should retain both findings rather than forcing the data into a simple success narrative.
• Protect sensitive voices: The study described informed consent, translation checks, anonymization, secure storage, and support for distress. AI workflows should apply equivalent safeguards to transcripts, photographs, translations, and quotations.

The study's limitations are especially relevant to AI-enabled research. According to the authors, the sample was small, women represented 4 of 17 participants, there was no baseline or comparison group, interviews took place within an institutional setting, and translation between Urdu and English could introduce semantic discrepancies. AI-generated themes should therefore be treated as analytic support, not causal proof.

The 2026 Pakistan storytelling study shows that qualitative research can capture meaningful shifts in identity and confidence while also documenting persistent structural barriers. AI qualitative analysis can help researchers synthesize interviews, compare segments, retrieve evidence, and connect themes across multilingual or mixed-format datasets. The strongest workflow keeps participant voice, methodological limitations, and human interpretation at the center. Researchers working with interviews, open-ended surveys, or health documents can try Evidano for free to begin a secure AI-assisted qualitative analysis workflow.