AI-enabled Qualitative Analysis: LLM Adoption in Primary Care

This post shows how to run a rigorous qualitative analysis of LLM adoption in primary care, using the PLOS ONE protocol published on August 7, 2026, as a worked example. The primary keyword for this guide is qualitative analysis of LLM adoption. The intended audience is qualitative researchers and clinical research teams who need pragmatic decisions on sampling, interview design, coding, and AI-assisted workflows.

Key Takeaways

According to the PLOS ONE protocol (Pokharel et al., 2026), a focused qualitative design using semi-structured interviews, purposive sampling, and the Technology Adoption Behavior (TAB) framework will reveal how family physicians weigh benefits and risks of LLMs PLOS ONE.

Evidano is an AI-powered qualitative data analysis platform that helps researchers analyze interviews, open-ended surveys, and documents.

The PLOS ONE protocol was published on August 7, 2026 and reports ethics approval from the UBC Behavioral Research Ethics Board (H25-01658).
The PLOS ONE protocol frames physician sampling into three LLM-experience groups: 50+ uses, 5–50 uses, and fewer than 5 uses, with an anticipated recruitment window from August 1, 2025 to January 31, 2026.
The PLOS ONE protocol cites that the number of Canadians without a family doctor rose from 4.5 million in 2019 to about 6.5 million in 2023, motivating focus on primary care (Pokharel et al., 2026).

What Happened and How the Protocol Works

Answer: The PLOS ONE study protocol outlines a descriptive qualitative study of primary care physicians in British Columbia using semi-structured interviews and the TAB framework to map perceptions of LLM adoption.

According to PLOS ONE (Pokharel et al., 2026), the research team will purposively sample practicing family physicians in British Columbia to achieve variability in LLM experience, years of practice, and clinical roles.

According to PLOS ONE (Pokharel et al., 2026), interviews will be recorded, transcribed verbatim, deidentified, and analyzed with both deductive and inductive thematic coding, and the team plans to use NVivo for analysis.

According to PLOS ONE (Pokharel et al., 2026), the TAB framework integrates perceptions (perceived ease of use and usefulness) with task-technology fit to explain adoption dynamics.

Direct quote: Pokharel et al. write, "we aim to use thematic analysis drawing on deductive and inductive approaches" (PLOS ONE, 2026).

Findings Snapshot

Date / Source	Metric	Value / Definition	Implication for qualitative research
August 7, 2026 (PLOS ONE)	Protocol publication	Study protocol and TAB framework described	Use the TAB framework to design interview guides and coding schemas
Aug 1, 2025 to Jan 31, 2026 (PLOS ONE)	Planned recruitment window	6 months anticipated for recruitment and data collection	Plan resources for transcription, deidentification, and reflexive memos in that timeline
2019 and 2023 (PLOS ONE citing national data)	Primary care access	Patients without a family doctor rose from 4.5 million in 2019 to about 6.5 million in 2023	Justifies focusing on family physicians as high-impact users and gatekeepers
Defined in protocol (PLOS ONE)	LLM experience categories	Frequent: 50+ uses; Occasional: 5–50 uses; Rare: <5 uses	Use these thresholds to stratify sampling and cross-segment analysis

Implications for Qualitative Researchers and Clinical Teams

Answer: Use purposive sampling, a theory-driven guide, and mixed deductive-inductive coding to capture how clinicians frame value and risk, as recommended in the PLOS ONE protocol.

According to PLOS ONE (Pokharel et al., 2026), purposive sampling with explicit LLM-experience strata (50+, 5–50, <5) helps reveal adoption gradients rather than a binary user/non-user split.

According to PLOS ONE (Pokharel et al., 2026), combining deductive codes based on TAB with inductive coding lets the study test theoretical expectations while remaining open to novel clinician concerns.

According to PLOS ONE (Pokharel et al., 2026), reflexive memos and a multidisciplinary team are essential to mitigate researcher bias and to interpret professional and ethical tensions around clinical LLM use.

How Evidano Helps: Practical Feature Mapping

Problem: Slow transcription and inconsistent deidentification

Solution: Evidano provides automated transcription with custom dictionaries and PII redaction to produce deidentified transcripts ready for analysis.

For details see Evidano features: Evidano transcription features.

Problem: Managing deductive and inductive coding at scale

Solution: Evidano supports thematic, frequency, and cross-segment analyses and lets teams iterate codebooks while tracking code provenance, which maps directly to the PLOS ONE protocol’s use of deductive and inductive approaches.

Explore how thematic workflows work in Evidano features.

Problem: Synthesizing findings across LLM-experience strata

Solution: Evidano’s cross-segment analysis and visualizations (code hierarchies and co-occurrence networks) speed identification of where frequent, occasional, and rare users diverge in perceptions.

Evidano also offers an AI chat over your documents so teams can query transcripts and code outputs interactively: Evidano AI chat.

Problem: Ethical data handling and collaborator trust

Solution: Evidano documents secure storage and encryption practices and provides team access controls to align with ethics protocols; read about our approach at Evidano data security.

FAQ: qualitative analysis of LLM adoption

How many participants should a qualitative study of LLM adoption recruit?

Answer: Aim for purposive sampling until thematic sufficiency rather than a fixed N, as recommended in the PLOS ONE protocol.

Supporting detail: The PLOS ONE protocol (Pokharel et al., 2026) cites the concept of sufficiency and references Vasileiou et al. (2018) on justifying sample-size adequacy in interview studies.

How do I apply the TAB framework when coding interviews about LLMs?

Answer: Start with deductive codes for perceived ease of use, perceived usefulness, task-technology fit, and social/organizational factors, then add inductive codes that emerge from transcripts.

Supporting detail: The PLOS ONE protocol (Pokharel et al., 2026) provides a TAB-based codebook scaffold and recommends monthly team reviews to refine codes and ensure analytic depth.

Can AI tools assist with transcription and coding without compromising ethics?

Answer: Yes, if you implement verified PII redaction, secure storage, and ethics-approved deidentification workflows as described in the PLOS ONE protocol.

Supporting detail: The PLOS ONE protocol (Pokharel et al., 2026) specifies that audio will be destroyed after transcription and that only deidentified transcripts will be analyzed; platforms used should match those safeguards and institutional ethics approvals.

What are practical interview prompts for exploring clinician risk perceptions of LLMs?

Answer: Use open prompts on specific tasks (documentation, diagnosis, patient communication), probe for illustrative examples, and actively solicit disconfirming evidence, as recommended in the PLOS ONE guide.

Supporting detail: The PLOS ONE protocol (Pokharel et al., 2026) includes an interview guide that asks about intended purpose, target users, settings of use, and prompts for contradictory experiences.

Conclusion & Next Steps

Answer: The PLOS ONE protocol (Pokharel et al., 2026) provides a replicable model for qualitative analysis of LLM adoption that combines purposive sampling, the TAB framework, and mixed deductive-inductive thematic analysis.

Researchers should plan for targeted strata (50+, 5–50, <5 LLM uses), clear ethics-approved deidentification, and iterative team reflexivity, consistent with the PLOS ONE protocol.

To accelerate these steps with secure transcription, collaborative coding, and cross-segment visualizations, try an AI-enabled qualitative workflow.

Get started with a hands-on trial: Try Evidano for free.

Topics

qualitative analysis of LLM adoption
LLM adoption primary care
technology adoption behavior framework
AI-enabled qualitative research

Keep reading

Browse all articles

AI-enabled Qualitative Analysis: LLM Adoption in Primary Care

Key Takeaways

What Happened and How the Protocol Works

Findings Snapshot

Implications for Qualitative Researchers and Clinical Teams

How Evidano Helps: Practical Feature Mapping

Problem: Slow transcription and inconsistent deidentification

Problem: Managing deductive and inductive coding at scale

Problem: Synthesizing findings across LLM-experience strata

Problem: Ethical data handling and collaborator trust

FAQ: qualitative analysis of LLM adoption

How many participants should a qualitative study of LLM adoption recruit?

How do I apply the TAB framework when coding interviews about LLMs?

Can AI tools assist with transcription and coding without compromising ethics?

What are practical interview prompts for exploring clinician risk perceptions of LLMs?

Conclusion & Next Steps

Topics

Keep reading

Company

About

Preview Evidano

Account

Newsletter