Vahtian
← All Learn guides

The ideas behind the methods

A short history of qualitative research

Fifteen stops from early ethnography to AI-assisted analysis, and the three questions that never went away.

Keep these three questions beside the timeline

This is a teaching timeline, not a claim that the field moved in one neat sequence. Traditions overlap, disagree, and continue.

01What counts as knowledge?
02Who has the authority to interpret experience?
03How is an interpretation made credible?

01 / Roots

Observation, meaning, method

Qualitative research began as several traditions, not one unified method. Each offered a different answer to what researchers can know from lives, language, and social worlds.

Observation and the colonial archive

Early ethnography made prolonged observation and field notes central to inquiry. It also grew inside colonial projects in which Western researchers often claimed the authority to represent people who had little control over the account. [1]

What AI changes: a model can become another distant interpreter, smoothing culturally specific language into categories shaped elsewhere.

The Chicago School

Life histories, case studies, and urban ethnography shifted attention toward meaning, process, and the texture of everyday social life. The resulting knowledge was rich and situated, but often challenged as too local to generalise.

What AI changes: processing thousands of narratives may look like scale. It does not make the sample representative or the interpretation valid.

Experience is interpreted, not simply collected

Phenomenology, hermeneutics, and symbolic interactionism placed meaning and interpretation at the centre. Language does not merely report a world; it helps make that world intelligible.

What AI changes: a model can produce a coherent interpretation without having shared the participant's world. Fluency is not situated understanding.

Grounded theory

Glaser and Strauss argued that theory could be developed from data through coding, comparison, and theoretical sampling, rather than only tested after being supplied in advance. [2]

What AI changes: asking for a theory before close engagement with the material reverses the method and lets the first plausible structure anchor everything that follows.

Fluent interpretation can look like understanding. The history of the field gives us reasons to keep those apart.

02 / Authority

Who gets to interpret?

The field widened. Neutrality, voice, power, and the form of the research account became methodological questions rather than background concerns.

One standard gives way to many paradigms

Constructivist, critical, feminist, and participatory traditions rejected the idea that one neutral procedure could settle every qualitative question. Reflexivity, standpoint, and power moved closer to the centre.

What AI changes: a general model tends to blend terms that often appear together, even when their assumptions are incompatible. A polished methods paragraph can hide the mismatch.

Computer-assisted analysis

Qualitative software made transcripts searchable and connected codes, memos, and source passages. Its role was to store and retrieve analytic work while the researcher remained responsible for interpretation.

What AI changes: a language model does not only organise. It generates analytic content. That is a change in role, not simply a faster version of the same tool.

Feminist, critical, participatory, and decolonising turns

Researchers asked who benefits, who controls the categories, whose language is privileged, and who owns the knowledge. In participatory work, people previously treated as subjects could become co-researchers.

What AI changes: models can quietly re-centralise dominant-language categories. Under community control, local tools may instead support translation and participant-led analysis.

Narrative, discourse, and conversation

Form became part of the evidence. Sequence, audience, pause, repetition, code-switching, and silence could do social work that a content summary would miss.

What AI changes: a cleaned transcript is already one transformation. A model that treats it mainly as semantic content adds another and can erase what was never captured as plain text.

Thematic analysis becomes easier to name

Braun and Clarke set out a flexible, explicit account of thematic analysis. Later work distinguished reflexive thematic analysis from coding-reliability approaches that answer different methodological questions. [3]

What AI changes: models readily produce topic summaries. A theme is not a frequent topic; it is an interpretive pattern that answers the research question.

03 / Records

From craft to documented workflow

Digital settings and formal reporting guidance made the analytic record more visible. They also created new ways to mistake documentation for rigour itself.

Digital ethnography and large text corpora

Social life moved online, and qualitative inquiry followed through digital ethnography, social-media research, and the study of digital traces. Availability did not remove the need for consent, context, or care.

What AI changes: scale makes it easy to detach a statement from platform culture, thread history, timing, and identity performance, then treat the fragment as self-contained evidence.

Reporting standards make the record explicit

COREQ and SRQR asked authors to report the team, study design, analysis, interpretation, and links between claims and material more transparently. They are reporting guides, not substitutes for methodological judgment. [4] [5]

What AI changes: a model can reconstruct a plausible methods section after the fact. The better use of technology is to preserve prompts, versions, edits, rejected suggestions, and source links while the work happens.

Remote research becomes ordinary

The pandemic accelerated video interviews, online groups, remote recruitment, and automated transcription. Researchers also had to confront digital exclusion, rapport at a distance, and uncertainty about where recordings and transcripts travelled. [6]

What AI changes: the infrastructure for remote research made it easy to add cloud models to a workflow before data location, consent, and third-party processing had been settled.

Use software to preserve the record, not to manufacture the appearance of one.

04 / Models

When software begins to suggest meaning

Generative models entered planning, coding, interpretation, and writing. The central question became less about access to the model and more about the role it is allowed to play.

Generative models enter the workflow

Researchers began testing public language models for interview guides, summaries, coding, theme development, and drafting. AI-supported qualitative analysis was not new, but conversational tools made it newly accessible. A 2025 review identified 130 articles on AI-supported qualitative data analysis through May 2024; not all concerned generative LLMs. [7]

What AI changes: access is no longer the main barrier. Method fit, privacy, provenance, and the weight given to a generated suggestion become design decisions.

Comparisons replace easy claims

The evidence does not support one simple verdict. In interviews with Rohingya refugees and Bangladeshi hosts, LLM annotation errors varied with participant characteristics and could bias later inference. A review of 11 qualitative nursing studies found promise in text analysis and theme generation but weakness in applying theory, developing codebooks, and generating interview questions. [8] [9]

What AI changes: agreement with a human output is not enough. Researchers need to test whose material is misread, which tasks fail, and whether the evaluation reflects the method's actual aims.

Workflow design matters more than model choice

The useful unit of evaluation is increasingly the whole human-AI workflow: what the model sees, what it produces, who reviews it, what gets rejected, and how the final claim remains connected to source material. The literature is still young and does not justify a universal division of labour. [7] [8] [9]

What AI changes: the researcher must define the model's role before using it and keep enough of the path visible for another person to challenge the result.

What survives every era

The tools change. The responsibility does not.

An LLM can sort, compare, summarise, and suggest. It has no lived relationship with participants, no stable position it can account for, and no responsibility for what a research claim does in the world.

You make the research decision. The record should make it possible to see how.

Sources and further reading

  1. Denzin NK. Critical Qualitative Inquiry. Qualitative Inquiry. 2017.
  2. Glaser BG, Strauss AL. The Discovery of Grounded Theory: Strategies for Qualitative Research. Aldine; 1967.
  3. Braun V, Clarke V. Using thematic analysis in psychology. Qualitative Research in Psychology. 2006.
  4. Tong A, Sainsbury P, Craig J. Consolidated criteria for reporting qualitative research (COREQ). 2007.
  5. O'Brien BC, et al. Standards for Reporting Qualitative Research (SRQR). 2014.
  6. Keen S, Lomeli-Rodriguez M, Joffe H. From Challenge to Opportunity: Virtual Qualitative Research During COVID-19 and Beyond. 2022.
  7. Cook DA, et al. Artificial Intelligence to Support Qualitative Data Analysis: Promises, Approaches, Pitfalls. 2025.
  8. Ashwin J, Chhabra A, Rao V. Using Large Language Models for Qualitative Analysis can Introduce Serious Bias. Sociological Methods & Research. 2026.
  9. Zhou T, et al. The application of large language models in qualitative nursing research: A scoping review. Nursing Outlook. 2025.