Last updated: 17 September 2026
Quick Answer: Choose the qualitative research design whose output your decision needs. Phenomenology yields the essence of a lived experience from 10 to 15 interviews, grounded theory a theory, ethnography a cultural description built over 4 weeks or more, case study a bounded account, and narrative inquiry a story arc. Pick the output first.
A 2017 methods study ran 25 in-depth interviews. Every theme had surfaced by nine; understanding what they meant took 16 to 24. Two defensible stopping points on one dataset, and the right one depended on what the study had committed to produce.
That commitment is the design: a sampling logic, a data type, an analysis method, and a definition of what counts as a finding. Each link constrains the next, and the chain is fixed before recruitment.
Most published examples of qualitative research, and most lists promising seven types, mix two levels. Interviews, focus groups and observation are data collection methods that sit inside a design. A separate guide covers what qualitative research is and the methods it uses; designs are the layer above, and they decide which of those methods are even available.
Seven Worked Examples of Qualitative Research Designs
Seven designs cover almost all qualitative work, five of them the canonical majors doctoral programs teach, as Grand Canyon University's February 2026 guide also lists. Each is defined by what it produces.
-
Phenomenology asks what an experience is like for those living it. Worked example: a telehealth provider interviews twelve patients managing a chronic condition, describing the experience, not who felt what.
-
Ethnography asks what a group's shared rules are, including unstated ones. Worked example: a food brand embeds a researcher in fifteen households for four weeks, delivering the logic behind the cart.
-
Grounded theory asks how a process works when no framework fits. Worked example: a fintech runs eight interviews on merchant drop-off and lets analysis pick each next recruit, ending in a staged theory of abandoned onboarding.
-
Case study asks why one bounded situation turned out as it did. Worked example: a hospital whose scheduling rollout stalled pulls together interviews, minutes, ticket logs and usage data to explain the mechanism.
-
Narrative inquiry asks how something unfolded over time for one person. Worked example: a bank has nine customers narrate a first mortgage end to end, with analysis preserving sequence rather than coding.
-
Qualitative content analysis asks what patterns sit in text that already exists. Worked example: two years of tickets and reviews are coded against a frame drawn from the first few hundred, reported as a distribution over time.
-
Action research asks whether a change works, with the people affected evaluating it. Worked example: a retailer rewrites its returns script, store teams test and critique it, and the revision ships with what it changed.
What Does Choosing a Qualitative Research Design Commit You To?
Four things, decided together before recruitment starts. Change one and the other three change with it.
-
A sampling logic. Qualitative sampling is purposeful, not probabilistic: cases are chosen because they are information rich, not to generalize. Criterion sampling suits phenomenology, theoretical sampling grounded theory, and a study must say which it used.
-
A data type. Ethnography needs observational records, narrative inquiry accounts with a shape, and a focus group design a room where the dynamics are the point. A cheaper input changes the design silently.
-
An analysis method. Ethnographic, phenomenological, grounded theory and content analysis studies yield structurally different findings, as that European Journal of General Practice series sets out: a culture described, an essence, a theory, coded categories. None converts into another.
-
A definition of what counts as a finding. Decide in advance whether the deliverable is a theme, a theory, a description, an arc or a distribution. That tells you when fieldwork stops, and it is the commitment most often left implicit.
What Goes Wrong When Two Designs Get Mixed in One Study?
You get a study that answers neither question. The common failure is sampling theoretically, as grounded theory requires, then reducing the accounts to a shared essence, as phenomenology does: theoretical sampling seeks difference, phenomenological analysis needs shared experience, and one cancels the other. COREQ's 32-item checklist asks for methodological orientation, sampling method and analysis approach separately, so a study that cannot state all three without contradicting itself has a design problem.
Which Design Fits the Research Question You Actually Have?
The grammar of the question already names the design. "What is it like to" points at phenomenology, "how does this group" at ethnography, "how does this process work" at grounded theory.
| Design | Sampling logic | Typical sample |
|---|---|---|
| Action research | The group doing the work | 1 to 3 teams, several cycles |
| Case study | The bounded case itself | 1 case, many sources |
| Content analysis | A corpus with stated rules | Hundreds of documents |
| Ethnography | A setting, entered over time | 10 to 20 households, 4 weeks or more |
| Grounded theory | Theoretical, round by round | Until concepts stop appearing |
| Narrative inquiry | Small, purposive, richly told | 5 to 12 accounts |
| Phenomenology | Criterion, all have lived it | 10 to 15 interviews |
Rows are alphabetical. If more than one fits, the question is too broad to design against; if it asks how many, a quantitative design fits.
Can Your Design Actually Reach the People It Assumes?
Every design assumes access to a population, and that assumption fails quietly: phenomenology needs people who have lived the experience and can articulate it, ethnography needs presence in the setting.
91% of US adults own a smartphone, but 16% have no home broadband, so a browser-link study thins the sample a design depends on, and sample validity is settled at recruitment. Alchemic runs interviews inside WhatsApp, with no link and no app, as text or voice notes, places outbound AI phone calls, and offers managed fieldwork or bring your own across 14 markets in 57+ languages including Spanish, Arabic and Mandarin.
Ethnography is where the channel decision bites hardest, and it is also the one honest limit. A full ethnography needs a researcher physically in the setting, because the unstated rules the design exists to recover are the ones nobody narrates, and no remote channel substitutes for that. Alchemic runs ethnographic deep-dives across WhatsApp, phone and camera-on video, which reaches households a fieldwork team cannot enter and records what people choose to show. Decide which of those two outputs your question needs before you choose the fieldwork.
Where Qualitative Research Designs Genuinely Fall Short
No qualitative research design can tell you how many. It produces no projectable number, and treating a theme count as a market size is the most common misuse of this work. Two further limits matter.
- Saturation is a judgment, not a threshold. That 2017 study found code saturation at nine interviews and meaning saturation at 16 to 24 on one narrow dataset, so report it with its reasoning.
- Interpretation is researcher-dependent. Two analysts can read the same transcripts into different themes, so a design must say who analyzed the data and how disagreements were resolved.
Where an AI moderator runs fieldwork, the tested limitation is narrow. Nielsen Norman Group's January 2026 evaluation of ten participants across two platforms found those tools followed the script rather than the insight, a finding about the two systems tested. That is a question about in-session latitude rather than guide quality, so ask any vendor how far its moderator may depart from the guide mid-interview. Alchemic tailors the discussion guide from the brief before fielding, which is a different lever on the same risk.