Qualitative Research

Qualitative research is the systematic study of non-numerical evidence, such as interviews, observation, and open-ended responses, to understand why people think and act the way they do. It explains reasons and context, but it cannot tell you how many people share a view.

That second sentence is where marketing teams most often get into trouble. A handful of customer interviews can reshape a positioning statement, a landing page, or an entire campaign brief, and often they should. The same interviews cannot tell you what share of the market feels that way, or whether the new page will convert better than the old one. This entry focuses on that boundary: what qualitative research can legitimately prove for a marketer, what it cannot, and how to move a finding from one side of the line to the other. For the wider discipline of studying markets, competitors, and demand, see market research.

What Counts as Qualitative Evidence

Interviews, observation, and other direct sources

The defining feature is not the absence of numbers, and it is not the collection tool. It is the kind of evidence and analysis. Qualitative work interprets non-numerical evidence, such as what people say, write, or do; the CDC Field Epidemiology Manual notes that data in qualitative studies typically take the form of text or visual images (CDC, Collecting and Analyzing Qualitative Data). Open-ended survey responses, interviews, field notes, and documents can all be qualitative sources, even when a survey platform collects them. Within UX research, the Nielsen Norman Group frames the distinction as direct methods that observe or hear behavior and attitudes first-hand versus indirect methods that gather them through an instrument (NN/G, When to Use Which User-Experience Research Methods); that direct-versus-indirect frame is their UX method landscape, not a definition of all qualitative methodology. Because the researcher is often present in direct work, they can ask a follow-up question, probe an unexpected answer, or adjust the study protocol while it runs. Analysis is usually interpretive rather than mathematical.

UK government evaluation guidance lists the common qualitative methods as contextual inquiry, ethnographic study, focus groups, interviews, think-aloud studies, and usability testing (GOV.UK, Choose evaluation methods). The guidance was written for digital health products, but the list carries over to marketing work almost unchanged: a buyer interview, a session watching someone compare pricing pages, or a diary kept during a long purchase decision.

Why a pile of comments is not yet research

Reading reviews, social replies, and support tickets is useful. It is not, by itself, qualitative research. NN/G classifies customer feedback as information supplied by a self-selected sample of users through a link, form, or email, which means the people who chose to write in decide what you hear and who stays silent. Qualitative research starts when you decide whose experience you need, collect it deliberately, and analyze it against a defined question. Gathering voice of customer material is a valuable input to that process. It is not a substitute for it.

What Qualitative Research Can and Cannot Prove

What it can support

Qualitative research is strong evidence for reasons, context, and the range of views that exist among the people studied. The GOV.UK guidance gives a clean illustration: a quantitative approach can describe patterns of disengagement with an app and which demographic factors predict them, while a qualitative approach can describe why users stopped. NN/G draws the same line, noting that qualitative methods are much better suited to questions about why or how to fix a problem, while quantitative methods answer how many and how much.

For a marketer, a well-supported claim in this category sounds like this: buyers in these interviews delayed the purchase because they could not tell the two plans apart. The claim is specific. Its scope is the people you spoke with, and it says so.

What it cannot support

Three kinds of claim sit outside what qualitative research can carry. The first is prevalence. GOV.UK states that qualitative studies can describe the range of views among participants but are weak at saying how many participants hold each view. The second is generalization, since the same guidance notes that qualitative results are hard to generalize to other participants. The third is causation. Knowing why people hesitate does not prove that removing the reason will change what they do. For that, GOV.UK points to comparative studies, such as A/B testing or randomized controlled trials, which compare an outcome against what would have happened without the change.

None of this is a weakness to apologize for. It is a scope, not a flaw. Problems begin only when a finding is quietly promoted beyond it, usually somewhere between the research readout and the planning meeting.

When what customers say and do disagree

NN/G frames a second dimension of research as the contrast between what people say and what people do, and observes that the two are very often quite different. Attitudinal research is limited by what people are aware of and willing to report. Most interviews sit on the attitudinal side. Field studies and usability sessions mix both kinds of evidence, which is why NN/G generally recommends leaning toward observed behavior.

The GOV.UK guidance on mixed methods studies summarizes a smoking cessation app evaluation that shows the gap clearly. Location data from the app was largely accurate, but participants often failed to log when they smoked, and the follow-up interviews explained why: they were driving, the phone was not at hand, or they simply forgot. Neither data source told the full story alone. The interviews did not replace the usage data. They explained it.

One study, four separate claims

A practical discipline is to split every qualitative finding into the separate claims it contains, because each one needs a different kind of evidence before it can be reported with confidence:

• What participants said. Supported by the transcripts, within the sample you recruited.
• What participants did. Supported only if you observed it, in a session, a field visit, or behavioral data.
• What the team infers. An interpretation, to be treated as a hypothesis and checked against the cases that do not fit.
• What a change will achieve. A prediction that needs a comparative test before anyone reports it as a result.

Most overreach happens on the jump from the third claim to the fourth. A finding that customers find the pricing confusing becomes a promise that clearer pricing will lift conversions, and nobody notices that the evidence changed type along the way. Labeling each claim at the moment you write it is the simplest guard against that drift.

Matching the Method to the Question

Interviews, field studies, diary studies, and focus groups

NN/G's method catalog gives a practical starting point. Interviews are one-on-one conversations exploring what a participant thinks about a topic in depth. Field studies observe people in their own environment, which raises realism at the cost of control. Diary studies have participants record relevant moments over time, which suits long or irregular purchase journeys. Focus groups bring three to twelve people into a moderated discussion, and NN/G notes they are less useful for usability questions but can give a top-of-mind view of a brand or product concept in a group setting.

Pick the method by the claim. If you need to know how a buyer actually compares options, watch them do it. If you need the words they use to describe the problem, interview them. If the decision unfolds over several weeks, ask them to keep a diary.

Where quantitative research takes over

The two approaches divide the work roughly as follows, based on the NN/G and GOV.UK guidance cited above. Treat it as a simplification, because many methods can produce both kinds of data.

• Qualitative research answers why and how to fix. Evidence is mostly non-numerical, such as interviews, open-ended answers, field notes, or documents, samples are small and chosen on purpose, analysis is interpretive, and the strongest claims concern reasons and the range of views.
• Quantitative research answers how many and how much. Evidence is numerical, samples are larger, analysis is mathematical, and the strongest claims concern prevalence and, in comparative designs, effect.

Combining them is often the right call. GOV.UK notes that a mixed design can balance the limitations of each method and give more confidence in the findings, while also warning that it costs more time, money, and expertise than either method alone.

Sample Size Without a Magic Number

Why fewer participants can be enough

GOV.UK observes that qualitative evaluations can need far fewer participants than other methods, because each participant yields much more information and more time is spent with each one. That is the logic of qualitative sampling. You are not estimating a population figure, so you do not need the sample a population estimate requires.

There is no universal count that makes a study sufficient. A sample is adequate when it contains the people whose experience bears on the decision and when the conversations go deep enough to explain what you are seeing. In a churn study, a small group of recent switchers may tell you more than a larger group of loyal customers who never considered leaving.

Who you recruit decides what you can claim

Selection shapes findings more than size does. GOV.UK points out that qualitative work can give a voice to users who might otherwise be overlooked in an evaluation, which only happens if you recruit them deliberately. Record who was in the sample and who was not, and carry that boundary into every slide. Findings drawn from one customer segment describe that segment, however persuasive the quotes sound.

Collecting and Analyzing Evidence That Holds Up

Asking about events, not opinions

Because attitudinal answers are limited by what people notice and are willing to report, the most useful interview questions ask about specific past events. A request to walk through the last time someone compared providers produces evidence. A question about whether they would use a new feature produces a guess. General rules for neutral question wording are covered in the market research entry, so the point here is narrower: anchor every question in something the participant actually did.

From transcripts to themes

Analysis turns raw material into a small set of interpretations. In practice that means reading everything closely, labeling passages, grouping labels into candidate themes, and then testing each theme against the cases that do not fit. The step teams skip most is the last. A theme that survives its contradictions is worth presenting to stakeholders, while a theme built only from the quotes that agreed with the original brief is closer to confirmation than to research.

Where AI fits, and where it does not

AI now touches every stage of the workflow, and the published evidence is mixed. A Nielsen Norman Group study of AI-moderated interviews found the tools useful for structured work such as product feedback, recruitment screening, and interviews in languages the team does not speak, but not yet suited to semistructured or in-depth discovery interviews (NN/G, AI-Moderated Interviews, January 2026). Only 3 of the 10 participants agreed that the conversation felt natural (source: NN/G; metric: agreement in a post-interview questionnaire; sample: research leaders and ResearchOps professionals from 8 countries; confidence: VERIFIED). Some participants also admitted holding back information they would have shared with a person.

Synthetic users, meaning AI-generated stand-ins for real participants, fare worse. A review of 12 papers by the UX research firm MeasuringU tagged 9 findings as encouraging and 14 as discouraging, with no encouraging findings on qualitative depth (source: MeasuringU, April 2026; metric: count of findings across published studies; geography: published literature, no single region; confidence: VERIFIED). The authors concluded it is premature to rely on synthetic users for critical decisions.

AI also changes who shows up. A UK-wide interview study reported receiving hundreds of false expressions of interest generated or assisted by large language models, and filtered them with postcode checks, telephone confirmation, and short introductory calls (source: Quinn et al., International Journal of Qualitative Methods, 2026; metric: suspected false expressions of interest; geography: United Kingdom; confidence: VERIFIED).

On analysis, researchers disagree openly. A published open letter signed by 419 qualitative researchers from 32 countries argues that reflexive thematic analysis requires a subjective, positioned researcher, so generative AI does not fit that method (source: Jowsey et al., Qualitative Inquiry; metric: signatories in the published version; geography: international; confidence: VERIFIED). A published response argues that AI can serve as a human-led analytic aid if interpretive decisions stay with the researcher and its use is reported transparently (Friese et al., Qualitative Inquiry, published by SAGE on 17 March 2026). For marketers, the practical floor is the same on both sides of that debate: a person who can defend the interpretation has to own it.

Turning Findings Into Marketing Decisions

From insight to hypothesis to test

Qualitative findings earn their keep when they change what the team builds or tests. GOV.UK describes mixed methods as a way to combine the strengths of each approach: quantitative data is better at answering what effect something had, and qualitative data can show how and why those results happened. The same pattern works in marketing. Interview findings shape a buyer persona or a customer journey map, those artifacts generate hypotheses, and an experiment or an analytics check decides which hypotheses hold.

Keep the label honest at each stage. A persona built from a dozen interviews is a model of a dozen people's accounts, useful for writing and design, not a market estimate. Budget for the follow-up test before the research starts, so the finding has somewhere to go once the readout ends.

Frequently Asked Questions

What is qualitative research in marketing?

It is the systematic study of how people experience, describe, and decide about products, brands, and messages, using interviews, observation, and similar methods. Its purpose is to explain reasons and context, not to measure how common a view is.

How is qualitative research different from quantitative research?

Qualitative research interprets non-numerical evidence, such as interviews, open-ended answers, field notes, or documents, and answers why and how to fix. Quantitative research measures and counts, and answers how many and how much. Many studies combine both.

How many interviews do you need?

There is no universal number. A sample is adequate when it includes the people whose experience matters for the decision and the conversations are deep enough to explain the problem. GOV.UK notes that qualitative work usually needs far fewer participants than other evaluation methods.

Can qualitative findings be generalized to all customers?

Not directly. Qualitative studies describe the range of views among participants but are weak at saying how many people hold each view, and their results are hard to generalize. To estimate prevalence or effect, follow up with a quantitative or comparative study.

Can AI replace qualitative interviews?

Not for in-depth work, based on current evidence. AI interviewers handle structured interviews reasonably well but struggle with semistructured discovery, and reviews of synthetic participants find too many discrepancies with real human data to support critical decisions.