Of all qualitative data analysis methods available to a researcher, thematic analysis is frequently described as straightforward, yet challenging to execute right. Straightforward because the logic is simple and linear: read the dataset, find patterns, label them. It gets executed inadequately for the same reason. Without a structured process underneath it, what passes for thematic analysis is often a researcher picking quotes that support a hypothesis they already held, labelling those quotes with broad category names, and presenting the result as findings.
When done properly, thematic analysis can be one of the most powerful tools in the qualitative researcher's toolkit. Flexible enough to work across interview & focus group transcripts, open-ended survey responses and diary entries; it is rigorous enough to produce findings that hold up in a client debrief, a board presentation, or even a peer-reviewed academic paper.
Thematic analysis is a method of extracting, identifying, analysing, and interpreting patterns of meaning across a qualitative dataset. The output is a set of themes, each one representing a coherent pattern in the data that is relevant to the research question, supported by evidence across multiple participants or sources.
The most widely cited framework for thematic analysis came from Braun and Clarke – a six-phased, clearly articulated framework that most research teams are working from, knowingly or otherwise.
A theme in thematic analysis is not defined by ‘how many times’ it occurs in the dataset. Instead, it is defined by whether it represents a meaningful pattern in relation to the research question. One participant saying something in a way that amplifies what six others gestured at vaguely, can constitute a theme. However, twenty participants mentioning something that has no bearing on the research question, is not a theme.
Before you touch the dataset, you need to make a foundational decision that shapes every step that follows.
Inductive thematic analysis lets themes emerge from the dataset, without a pre-existing theoretical framework. This is the right approach when the research is exploratory, when you do not know what you will find, and when participant language and unexpected patterns are as important as confirming what you suspected. A typical example of such analysis would be when the research question is – Which consumer motivations are driving category usage today?
Deductive thematic analysis starts with an existing framework, a set of theoretical concepts, a prior research model, or a client-defined hypothesis, and uses the dataset to validate, test, or extend it. You apply the framework to the data, rather than building one from it, ground up. This is the right approach when the research is confirmatory, when you are revisiting a topic with prior evidence, or when a client needs findings mapped to a specific strategic framework. A typical example of such analysis would be when the research question is – Are consumer motivations more Convenience-led today?
Most applied market research sits somewhere between inductive and deductive. You enter the data with a discussion guide that reflects certain hypotheses, and you code both to those hypotheses and to what emerges beyond them.
Throughout this section, the working example is a study of ten in-depth interviews with frequent consumers of fast food. The study is intended to explore consumers' relationships with healthier menu options. The research question being: Why do consumers who express interest in healthier fast food consistently fail to order it?
Reading transcripts is the step most researchers cut short under time pressure and is the step that determines the quality of everything that follows. It involves making notes, noticing patterns, flagging surprises, and developing an intuitive sense of the dataset before any formal analytical structure is applied.
In our example, this means when reading transcripts, you might notice that participants talk differently about healthier options when they are discussing what they order for themselves versus what they order for their children.
Coding is the process of labelling segments of data with a descriptor that captures what is meaningful about that segment in relation to the research question. At this stage, codes should be close to the data: descriptive, specific, and numerous. It is perfectly okay for the researcher to ‘overcode’, at this stage. Pruning too many codes too early in the analysis process is how you run the risk of missing nuanced patterns that might actually reflect a unique theme (at a later stage in the analysis process).
In our example, a participant saying "I always mean to get the salad but by the time I get to the front of the queue I just order what I know" might be multi-coded as: decision-making under pressure, habitual ordering, salience of healthy options at point of decision, and gap between intention and behaviour.
Once you have coded across the full dataset, group your codes into potential themes. This is where the analytical work becomes interpretative rather than descriptive. You are not just sorting codes into buckets; you are asking how these codes, taken together, are addressing the research questions.
The commitment gap (the distance between what participants say they want versus what they do
The social contract of fast food (participants feel that ordering healthier options violates an unspoken rule that fast food is best enjoyed at a restaurant rather than at home)
The effort demanded (additional cognitive work required of the individual, to evaluate a healthier menu item versus defaulting to a familiar order).
These 3 now become ‘theme candidates’ - clusters that could potentially be collapsed into a single, homogeneous theme.
See how flowres.io handles transcription, tagging and thematic analysis – all in a single, connected environment.
Take your theme candidates back to the data and test them. Does each theme work at two levels:
First, does it cohere internally: do the coded extracts within it actually belong together?
Second, does it differentiate externally: is it meaningfully distinct from the other themes in your set?
This review often collapses themes that seemed distinct in the earlier step, splits themes that seemed unified or surfaces a new theme that was hidden inside a code cluster you had labelled too broadly in the earlier step.
In our example, 'commitment gap’ and 'effort demanded’ turn out to be expressions of the same underlying dynamic (say, individual behavior/ preferences/ beliefs) and merge into a single theme. Instead, ‘social contract’ holds up as a separate theme, because it is driven by a social dimension (rather than an individual dimension).
Theme definition should capture what the theme is about, why it matters in relation to the research question, and what it does not include. The name should be specific enough to be meaningful and accessible enough to communicate clearly to a non-researcher.
In our example, a theme name for 'social contract’ might be: "Fast food as a guilt-free zone: why healthier options feel like a category violation."
The written output of thematic analysis is an analytical narrative that uses quotes for evidence. Each theme should be introduced, explained, evidenced with selected participant quotes, and connected back to the research question. The quotes you select should be the ones that most precisely evidence the claim being made – even if they don’t sound very evocative and exciting.
AI has a legitimate role in the thematic analysis process and a clearly defined ceiling.
It helps cut down the manual effort involved in transcript cleaning, suggesting initial codes in large datasets, flagging patterns across sessions, summary generation. A study comprising 15 focus groups that would take an analyst four days to code manually, can be given an initial AI-assisted code structure in minutes. The analyst applies interpretive judgement to refine the code structure, rather than building it from scratch.
However, AI cannot recognise when a participant is ‘performing’ or expressing only socially acceptable opinions. It cannot exercise the analytical judgement that turns a code cluster into a distinct theme that can stand strong, by itself.
flowres.io is built with this in mind. Its AI analysis layer suggests thematic structures and surfaces patterns across sessions, with one-click citations that take you directly to the participant quote that generated each suggested theme. The analyst reviews, accepts, refines, or rejects those theme suggestions. Thus, each theme that goes into a client deliverable is traceable back to source data; and has passed through human interpretative judgement.
Watch scheduling, backroom, AI transcription and analysis in action.
The bottom line
The six steps of Thematic analysis are what amplify its true power in Qualitative Data Analysis. Teams that code from a discussion guide end up producing topics that masquerade as themes. Following the process ensures that findings hold up to stakeholder scrutiny. A platform that handles grunt work yet keeps the human analyst in charge of interpretative judgement, helps conduct robust thematic analysis to produce evidence-backed, source-cited, defensible findings.
A method for identifying, analysing and interpreting patterns of meaning across a qualitative dataset, producing a set of themes that answer the research question; backed by evidence from the dataset.
Inductive lets themes emerge from the data; deductive applies a pre-existing framework to the data to test or extend it.
There is no fixed number; the right set of themes is the one that answers the research questions robustly and is supported by evidence across the dataset.
A code labels a specific segment of data; a theme is a higher-order pattern that captures what multiple codes, taken together, are saying about the research question.
AI can assist with the mechanical layer: initial coding, pattern flagging, and summary generation. The interpretive decisions that turn codes into defensible themes still require human analytical judgement.
She is a content writer specializing in the intersection of human inquiry and modern efficiency. Through her work at flowres.io, she explores how qualitative research is evolving and highlights the tools that help researchers maintain their creative flow.
Posted on: May 27, 2026
