How to Conduct a Thematic Analysis

By Kevin Fernando · Oct 6, 2026 · 9 min read

Thematic analysis is a six-phase method for organizing qualitative data into meaningful patterns. You read through your data, assign codes to meaningful segments, group codes into candidate themes, refine and verify those themes for coherence and distinctness, then document each theme with supporting evidence from the original data.

Thematic analysis is a systematic method for finding patterns in qualitative data, interviews, documents, observations, open-ended survey responses. It works without specialized software and requires no prior coding experience, making it accessible to researchers new to qualitative work. The process is iterative: you move through six phases, often cycling back to refine earlier steps as your understanding deepens.

This guide walks through each phase in order, showing exactly what to do at each stage and how to recognize when you've completed it. The method is teachable because it is concrete: you assign labels to text segments, group those labels into larger categories, test whether those categories hold together, and then document what you found with direct evidence from your data.

What you'll be able to do, and what you need first

After completing thematic analysis, you will have identified the major patterns in your data and be able to describe what each pattern represents, supported by direct quotes or examples from the original material. You will know which themes are distinct from one another and which codes belong in each theme. You will also know when you have reached the point where additional analysis yields no new insights.

Before you begin, you need a research question that is clear enough to guide your coding. Vague or poorly articulated questions lead to confusion during analysis and ambiguous findings. You also need your data in a form you can work with repeatedly: transcribed interviews, typed field notes, or documents saved as text. The data should be complete and collected using methods appropriate to your research question; inadequate data collection compromises the quality of your analysis.

Familiarize yourself with your data

Read through your entire dataset once before you begin formal coding. Do not take notes or mark anything yet. Your goal is to absorb the overall patterns, tone, and scope of what you have.

On a second pass, read more actively. Jot down initial impressions: recurring words, surprising statements, contradictions, or concepts that appear across multiple data sources. Notice what seems important to your research question. This is not coding, it is orientation. You are building familiarity so that when you assign codes, you do so with a sense of the whole dataset, not in isolation.

Generate initial codes from your data

A code is a label for a meaningful segment of your data. Codes are more specific than themes and capture a single idea associated with a segment of data. You assign codes by reading through your data line by line and marking passages that relate to your research question, then labeling each marked passage with a brief, descriptive label.

Work systematically through your data. For each segment you identify as meaningful, write a code label that captures what is analytically interesting about it. Use short phrases: "time pressure", "lack of support", "unexpected benefit". The code should evoke the idea you are capturing. Multiple codes typically combine to create themes, so think of codes as building blocks, specific enough to stand alone, but designed to fit together later.

You can code on paper, in a spreadsheet, or in qualitative analysis software. Paper and spreadsheets are sufficient for small datasets; software becomes helpful as your dataset grows or if you need to track which codes appear in which sources.

Code inductively or deductively

Inductive coding means you generate codes directly from the data without predetermined categories. You read the data and let codes emerge. This approach is appropriate when your research question is exploratory: you are asking what patterns exist, not testing whether a specific pattern is present.

Deductive coding means you apply a set of codes you have defined in advance, based on theory, prior research, or your research framework. You then mark passages that fit each predetermined code. This approach is appropriate when your research question is hypothesis-driven: you are testing whether the data contains specific concepts or behaviors.

Most thematic analyses blend both approaches: you begin with some codes derived from your research question or theoretical framework, then add new codes as patterns emerge from the data itself. Choose your starting point based on what your research question asks.

Decide on your coding unit

A coding unit is the size of text you assign a single code to. You can code at the word level (marking individual words that carry meaning), the phrase level (short expressions), the sentence level (complete thoughts), or the passage level (multiple sentences or paragraphs that develop a single idea).

Your choice depends on your research question and the density of your data. If you are analyzing interview transcripts where speakers often develop ideas across several sentences, code at the passage level. If you are analyzing written documents with tightly packed information, code at the sentence or phrase level. Be consistent within your analysis: if you code one interview at the sentence level, code all interviews at the sentence level. Inconsistency produces unreliable results.

Organize codes into candidate themes

Once you have coded your entire dataset, sort your codes into groups. Codes that share a common idea or relate to the same concept belong together. These groupings are your candidate themes, the larger conceptual buckets that may become your final themes.

Read through your complete list of codes and look for patterns. Some codes will naturally cluster. Others will stand alone or fit into multiple groups; set those aside for now. You are not finalizing themes yet; you are testing which groupings make sense. Write each candidate theme as a short phrase that captures the idea binding its codes together. Expect to have more candidate themes than your final analysis will include; refinement happens in the next phase.

Review and refine your themes

Return to your original data and check whether each candidate theme holds together. Read all the coded passages assigned to one theme and ask: do these passages genuinely belong in the same category? Does the theme make sense as a coherent concept? If passages within a theme feel disconnected, the theme may need to be split, or some codes may need to move to a different theme.

Themes are overarching categories of common information related to your research phenomenon that tell a story about its dimensions. Each theme should tell part of that story. If a theme does not, it is not yet ready.

Check theme internal consistency

For each theme, verify that all codes within it genuinely belong together. Read the original data passages for each code in the theme. Ask whether they share a coherent meaning or whether some codes feel like outliers.

If you find codes that do not fit, move them to a more appropriate theme or create a new theme for them. If a theme has only one or two codes and they seem isolated, consider whether that theme is substantial enough to report, or whether those codes belong elsewhere. A theme should represent a meaningful pattern across your data, not a single occurrence.

Check theme distinctness

Ensure that your themes do not overlap. Read the definition of each theme and ask: could the same passage fit into two different themes? If yes, your themes are not distinct enough.

Distinctness does not mean themes are unrelated. Themes can be connected or nested within a larger story. But each theme should have clear boundaries. If two themes describe nearly the same concept, merge them or redefine one so the difference is clear. If a passage could reasonably belong in multiple themes, choose the theme it most directly illustrates and remove it from the others.

Define and name each theme

Write a clear description of what each theme represents. This description should explain the concept the theme captures, not simply list the codes within it. Your description answers: what does this theme tell us about the data in relation to the research question?

Then assign each theme a precise, memorable name. The name should evoke the theme's meaning and be specific enough that someone reading your report understands what the theme covers. Avoid generic names like "Challenges" or "Experiences". Instead, use names that capture the specific content: "Time scarcity as a barrier to adoption" is more useful than "Barriers".

Write up your findings with supporting evidence

For each theme, document what the data shows using direct quotes or examples from the original material. Do not paraphrase or summarize; include the actual words or passages from your data. This grounds your analysis and allows readers to verify that your interpretation fits the evidence.

For each theme, write a paragraph or two explaining what it represents, then provide one or more representative quotes or examples that illustrate it. Explain briefly what each quote shows. This structure demonstrates that your themes emerged from the data rather than being imposed on it.

Recognize when your analysis is complete

Data saturation is the point in analysis when new information no longer brings fresh insights to your research questions. Saturation is used in qualitative research as a criterion for discontinuing data collection and analysis. Saturation takes 4 forms: theoretical saturation, data saturation, code or thematic saturation, and meaning saturation.

For thematic analysis, you are typically working toward code or thematic saturation: the point where additional coding yields no new codes and no new themes. When you code new data and find that all segments fit into themes you have already identified, saturation has been reached. At that point, additional analysis is unlikely to reveal new patterns. You can stop coding and move to writing your findings.

Watch for common pitfalls

Thematic analysis fails most often in predictable ways. Understanding these pitfalls helps you recognize them in your own work and correct course before they compromise your analysis.

Forcing data into predetermined themes

The most serious pitfall is deciding what your themes will be before you code, then forcing data into those predetermined categories. This inverts the method. Thematic analysis is inductive: themes emerge from the data. When you impose themes in advance, you are no longer analyzing the data; you are sorting it into bins you created beforehand.

This happens when researchers begin with a strong hypothesis about what patterns they expect to find. They then code selectively, emphasizing data that fits their prediction and downplaying data that does not. The result is an analysis that confirms what the researcher already believed rather than discovering what the data actually shows.

Prevent this by generating codes from the data first, then organizing codes into themes. Let the themes emerge. If your data does not support the patterns you expected, that is a finding worth reporting.

Confusing themes with topics

A theme is not the same as a topic. A topic is what the data is about. A theme is a pattern that emerges across the data in relation to your research question.

Suppose you are analyzing interviews about workplace stress. "Workload" is a topic, it is what people are discussing. A theme might be "Workload as a source of stress" or "Strategies for managing workload demands". The theme captures a pattern or insight, not just a subject matter.

Check your themes by asking: does this theme tell me something about the data, or does it just name a subject? If you could replace your theme name with "stuff about X", it is probably a topic, not a theme. Reframe it to capture the pattern or relationship you are actually observing.

Coding without consistency

Inconsistent coding produces unreliable results. If you apply different standards to different sections of your data, you will miss patterns and create false ones.

Consistency means applying the same coding criteria throughout. If you code a particular type of statement as "resistance to change" in one interview, code similar statements the same way in all interviews. If you decide to code at the sentence level, maintain that standard across your dataset. If you change your coding approach midway through, go back and recode earlier material using the new standard.

To maintain consistency, document your coding rules as you work. Write down what each code means and what kinds of passages warrant that code. Refer back to these definitions as you code new material. If you find yourself applying a code in a new way, update your definition and consider whether earlier coding needs adjustment.

Do this automatically

Manually conducting thematic analysis across hundreds of interview transcripts or survey responses requires reading each dataset multiple times, assigning codes phrase-by-phrase or sentence-by-sentence, then reorganizing those codes into candidate themes, a process that can consume weeks or months depending on dataset size. Software can automate the initial coding phase and theme organization, reducing the manual labor required to move from raw data to identified patterns. Academia research runs this kind of analysis end to end and cites every figure back to its original source, across as many subjects as you need, with Mimir attaching a citation to every figure. It will not write your paper for you. It gathers and cites the evidence; the argument stays yours, and deliberately so. Every plan includes it, including the free one.

Start your first report and run it on something you already know well.

This article is general information and research commentary. It is not professional advice, and it does not account for your particular circumstances.