Discourse analysis is a family of qualitative methods for studying language-in-use — spoken, written, or multimodal text — not to catalogue what was literally said, but to examine how that language constructs meaning, social reality, identity, and power. It treats talk and text as social action rather than as a neutral window onto pre-existing facts: the guiding question is not just what does this text say but what is this text doing, and how.
Because “discourse analysis” names several distinct methodological traditions rather than one fixed procedure, conducting one well starts with choosing an approach whose theoretical commitments match your research question, then applying that approach’s specific analytic moves to a bounded corpus of data. This guide walks through the major traditions, a practical step-by-step process, and how discourse analysis relates to other qualitative methods already covered on this site.
What Discourse Analysis Is (and Isn’t)
At its core, discourse analysis studies patterns of language above the level of the individual sentence — how word choice, framing, narrative structure, and rhetorical strategy work together across a text or a body of texts to produce particular versions of events, identities, or relationships. It is distinguished from other forms of qualitative text analysis by its focus on language as constitutive: discourse is understood not merely to describe reality but to help build it, making certain interpretations available and others harder to voice.
It is not a synonym for close reading in general, nor for simple content analysis (counting or categorizing the presence of words or themes). Content analysis and basic thematic coding ask what topics appear and how often; discourse analysis asks how language is doing ideological, relational, or identity work, often below the surface of what a speaker or writer might consciously intend.
The Major Traditions
Researchers new to the method are often surprised that there is no single “discourse analysis” procedure to follow — the label covers several traditions with different theoretical roots and different units of analysis. Choosing among them should be driven by your research question, not by convenience.
Critical Discourse Analysis (CDA)
Associated most closely with linguist Norman Fairclough, Critical Discourse Analysis examines how language use reproduces or resists social power, dominance, and inequality. CDA treats every text as operating on three interconnected levels — the text itself (description), the discursive practice through which it is produced and consumed (interpretation), and the wider social practice it is embedded in (explanation) — and is explicitly critical in orientation, typically concerned with exposing how discourse serves the interests of dominant groups.
Foucauldian Discourse Analysis
Drawing on the work of philosopher Michel Foucault, this tradition treats discourses as historically situated systems of knowledge that make certain objects, categories, and subject positions available to think and speak about, while foreclosing others. Rather than asking who is “biased,” Foucauldian analysis asks what a discourse makes possible to know, say, or be — for example, how a particular clinical or institutional discourse constructs what counts as a legitimate patient, a productive worker, or a normal family.
Discursive Psychology
Developed by Jonathan Potter and Margaret Wetherell, discursive psychology re-specifies psychological concepts — memory, attitude, emotion, identity — as things people actively construct and do in talk, rather than as inner mental states that talk simply reports. It pays close attention to interpretive repertoires: the recognizable, culturally available patterns of terms and figures of speech people draw on to construct accounts, justify positions, and manage accountability in interaction.
Conversation Analysis (CA)
Conversation Analysis, developed from the sociological work of Harvey Sacks, Emanuel Schegloff, and Gail Jefferson, is the most technically demanding of the traditions. It examines the fine-grained sequential organization of naturally occurring talk — turn-taking, adjacency pairs (such as question-answer or invitation-acceptance), repair, and overlap — using detailed transcripts that capture pauses, overlaps, and prosody, most commonly using the Jefferson transcription system. CA is less concerned with broad social power than with how interactional order is produced turn by turn, though some researchers combine CA with CDA to bridge micro-interactional detail and macro social critique.
Some researchers also draw on narrative analysis alongside or in combination with discourse analysis when the data centers on how people construct and sequence stories about their experience; the two overlap but are not interchangeable, and it’s worth being explicit in your methodology section about which analytic lens you are actually applying.
How to Conduct a Discourse Analysis: A Step-by-Step Framework
1. Formulate a Discourse-Analytic Research Question
A workable research question names both the discursive object of interest (a policy debate, a set of clinical consultations, social media commentary on a topic) and the analytic lens (are you asking how power operates, how a category is constructed, how psychological states are managed in talk, or how a conversation’s structure is organized?). A vague question like “how is X discussed” is a starting point, not a finished research question — it needs to specify what kind of discursive work you expect to trace.
2. Select and Bound Your Corpus
Define your dataset — the specific texts, transcripts, or media you will analyze — and be explicit about the boundaries: date range, source, genre, and inclusion/exclusion criteria. Because discourse analysis is intensive, close-reading work rather than a method aimed at statistical generalization, corpora are typically small and purposively selected rather than large and randomly sampled; see the FAQ below on typical corpus size.
3. Prepare the Data: Transcription
If your data is spoken (interviews, meetings, clinical encounters), transcription decisions are analytic decisions, not just clerical ones — how much detail you capture depends on your tradition. Conversation Analysis typically requires the most granular transcription, commonly using Jefferson notation, which records overlaps, pauses timed to tenths of a second, stress, and other prosodic features, because CA’s claims rest on exactly how talk unfolds turn by turn. CDA, Foucauldian analysis, and discursive psychology can often work from a less granular orthographic transcript, though discursive psychology still attends closely to hesitations, repairs, and reported speech as they may be doing interactional work.
4. Close Reading and Coding
Read and re-read the corpus closely, coding for the discursive features relevant to your chosen tradition — for example, framing devices, presuppositions, and nominalizations (CDA); the categories and subject positions a text makes available (Foucauldian analysis); interpretive repertoires and rhetorical strategies of accountability (discursive psychology); or turn-taking and repair sequences (CA). This stage is iterative: initial codes are refined and reorganized as patterns become visible across the corpus.
5. Identify Discursive Strategies and Patterns
Move from cataloguing individual instances to identifying recurring strategies across the corpus — how particular groups, events, or categories are consistently positioned, what alternative framings are absent, and where contradictions or tensions in the discourse appear. Contradiction and variability are often analytically productive in discourse analysis, unlike in methods that treat inconsistency as a coding problem to be resolved.
6. Move from Description to Interpretation
Description alone — “this text uses passive voice here” — is not yet discourse analysis. The interpretive step connects the linguistic pattern to its social function or effect: what work does this passive construction do in obscuring agency, and for whose benefit? This is where the theoretical framework you chose in step 1 does its work, providing the vocabulary to move from what is said to what it accomplishes.
7. Reflexivity and Positionality
Because discourse analysis treats interpretation as constructed rather than simply extracted from the data, researchers are expected to be explicit about their own position: disciplinary background, relationship to participants or data, and the theoretical commitments shaping their reading. This is a rigor requirement, not a formality — see the related guide on research ethics in qualitative research, which covers positionality in more depth, and the guide on research paradigms for the epistemological groundwork (most discourse-analytic traditions sit within social constructionist or critical/interpretivist paradigms) underlying why reflexivity matters here.
8. Establish Rigor and Trustworthiness
Discourse analysis does not use quantitative validity or reliability in the traditional sense — you are not aiming for inter-coder statistical agreement on a fixed coding scheme. Instead, trustworthiness is typically demonstrated through a transparent, well-evidenced argument: showing your working by quoting extensively from the data, being explicit about how you moved from data to claim, addressing deviant or contradictory cases rather than ignoring them, and (in CA especially) grounding claims in participants’ own next turns rather than the analyst’s outside assumptions about what talk “really” means.
Discourse Analysis vs. Other Qualitative Methods
Discourse analysis is one option among several qualitative analytic approaches, and it’s worth being clear about how it differs from methods already covered on this site so you can justify your choice in a methodology section.
- Thematic coding / content analysis catalogues what topics or themes appear and how frequently, generally treating language as relatively transparent evidence of content. Discourse analysis instead asks how language actively constructs the phenomena it appears to describe.
- Grounded theory aims to build theory inductively from data, typically through systematic, iterative coding toward theoretical saturation. Discourse analysis is less focused on generating a novel theoretical model and more focused on the close analysis of how a specific body of language does its work — though grounded theory and discourse analysis are sometimes combined.
- Narrative analysis focuses specifically on how people structure and tell stories about experience — sequence, character, and plot — and overlaps with discourse analysis without being identical to it.
For the broader landscape of qualitative approaches and how they contrast with quantitative methods, see Qualitative Research vs. Quantitative Research and the dictionary entry on qualitative research.
Common Pitfalls
- Description masquerading as analysis. Simply summarizing or paraphrasing what a text says, without connecting linguistic features to their social function, is the most common weakness reviewers flag in discourse-analytic manuscripts.
- Mixing traditions without acknowledging it. Borrowing vocabulary from CDA and CA interchangeably without a stated rationale can read as theoretically incoherent; if you are combining traditions, say so explicitly and justify it.
- Cherry-picking illustrative extracts. Presenting only the quotes that support your argument, without addressing counter-examples or the overall pattern across the corpus, undermines the trustworthiness case discussed above.
- Under-specifying the corpus. Reviewers and readers need to know exactly what was included, excluded, and why — vague descriptions of “media coverage” or “interview data” without boundaries make the analysis hard to evaluate or replicate in spirit.
- Insufficient attention to positionality. Especially in CDA and Foucauldian work, failing to situate the analyst’s own standpoint can leave interpretive claims looking asserted rather than argued.
Tools
Discourse analysis is fundamentally close, interpretive reading rather than automated processing, but qualitative data analysis software can help manage larger corpora, organize codes, and retrieve linked extracts. See the guide to NVivo: Qualitative Data Analysis Software for one widely used tool that supports coding and managing discourse-analytic data, though the interpretive work itself remains the researcher’s.
Frequently Asked Questions
Is discourse analysis qualitative or quantitative?
Discourse analysis is a qualitative method. It is fundamentally interpretive, focused on close reading of a bounded corpus rather than on statistical generalization from a large sample. Some corpus-linguistics-informed variants incorporate quantitative frequency counts of specific linguistic features as a starting point for closer qualitative analysis, but the core analytic claims remain interpretive.
How is discourse analysis different from content analysis?
Content analysis typically catalogues and counts the presence of themes, words, or categories, treating language as relatively transparent evidence of content. Discourse analysis instead examines how language actively constructs meaning, identity, and power — asking not just what appears in a text but what that language is doing socially.
How many texts or transcripts do I need for a discourse analysis?
There is no fixed number. Because discourse analysis involves intensive close reading rather than statistical sampling, corpora are often small — sometimes a handful of texts or a single extended interaction analyzed in depth, particularly in Conversation Analysis. What matters more than raw count is that the corpus is clearly bounded, purposively selected, and sufficient to support the specific claims being made; this should be justified in the methodology rather than driven by a target number.
Do I need special transcription notation for discourse analysis?
Only if your data is spoken and your chosen tradition requires it. Conversation Analysis generally requires detailed notation such as the Jefferson system, which captures pauses, overlaps, and prosodic detail because CA’s claims depend on the precise sequential unfolding of talk. Other traditions, such as CDA or Foucauldian analysis, more often work from a standard orthographic transcript, though discursive psychology typically retains attention to hesitations and reported speech.
Which discourse analysis approach should I use?
Let your research question drive the choice. If you’re asking how language reproduces power or ideology, Critical Discourse Analysis fits. If you’re asking how a discourse constitutes categories, knowledge, or subject positions historically, Foucauldian analysis fits. If you’re asking how people construct psychological states or manage accountability in talk, discursive psychology fits. If you’re asking how the structure of interaction itself is organized turn by turn, Conversation Analysis fits.
How is discourse analysis evaluated in peer review?
Reviewers typically look for a clearly justified choice of tradition, a well-bounded and described corpus, transparent movement from data extracts to interpretive claims (usually demonstrated through extensive, quoted evidence rather than summary), engagement with contradictory or deviant cases, and explicit researcher reflexivity. As with other qualitative methods, the rigor case rests on a transparent, well-evidenced argument rather than statistical tests.
