The types of discourse analysis,
explained and applied.

A different approach to discourse analysis.

Discourse analysis is a qualitative research method that examines how language is used in real-world contexts to construct meaning, shape identities, exercise power, and accomplish social actions. It goes beyond what is said to analyze how and why it is said in particular ways, and what effects those language choices have. The seven types below share that starting point but diverge sharply in theory, method, and the kind of question each is built to answer.

The seven types, theorist by theorist

Each type carries its own key theorists, its own typical applications, and its own way of reading a transcript. This is the fast reference; the sections below go deeper on the two that most researchers underestimate.

• Critical Discourse Analysis (CDA), Fairclough, van Dijk, Wodak. Studies political rhetoric, media bias, and policy language. Example: how metaphors of "flood" or "invasion" in immigration coverage construct people as threats rather than individuals.
• Conversation Analysis (CA), Sacks, Schegloff, Jefferson. Studies turn-taking, repair, and adjacency pairs in healthcare, courtroom, and service talk. Example: patients repeating symptoms when they feel unacknowledged, correlating with lower satisfaction.
• Foucauldian Discourse Analysis (FDA), Foucault, Parker, Willig. Studies how institutions construct knowledge, normality, and subject positions. Example: how the shift from "insanity" to "mental health condition" reflects changing power between medicine, patients, and society.
• Narrative Discourse Analysis, Labov, Riessman, Bamberg. Studies story structure, positioning, and identity in illness narratives, organizational stories, and life-history interviews. Example: cancer survivors who frame recovery as a "journey" reporting different outcomes than those who use "battle" language.
• Multimodal Discourse Analysis (MDA), Kress, van Leeuwen, Machin. Studies how text, image, sound, gesture, and layout combine in ads, social content, and video. Example: how a candidate photograph, slogan, color scheme, and campaign-ad music work together to build an identity no single mode could carry alone.
• Corpus-Based Discourse Analysis, Baker, Partington, Stubbs. Uses concordance, collocation, and keyword analysis across large text or speech collections. Example: 20 years of climate reporting tracked from "climate debate" to "climate crisis" through keyword shift.
• Interactional Sociolinguistics, Gumperz, Tannen. Studies contextualization cues, code-switching, and framing in cross-cultural and gatekeeping encounters. Example: interviewers misreading a candidate's cultural discourse strategies as a lack of qualification.

Reading the recording, not just the transcript

Most discourse analysis still starts and stops at a text file, even when the underlying data was spoken. That leaves a gap for the types built to read interaction and multimodal meaning: conversation analysis needs the pauses, overlaps, and intonation that a plain transcript strips out, and multimodal discourse analysis needs the images, gestures, and sound a transcript never captured in the first place.

Speak AI reads recordings on three layers at once: the words spoken, the voice behind them (tone, energy, pacing, and pauses), and what is visible in the recording itself (facial expression, gesture, on-screen content). For a CA researcher, that means timestamped speaker turns without hand-notating every overlap. For an MDA researcher working with video interviews or social content, it means one coded record instead of separately logging the visual and verbal channels. A frustrated pause, a rising pitch on a key word, a hesitation before naming a competitor, all become part of the data instead of getting lost between the recorder and the page.

How to conduct discourse analysis, step by step

• Define your research questions. Focus on how language functions, not just what is said. Many researchers sharpen this stage against a grounded theory topic list, even for studies that don't test formal hypotheses (most qualitative studies don't need one).
• Select and collect your data. Written texts, transcribed speech, social posts, interview recordings, or policy documents. Speak AI's qualitative research platform og automatisert transkripsjon speed up collection considerably.
• Transcribe and prepare your data. CA requires precise notation of pauses and overlaps; CDA can work with standard transcriptions. AI-powered tools like Speak's audio-to-text converter handle the first pass, which you refine for your analytical needs.
• Conduct initial coding. Read through multiple times, letting your theoretical framework guide what you attend to. A transkripsjonanalysator og AI-assisted qualitative coding software can surface first-pass themes before you code line by line.
• Analyze in depth. Connect specific language choices, word choice, metaphor, grammar, rhetorical strategy, to the macro-level social processes and theory your framework predicts.
• Interpret and write up. Present rich examples that show the evidence behind your interpretation, and include reflexivity about your own position as analyst.

AI tools do not replace the interpretive work that defines discourse analysis, and the theoretical judgment behind a CDA reading of a metaphor or an FDA reading of a subject position still has to come from the researcher. What changes is how much time reaches that judgment: less spent transcribing, coding first passes, and searching across files, and more spent on the analysis that actually answers your research question. Some teams also connect this workflow to an AI meeting assistant for live recruitment interviews, or AI-agenter to automate repetitive collection tasks.