Fairclough Critical Discourse Analysis

Getting past the theory to actually using this stuff

I started using Fairclough Critical Discourse Analysis on a corporate sustainability report for a client last year. The problem wasn't the framework itself. The problem was that the text was three hundred pages of carefully calibrated evasion. Fairclough's model is good, but it requires you to be extremely systematic or you end up writing your own prejudices into the analysis. Normally people lead with the three-dimensional model. Description, interpretation, and explanation. But if you're actually doing the work, you need to start with description because that's where everything falls apart if you rush it. Description is the bottom level. You take a stretch of text and you break it down into its formal features. Vocabulary choices, transitivity patterns, modality, cohesion, speech acts, nominalizations. You list what is there without saying what it means yet. I learned this the hard way with a set of hospital patient communications. I was analyzing consent forms for a NHS trust. I jumped straight to interpretation because I already knew what I thought the discourse was doing. That cost me two weeks. I found myself arguing against the tone instead of against the actual grammar of the text. Once I went back and did proper description, I caught something I had completely missed. The forms used extreme agency reduction throughout. Verbs were consistently passive. Patients were rendered grammatically invisible in the process descriptions. The interpretation part became almost trivial once the description was solid.

The second dimension is interpretation. This is where you connect the textual features to the processes that produced them and the processes they draw on. You ask how the discourse is being used, by whom, in what context. What conventions of discourse are at play here? How is the text being produced and consumed? A corporate press release about layoffs is produced under different constraints than a union pamphlet about the same event. The same lexical choices carry different meanings depending on the production context. This layer requires some ethnographic or institutional knowledge. You cannot do it from the text alone. The third dimension is explanation. This is where you link the discursive practices to the broader social and power structures. Ideology, hegemony, power relations. Fairclough draws heavily on Foucault here. The question is what social effects does this discourse produce and reproduce? Whose interests are served? This is the critical part. Not in the sense of being negative, but in the sense of uncovering how power operates through language rather than through explicit coercion. One thing that trips people up is treating the three dimensions as a linear pipeline. They are not. They are interrelated and you move back and forth between them iteratively. My usual workflow is to do a full description pass first, then start interpreting, then go back and refine the description with what the interpretation reveals. Then you cycle through again. A first pass on a medium-length text takes me about forty-five minutes. A full three-dimensional analysis of the same text takes roughly three hours including the iterative refinement.

Here is a specific example. Consider this sentence from a government briefing: "Mistakes were made in the implementation of the policy." Description would note the passive construction, the agentless verb, the nominalization of "implement" into "implementation." Interpretation would flag this as a classic political hedge that distributes responsibility while attributing none. Explanation would connect it to the broader discourse of managerialism in government, where institutional accountability is systematically dissolved into procedural language. The sentence is not bad writing. It is functioning exactly as designed. There is a real limitation to this method that most introductions skip over. Fairclough's approach assumes that discourse reflects and reproduces power structures, but it does not give you a reliable way to measure the strength of that relationship. You can demonstrate that a particular discourse exists and show how its grammatical features might support a certain power dynamic. You cannot prove that the discourse caused anything or that changing the discourse would change the outcome. Several of my colleagues who tried to use FCDS in policy evaluation runs into this wall. They end up making claims that go beyond what the textual analysis can actually support. If you need causal claims, you should combine it with something else. Quantitative content analysis or a corpus-based approach alongside it will give you more grounding. Another pitfall that nobody warns about is confirmation bias in the description stage. Your theoretical commitments shape what you notice. I spent months analyzing financial sector regulatory documents and I became so attuned to neoliberal discourse features that I started reading them everywhere. Even a straightforward customer service email would trigger my pattern recognition. The workaround was to have a colleague independently code a sample of the same texts and compare results before proceeding. That process cut my analysis time by roughly a third but saved me from publishing something I would have been embarrassed by later.

The most useful practical tip I can offer is to work with a coding frame early. Create a spreadsheet or table that maps textual features to analytical categories before you start reading. Transitivity, agency, modality, nominalization, metadiscourse, presupposition, argumentation schemes. When you encounter features in the text, tag them systematically. This keeps the description phase grounded and makes the transition to interpretation much smoother. Without this structure, you will spend more time organizing your notes than actually analyzing. I also recommend starting with shorter texts. A two-page speech or a single policy document. Fairclough himself often uses whole books or extended corpora, but that is after you have internalized the method. Your first three or four applications should be manageable in length so you can focus on getting the mechanics right rather than managing scale. Once description becomes automatic, you can expand to longer and more complex materials without losing analytical control.