"We used grounded theory, with open, axial and selective coding." The sentence is common, but it hides a real choice: the three labels come from one school of grounded theory, and other schools code differently. This guide explains what each stage does, how the three main traditions differ, and where software help is reasonable and where it is not.
What coding is for in grounded theory
In grounded theory, coding is not sorting data into boxes. It is the way you move from description to concepts, and from concepts to a theory that is grounded in the material. It runs alongside data collection, uses constant comparison (incident with incident, then concept with concept), and is documented in memos. If your study collected all interviews first and coded afterwards from a fixed list, you are probably doing something closer to content analysis, and the label should say so.
The three stages in the Strauss and Corbin tradition
The version taught most often goes back to Strauss and Corbin. The stages overlap in practice, so treat them as emphases, not as steps you tick off.
- Open coding. You break the material open, often line by line at the start, and name what you see. Codes should be concepts, not summaries. Similar concepts are grouped into categories, and you work out their properties and dimensions.
- Axial coding. You take one category at a time as an axis and relate it to others: what conditions lead to it, in what context it occurs, which strategies people use, and with what consequences. The coding paradigm provides these questions.
- Selective coding. You decide on a core category, relate the other categories to it, check these relations against the data, and fill gaps, if necessary by collecting further cases.
An invented example: in a study of new teachers, an open code might be "checking with colleagues first". In axial coding it becomes a strategy for dealing with uncertainty about grading, shaped by how approachable the head of department is. In selective coding the core category might be something like "securing oneself", with several strategies linked to it.
Typical mistake: using these three words as chapter headings for what was really a theme list. Open coding that only paraphrases passages produces topics, and topics do not become a theory by being renamed.
Glaser, Strauss and Corbin, Charmaz in one paragraph each
Glaser distinguishes substantive coding (open coding, then selective coding once a core category is found) from theoretical coding, where you think about how your categories relate to each other, helped by so-called coding families. He objects to a fixed paradigm and argues that the relations should emerge from the data instead of being imposed. He is also sceptical of an early, thorough literature review on the topic.
Strauss and Corbin offer more structure: open, axial and selective coding, the coding paradigm, and room for prior knowledge and literature as sources of sensitivity. Many newcomers find this easier to follow; critics see the risk of forcing data into the paradigm. Corbin's later writing stresses that the process is less linear than the stage names suggest.
Charmaz represents a constructivist position. She treats categories as constructed in the interaction between researcher and participants, not as simply discovered. Her coding has an initial phase, close to the data and often phrased with action words, and a focused phase, in which the most significant or frequent initial codes are used to sort larger amounts of material. Theoretical coding is optional and is used to integrate the analysis.
Pick one tradition, cite it, and use its vocabulary consistently. Mixing "selective coding" in Glaser's sense with "axial coding" from Strauss and Corbin without saying so will confuse a careful reader. For a German-language walk-through of the same stages, see Grounded Theory kodieren.
Typical mistake: coding first, then sampling
Grounded theory ties coding to theoretical sampling: what you analyse decides whom you ask or observe next. If you cannot name a single decision about sampling that came out of your analysis, say so openly in the methods section and describe your approach as inspired by grounded theory techniques. That is accurate and defensible. Claiming the full methodology without its core mechanism is not.
What software can and cannot do here
The decisions that make grounded theory what it is cannot be delegated: which case to sample next, what belongs in a memo, which category becomes the core. An AI tool can still be useful as a second reader. In Themera every code is backed by a verbatim quote, and quotes that are not exactly in the material are rejected automatically, so you can see what a code rests on. A coverage indicator shows which responses got no code, which helps you spot material you have not yet looked at closely. You can edit the codebook and re-code the material, and an intercoder agreement check (blind re-coding, Cohen's kappa) lets you compare your own coding with the AI's.
Be honest about the fit, though. A tool that applies a codebook to the material works in a codebook logic, which sits closer to content analysis than to open coding in Glaser's sense. Used for grounded theory, it is a comparison to your own open codes and a check on coverage. It does not replace line-by-line work, memoing or the theoretical decisions. Pseudonymise your transcripts before analysis, and ask your supervisor or the data protection office what your institution requires.
Describing AI support in your methods section
If you used AI at any stage, state at which one, what you did with its output, and what remained your interpretation. For the general mechanics of coding and the role of codebooks, see how to code qualitative data; for a method that is often confused with grounded theory, see thematic analysis in six phases.
Wording to adapt
A template with placeholders, not a description of a real study. Delete whatever does not match what you did.
Data were analysed following the grounded theory approach of [Strauss and Corbin / Glaser / Charmaz, citation]. Data collection and analysis alternated: after the first [number] interviews, further participants were selected to [theoretical sampling criterion]. Transcripts were coded [line by line] in an initial/open phase; codes were compared constantly and grouped into categories, and analytic ideas were recorded in memos. [Axial coding related categories using the coding paradigm. / Focused coding was used to sort larger parts of the material.] [Core category] was identified as the core category and tested against the full data set. [If applicable: An AI-assisted tool was used to generate a comparison coding; each code was checked against its supporting quote, and differences from the researcher's own coding were discussed.] Sampling ended when [justification for saturation].
Conclusion
Open, axial and selective coding describe one tradition, not grounded theory as such. Choose the tradition that fits your question and your experience, show the link between coding and sampling, and be precise about which parts of the methodology you actually used.
See a sample analysis with codes, verbatim evidence and the coverage indicator, or try it with your own material: three analyses free, no card needed.