Open, Axial, and Selective Coding: A 2026 Guide
Qualitati Research Team · 2026-09-24 · 11 min read
Short answer: Open, axial, and selective coding are the three analysis stages of Strauss and Corbin's grounded theory. Open coding breaks data into labeled concepts. Axial coding relates those concepts into categories by asking about conditions, actions, and consequences. Selective coding integrates the categories around one core category that explains the main story. The stages overlap and loop; they are not a one-pass pipeline.
Open, axial, and selective coding is the vocabulary many PhD students meet first when a supervisor says "use grounded theory." It is also where many methods chapters go wrong: the three terms get used as generic labels for "first pass, second pass, final themes," which is not what they mean. This guide explains each stage, works through one running example, and adds a checklist you can use to show examiners and reviewers that the procedure was actually followed.
The topic is timely. In August 2026, a multi-agent system called AutoTraceGT automated open, axial, and theoretical coding on AI-agent logs, and a July 2026 preprint on human-LLM collaborative inductive coding kept humans in charge of category definitions. Knowing what each stage is supposed to do is now the only way to judge whether a tool is doing it.
Last updated: September 24, 2026
Key takeaways
- Open coding produces concepts; axial coding produces relationships; selective coding produces an integrated theory around one core category.
- The three-stage vocabulary belongs to the Straussian tradition (Strauss and Corbin, 1990). Glaser's classic grounded theory and Charmaz's constructivist version use different stage names.
- Axial coding is the stage most often skipped. Grouping codes into buckets is categorizing, not axial coding, unless you specify how categories relate.
- The stages run iteratively with constant comparison, memo writing, and theoretical sampling. Coding everything once, in order, is not grounded theory.
- AI tools can speed up open coding and propose groupings. Relating categories and choosing a core category remain interpretive decisions a researcher must own and document.
Where open, axial, and selective coding come from
Grounded theory began with Barney Glaser and Anselm Strauss's The Discovery of Grounded Theory (1967). The two later diverged. Strauss and Juliet Corbin's Basics of Qualitative Research (1990) set out the three-stage procedure of open, axial, and selective coding, together with a "coding paradigm" for relating categories. Glaser (1992) rejected the paradigm as forcing data into a preset frame and kept substantive (open and selective) and theoretical coding. Kathy Charmaz's constructivist grounded theory (2006) uses initial and focused coding instead.
For a clear account of the Straussian procedure and the paradigm, see Vollstedt and Rezat (2019). The practical point: if your methods section says "open, axial, and selective coding," you are signalling the Straussian version, and reviewers will expect to see its logic, including the paradigm or an explicit reason for not using it. For how grounded theory differs from other approaches, see our thematic analysis vs grounded theory guide.
Stage 1: What is open coding?
Open coding is the line-by-line or incident-by-incident breaking down of data into concepts, each with a label and a short definition. The researcher stays close to the data, asks "what is happening here?", and compares each new incident with earlier ones (constant comparison). Codes are provisional and numerous.
Running example. Imagine interviews with early-career nurses about their first year on night shifts. An open-coding pass on one excerpt might produce:
- "Pretending to know" (an in vivo code): hiding uncertainty from senior staff.
- Checking with a peer instead of the charge nurse.
- Rehearsing handover in the car.
- Counting hours until morning.
In open coding you also note properties and dimensions: "checking with a peer" varies by how often, with whom, and how risky the question is. These dimensions become the raw material for axial coding.
Stage 2: What is axial coding?
Axial coding relates concepts and subcategories to a category along its "axis." The researcher asks under what conditions a phenomenon occurs, how people act and interact in response, and with what consequences. This is the coding paradigm: conditions, actions-interactions, and consequences. Later editions of Basics present the paradigm in this simplified three-part form rather than the fuller 1990 list (causal, contextual, and intervening conditions, strategies, consequences).
Running example. Several open codes cluster around a category we might call managing visible competence:
| Paradigm element | Question asked | Codes related to "managing visible competence" |
|---|
| Conditions | When and why does it happen? | Thin night staffing; evaluation period; senior staff perceived as judging |
| Actions-interactions | What do people do about it? | "Pretending to know"; checking with a peer; rehearsing handover |
| Consequences | What results? | Delayed escalation; peer network forms; exhaustion after shifts |
Notice the difference from simple grouping. A folder named "coping" holding the four codes says nothing about how they connect. The axial table makes a claim that can be checked against more data: thin staffing and evaluation pressure lead to hiding uncertainty, which leads to delayed escalation. Write that claim in an analytic memo and go looking for cases that break it.
Stage 3: What is selective coding?
Selective coding integrates the categories around one core category: the concept that appears frequently, connects to most other categories, and explains the most variation in the data. The researcher writes the "storyline," refines weakly developed categories, and fills gaps through theoretical sampling until new data stop changing the categories' properties (theoretical saturation; see our guide to data saturation).
Running example. After more interviews, managing visible competence might be subsumed under a core category such as earning the right to not know: a process in which new nurses move from hiding uncertainty, through peer-mediated checking, to openly asking once they feel their competence is established. The other categories (night-shift conditions, peer networks, exhaustion) now hang off that process.
How the three stages compare
| Stage | Core question | Main output | Common mistake |
|---|
| Open coding | What is happening here? | Many provisional concepts with properties and dimensions | Coding topics ("staffing") instead of actions and meanings |
| Axial coding | How do these concepts relate? | Categories with conditions, actions, consequences | Sorting codes into folders with no stated relationships |
| Selective coding | What is the main story? | One core category and an integrated theory | Listing five parallel themes and calling one "core" |
For how concepts, categories, and themes differ as analytic units, see codes, categories, and themes.
The Grounded Theory Coding Integrity Checklist
This checklist is a Qualitati-developed tool for PhD students and research teams. Use it before writing the methods chapter; each "no" is something an examiner is likely to ask about.
- Tradition named. Does the methods section say which grounded theory version you follow (Straussian, Glaserian, constructivist) and use that version's stage names?
- Open codes are active. Do most open codes describe actions, processes, or meanings rather than topics?
- Relationships are explicit. For each category, can you state at least one condition, one action-interaction, and one consequence, or a stated alternative logic?
- Constant comparison is visible. Is there evidence (memos, codebook versions) that codes changed after comparing incidents?
- Theoretical sampling happened. Did later recruitment or interview questions change because of emerging categories?
- One core category. Can you explain in two sentences why this category, not another, is central?
- Negative cases handled. Are cases that do not fit the storyline reported and explained? (See negative case analysis.)
- AI use documented. If a tool proposed codes or groupings, does your audit trail show what it suggested and what you accepted, changed, or rejected?
Can AI do open, axial, and selective coding?
AI can now attempt all three stages, but the evidence supports using it to accelerate open coding and propose structure, not to replace the researcher's interpretive decisions. Three 2025-2026 studies illustrate the state of the field:
- LOGOS (arXiv 2509.24294) combines LLM coding, semantic clustering, and graph reasoning to build a hierarchical theory, and reports 80.4% average alignment with expert-developed schemas. We covered it in Can AI build grounded theory?
- AutoTraceGT (Lu et al., submitted August 31, 2026) runs open, axial, and theoretical coding on AI-agent trajectories until saturation and recovers 73-91% of the failure modes in human-annotated taxonomies across six datasets. Its data are machine logs, not human interviews, so transfer to interview research is untested.
- Human-LLM collaborative coding (Liu et al., submitted July 30, 2026) coded 45,000 educator messages with an LLM proposing labels, while humans kept authority over category definitions and merges. Independent human coding of 2,560 messages added five codes the LLM phases had missed.
The pattern is consistent: alignment with expert schemas is substantial but incomplete, and the missing part is exactly what a thesis defends. Treat AI output as a set of candidate open codes and candidate relationships to test in memos, never as the finished axial or selective coding.
Where Qualitati fits
Qualitati is a European, budget-friendly qualitative research platform for universities and research firms, with GDPR and data privacy as central priorities. For a grounded theory project it can support the data-collection and early coding work, while the relating and integrating stays with you:
- AI-moderated interviews in text and voice help when theoretical sampling calls for new interviews with a revised guide.
- QDA Workspace supports AI-assisted inductive coding, which can serve as a first open-coding pass you then review line by line, plus deductive coding once your categories stabilize, codebook generation, and theme visualization.
- ThemeLens synthesizes themes with participant-anchored quotes across up to 100 transcripts, which is useful for checking whether a proposed category holds across the whole corpus. It performs thematic synthesis, not axial coding, so treat its output as evidence for your memos rather than as your paradigm model.
- Multilingual research in 10 languages, useful for cross-national grounded theory studies.
Qualitati does not have a dedicated coding-paradigm or core-category feature. If your design depends on formal axial diagrams, you may still draw them in your memos or in a traditional QDA package. See transparent pricing, or compare options in NVivo alternatives.
Limitations and methodological concerns
- The paradigm is contested. Glaser's "forcing" critique still matters. Some researchers use the paradigm loosely or not at all; either is defensible if stated.
- Stages are heuristic, not sequential. Writing "we first did open coding, then axial, then selective" can suggest a linear process that reviewers read as a misunderstanding.
- Small, fixed samples limit theory. Without theoretical sampling, the result is usually better described as grounded-theory-informed thematic analysis.
- AI studies cited here are preprints with different data types and evaluation criteria. Their alignment figures are not directly comparable and have not all been peer reviewed.
Human-review note: the running example is illustrative, not drawn from a real study. Check stage definitions against the edition of the methods text your department expects.
Who this is for, and when not to use this approach
This guide is for PhD students, supervisors, and research teams planning or defending a Straussian grounded theory study. Do not use open, axial, and selective coding if your aim is to describe themes across a fixed dataset (use reflexive thematic analysis; see our thematic analysis guide), if you are applying a predefined framework (use framework analysis or deductive coding), or if you follow Glaserian or constructivist grounded theory, which have their own coding vocabularies.
FAQ
What is the difference between open, axial, and selective coding?
Open coding labels concepts in the data. Axial coding relates concepts into categories by conditions, actions-interactions, and consequences. Selective coding integrates the categories around one core category into a theory.
Is axial coding the same as grouping codes into themes?
No. Grouping creates buckets. Axial coding specifies how concepts relate, for example which conditions trigger which actions and what consequences follow.
Do I have to use the coding paradigm in axial coding?
Not strictly. Many studies adapt it or use another relational logic. State what you did and why, since reviewers of Straussian studies will look for it.
What is the difference between selective coding and theoretical coding?
Selective coding is Strauss and Corbin's integration stage around a core category. Theoretical coding is Glaser's term for relating substantive codes using theoretical coding families. They serve similar purposes in different traditions.
Can I use AI for open coding in a grounded theory study?
Yes, as a first pass you review, provided you document what the tool proposed and what you changed. Recent studies show useful but incomplete alignment with expert coding.
How many interviews does grounded theory need?
There is no fixed number. Sampling continues until theoretical saturation, when new data no longer change your categories' properties.
Bottom line
Open, axial, and selective coding are three linked moves: name concepts, relate them, and integrate them around a core category. Axial coding is where most studies are weakest, and it is also where AI tools help least. Use AI to speed up open coding, use memos and the checklist above to make the relationships explicit, and name your grounded theory tradition clearly. When you are ready to collect and code data, start free with 30 credits, no credit card required.