Grounded theory is a qualitative methodology for generating theory from data through iterative coding, constant comparison, and theoretical sampling. Unlike thematic analysis, which describes patterns within a dataset, grounded theory aims to produce a substantive explanatory theory — an account of why and how something happens, not just what participants said about it. The process is circular rather than linear: data collection and analysis happen simultaneously, and early findings shape who you interview next.
The methodology was developed by Barney Glaser and Anselm Strauss in their 1967 book The Discovery of Grounded Theory, which has accumulated nearly 24,000 citations according to Semantic Scholar and remains one of the most influential methodological texts in social science. Since then, the approach has split into distinct variants, most notably Glaserian, Straussian, and Charmaz's constructivist grounded theory — each with different assumptions about the role of the researcher and the nature of theory.
This guide covers the 3 main variants, the full analytic process, and the practical question most researchers face: is grounded theory actually right for your project?
What are the 3 main variants of grounded theory?
The split in the field happened publicly in 1992, when Glaser published a rebuttal of the revised approach Strauss had developed with Judith Corbin. Kathy Charmaz later articulated a third position from within social constructionism. Understanding which variant you are using matters because they differ in coding procedures, the role of prior literature, and the criteria used to assess quality.
| Glaserian (classical) | Straussian/systematic | Constructivist (Charmaz) | |
|---|---|---|---|
| Philosophical stance | Post-positivist; theory is discovered | Post-positivist; theory is developed and verified | Constructionist; theory is co-constructed |
| Coding stages | Substantive coding (open + selective) → theoretical coding | Open coding → axial coding → selective coding | Initial coding → focused coding → theoretical coding |
| Prior literature | Avoid until analysis is complete; prevents contamination | Preliminary review permitted for conceptual clarity | Literature is engaged throughout as a theoretical resource |
| Researcher role | Neutral conduit between data and emerging theory | Active analyst applying systematic procedures | Co-participant who shapes the research through their position |
| Quality criteria | Fit, work, relevance, modifiability | Credibility, originality, resonance, and usefulness (adapted from Strauss & Corbin) | Credibility, originality, resonance, usefulness (Charmaz) |
| Best for | Pure inductive theory building; minimal preconceptions | Structured research environments; more prescriptive guidance | Social constructionist studies; focus on meaning and process |
For most researchers approaching grounded theory for the first time, Charmaz's constructivist approach and Strauss and Corbin's systematic version are the most accessible. Glaserian GT is the most demanding to apply faithfully and the least tolerant of pre-existing theoretical frameworks.
When should you use grounded theory?
Less often than people think. Grounded theory is frequently named as the methodology for studies that are actually doing something else — usually thematic analysis or framework analysis — because "grounded theory" sounds rigorous.
Use grounded theory when:
- Your research question is explanatory rather than descriptive ("how and why does X happen" rather than "what do people think about X")
- There is limited prior theory on your topic, or existing theory does not fit your context well
- Your dataset will grow as analysis proceeds — theoretical sampling requires flexibility in data collection
- You are prepared for an extended, iterative process; grounded theory projects typically take longer than thematic analysis
- You are aiming to produce a substantive theory that could contribute to the literature, not just a rich description
Do not use grounded theory when:
- Your data is already collected and fixed (GT requires ongoing theoretical sampling)
- Your research question is descriptive or evaluative
- You want a method that produces clean, reportable themes rather than an emerging theoretical framework
- You are working to a tight timeline (the iterative process is not compressible)
If your research question is "what are participants' experiences of X?", you are probably better served by thematic analysis or interpretive phenomenological analysis. If your question is "how does X process unfold over time and why?", grounded theory is worth the investment.
What is the grounded theory coding process?
Coding in grounded theory is more complex than in thematic analysis, and the stages differ across the three variants. Here is the Straussian/Charmaz sequence, which is the most widely used:
Initial or open coding
Read your transcripts line by line and code as you go. At this stage, codes should be close to the data — use gerunds (action words) where possible: "managing uncertainty", "negotiating expectations", "concealing effort". The point is to stay close to the participants' language while beginning to interpret.
Do not try to be efficient here. Code everything. You do not yet know what matters.
Focused or axial coding
Once you have coded a substantial portion of your data, begin to identify the most analytically important codes and develop them into categories. In Straussian terms, axial coding means examining the conditions, contexts, actions, and consequences associated with a category — essentially building a more complex picture of how things relate.
In Charmaz's framing, focused coding means selecting the codes that appear most frequently, analytically useful, or theoretically interesting, and using them to organise and sort large quantities of data.
Theoretical coding
In later stages, you look for how your categories relate to each other — what connects them into a coherent theoretical account. Glaser identified a range of theoretical codes (or "coding families") that describe the kinds of relationships between categories: causes, conditions, consequences, stages, strategies, and so on.
These families are close to what Skimle's agentic analysis encodes in its abductive templates. The classic skeleton is Antecedents, Mechanisms, Outcomes; an extended version adds Catalysts and Confounds; a process skeleton runs Initial Conditions, Triggering Events, Process Phases, Outcomes; and a critical-incidents skeleton runs Triggering Situation, Actor Response, Consequences, Learning. Picking one makes a theoretical-coding decision explicit rather than implicit, and every node can be renamed to your own concepts before the analysis runs.
Constant comparison
Throughout all stages, you compare: incident to incident, incident to code, code to code, category to category. The question you are always asking is "in what ways is this similar to, or different from, that?" Constant comparison is what drives theoretical sensitivity — the progressive ability to see what the data is telling you about your emerging theory.
Theoretical sampling
Unlike purposive or convenience sampling, theoretical sampling means choosing your next data source based on what your emerging theory needs. If your preliminary analysis suggests that a particular condition matters, you look for cases that vary on that condition. You cannot do genuine theoretical sampling if your data is already collected.
Theoretical saturation
Grounded theory does have a stopping criterion: theoretical saturation, which means that new data is no longer producing new theoretical insights. No new categories are emerging, and existing categories are fully elaborated. Note that this is different from thematic saturation — it is specifically about theory development, not just frequency of themes. See our guide on how many interviews you need for qualitative research for more on saturation and sample size.
What is the core deliverable of a grounded theory study?
A grounded theory study should produce a substantive theory — a conceptual account of how a process unfolds or how a phenomenon operates in a specific context. This is not a summary of what participants said, and it is not a list of themes. It is an explanatory framework with named categories, defined concepts, and propositions about how they relate.
The classic examples: Glaser and Strauss's original work on awareness of dying in hospitals; Charmaz's theory of "good days and bad days" in chronic illness. These produced frameworks that researchers could apply in different contexts, test against new data, and extend.
If your study ends with "three themes emerged: X, Y, and Z", you have not done grounded theory. You have done thematic analysis, which is also valuable — but it is a different thing.
How does memoing fit into the process?
Memo writing is not optional in grounded theory — it is a core analytic practice. Memos are theoretical notes to yourself: ideas about emerging categories, hunches about how things connect, reflections on your coding decisions, comparisons between cases.
Writing memos forces you to develop your thinking beyond coding and into theorising. A well-developed memo set is the scaffolding from which your final theoretical framework is built. Start writing memos from your first day of data collection, and do not wait until you "have enough data to say something."
In practice, this means having somewhere to capture your analytic thinking as it develops. Analyst notes and tags in Skimle serve this purpose — you can record your theoretical thinking alongside the coded data, linking observations directly to the excerpts that prompted them.
An agentic analysis also writes its own log memo, recording every step it took and the reasoning behind each one. Treat that as a complement rather than a substitute: it documents what the machine did, while your memos document what you thought. For a study you intend to publish, having both is useful, because the log gives you a ready account of the AI-assisted portion of the method when you come to write it up.
4 common mistakes in grounded theory research
1. Starting with hypotheses. Grounded theory is inductive (or abductive). Entering the field with hypotheses you want to test undermines theoretical sensitivity. You can have a general research focus; you should not have predetermined propositions.
2. Stopping at description. Many studies claim to use grounded theory but produce a descriptive account of what participants experienced rather than a theoretical account of why and how. Coding and categorising your data does not automatically produce theory — that requires the additional step of identifying conceptual relationships and building explanatory propositions.
3. Ignoring the literature until it is too late. Glaser's instruction to delay literature review is widely misunderstood. The point was to avoid letting prior theory prematurely close off theoretical sensitivity, not to remain ignorant indefinitely. Charmaz and Strauss both recommend preliminary engagement with the literature. By the time you are writing up, you must engage with existing theory to show how your theory extends or challenges it.
4. Confusing GT with just being thorough. "I used grounded theory because I was really systematic" is not a justification. Grounded theory is a specific methodology with specific procedures. If you used it, you should be able to describe your theoretical sampling decisions, your memo writing, and how you moved from codes to categories to theory.
How does AI-assisted analysis support grounded theory?
The iterative, corpus-wide nature of grounded theory coding is where AI tools can genuinely help. Manually tracking how a category evolves across 40 transcripts, and noticing when new data is and is not adding to theoretical understanding, is cognitively demanding.
Tools like Skimle support this by enabling you to see all coded excerpts across your full corpus simultaneously, run an inductive analysis to surface candidate patterns before you begin structured coding, and use metadata analysis to compare how categories manifest across different participant groups — which is central to theoretical sampling.
Where agentic analysis fits
Agentic analysis is the mode that maps most closely onto grounded theory work, because it starts from an explanatory question rather than from a request to describe a corpus. You state something like "why do some teams abandon the new protocol while others institutionalise it?", edit the framework Skimle proposes, and the analysis codes your documents against it, builds analytical themes, and writes up an answer with every claim cited back to a quote.
Three parts of it earn their place in a grounded theory project.
The abductive framing. Choosing a Diagnostic or Process type gives you a skeleton drawn from the same relational vocabulary as Glaser's coding families. This is the abductive move Charmaz's version of grounded theory already licenses: you bring a provisional conceptual structure, and the data pushes back on it.
Constant comparison as an explicit step. When you confirm the framework you also confirm two to four "broader dynamics", the analytical questions the synthesis stage investigates. For grounded theory the productive ones are close to what constant comparison asks: trace which mechanisms recur across cases and which are confined to particular ones; test whether the proposed antecedents are necessary, sufficient, or merely common; look for cases that contradict the emerging explanation. Those become the actual prompts driving the analysis rather than intentions you hold in your head.
Negative case analysis, run deliberately. After building each analytical theme, Skimle searches specifically for evidence that contradicts it, records what it finds, and rewrites the theme to make the claim conditional where the counter-evidence warrants. Grounded theory has always demanded this and it is the discipline most often skipped under time pressure. Within-case trajectory tracing is useful here too, since it shows how individual cases moved rather than only what was common across them, which is where process theorising usually finds its variation.
Re-runs matter for the same reason. Because a re-run preserves coding that already passed and redoes only the analytical layer, adding the next wave of theoretically sampled interviews and re-running gives you a revised account against the enlarged corpus rather than a fresh start. That is closer to the simultaneous collection-and-analysis cycle grounded theory actually requires than a one-shot analysis is.
What it does not do
Three limits worth stating plainly, because an academic reviewer will ask.
It does not do theoretical sampling. Deciding who to interview next based on what your emerging theory needs is a research judgement, and no part of the tool makes it. Skimle can show you which categories are thin and where your metadata has gaps, which informs the decision. The decision is yours.
Committing to a framework up front sits awkwardly with classical grounded theory. Glaserian purists will object, reasonably, that approaching data with a proposed structure is exactly what open coding is supposed to avoid. If you are working in that tradition, use the lighter modes for the open coding stage and reserve agentic analysis for later, once your categories have emerged. Straussian and constructivist grounded theory accommodate a provisional framework far more comfortably.
It does not produce your theory. It produces coded evidence, analytical themes with counter-evidence attached, and a written synthesis. Turning that into a substantive theory, with named concepts and propositions about how they relate, is the work grounded theory exists to do, and it stays with the researcher.
The theoretical work remains yours. AI tools accelerate the mechanics of coding; they do not do the constant comparison or the theorising. But they can reduce the hours spent managing data and increase the hours spent thinking about it. Our editorial on design criteria for AI-assisted qualitative analysis tools sets out the standards worth holding any such tool to before you use it in a study you intend to publish.
For academic researchers using Skimle in a grounded theory project, the academic researchers page covers the workflow in more detail.
Frequently asked questions
Is grounded theory inductive or deductive?
Grounded theory is primarily inductive, beginning with data and building toward theory rather than testing pre-existing theory against data. Charmaz's constructivist version is arguably abductive — moving between data and emerging theoretical ideas in a iterative cycle. See our guide on inductive, deductive, and abductive coding for a fuller explanation of these distinctions.
How many participants do you need for a grounded theory study?
There is no fixed number. Theoretical saturation drives sample size decisions, not a predetermined N. Published grounded theory studies range from fewer than 10 to over 60 participants depending on the phenomenon studied. Early-stage grounded theory projects often begin with 8-12 interviews and expand through theoretical sampling.
Can you use existing data (secondary data) for grounded theory?
Yes, with caveats. Theoretical sampling requires you to seek out specific data to develop emerging categories, which is difficult if your dataset is fixed. Secondary analysis is more compatible with Charmaz's constructivist approach than with classical GT. If using existing data, be transparent about the constraint and explain how you addressed it.
What is the difference between open coding in grounded theory and coding in thematic analysis?
Open coding in grounded theory is line-by-line, close to the data, and specifically oriented toward generating conceptual categories that will build into theory. Coding in thematic analysis is typically at the level of meaningful segments rather than line-by-line, and is oriented toward identifying patterns of meaning rather than building a theoretical framework. The end products are also different: themes vs theoretical categories.
Can I use agentic analysis in a grounded theory dissertation?
Yes, with the framing stated in your methods section. Agentic analysis suits the abductive middle and later stages of a constructivist or Straussian project, where you already have provisional categories and want to test how they relate, find contradicting cases, and build an explanatory account across a large corpus. It is a poorer fit for the open coding stage, and a poor fit for classical Glaserian work at any stage, because it begins from a proposed framework. Describe what the tool did, what you decided, and how you verified the coding, and the approach will survive examination. See our checklist of 20 questions to ask before publishing AI-assisted analysis.
Do I need to develop original theory, or can I extend existing theory?
You can do either. Grounded theory can be used to develop a substantive theory in a new area, to elaborate or refine an existing theory in a new context, or to generate a formal theory by integrating multiple substantive theories. The key criterion is that your theoretical contribution must emerge from systematic engagement with your data.
Ready to manage iterative qualitative coding without losing track of your emerging categories? Try Skimle for free and work through your coding across a growing corpus with full transparency from every category back to its source excerpts.
Related reading:
Qualitative evidence synthesis: thematic synthesis and meta-ethnography
Qualitative MEL data analysis: a guide for NGO and development teams
About the authors
Henri Schildt is a Professor of Strategy at Aalto University School of Business and co-founder of Skimle. He has published over a dozen peer-reviewed articles using qualitative methods, including work in Academy of Management Journal, Organisation Science, and Strategic Management Journal. His research focuses on organisational strategy, innovation, and qualitative methodology. Google Scholar profile
Olli Salo is a former Partner at McKinsey & Company where he spent 18 years helping clients understand the markets and themselves, develop winning strategies and improve their operating models. He has done over 1000 client interviews and published over 10 articles on McKinsey.com and beyond. LinkedIn profile
