Short answer: First-cycle coding is the initial pass through qualitative data, where you attach short labels to segments of text before grouping them into categories and themes. Johnny Saldaña's Coding Manual groups first-cycle methods into six families. Descriptive, in vivo, process, values, emotion and initial coding are the ones most researchers use. Choose the method that fits your research question, not the one your software makes easiest.
Last updated: October 29, 2026
Most methods chapters say "the data were coded inductively" and stop there. Reviewers increasingly ask a sharper question: coded how? First-cycle coding is where that answer lives. It is the first systematic pass through your interview transcripts, field notes or open-ended survey answers, and the method you pick decides what your later themes can and cannot say.
This guide explains the main first-cycle coding methods, shows the same interview excerpt coded six different ways, and gives you a selector matrix for matching a method to your research question. It also covers how far AI-assisted coding can help at this stage, and where it should not be trusted on its own.
Key takeaways
- First-cycle coding labels data; second-cycle coding organizes the labels. Pattern, focused and axial coding come later (see our second-cycle coding guide).
- Saldaña's fifth edition (SAGE, 2025) groups first-cycle methods into grammatical, elemental, affective, literary and language, exploratory, and procedural families.
- The method follows the question. "What happens?" points to process coding; "what matters to people?" points to values coding; "what is this about?" points to descriptive coding.
- Combining two methods is normal and defensible, as long as you report which ones and why.
- AI is most reliable for descriptive and structural codes and least reliable for interpretive ones such as values and emotion. A human should review every AI-suggested code before it enters the codebook.
What is first-cycle coding?
First-cycle coding is the set of methods used during the initial coding of qualitative data. Each segment, from a word to a paragraph, gets a short label that summarizes or captures its meaning. The output is a long, often messy list of codes. Second-cycle coding then reorganizes those codes into categories, themes or a theoretical model.
The distinction comes from Johnny Saldaña's The Coding Manual for Qualitative Researchers, now in its fifth edition. According to SAGE's product page, the 2025 edition covers more than 35 coding methods and adds material on CAQDAS and AI tools. Its table of contents splits first-cycle methods into six chapter families, plus a separate chapter on "themeing the data."
A review of the fifth edition in The Qualitative Report (Dos Santos, June 27, 2026) highlights that the book treats generative AI as a support tool. Interpretation, contextual judgment and ethical accountability stay with the researcher.
The six families of first-cycle coding methods
The family names below follow the fifth edition's table of contents. The example methods in each family follow the widely used fourth edition (2021); check the edition you cite for exact placement.
| Family | What it captures | Example methods |
|---|---|---|
| Grammatical | Basic properties of the data and its structure | Attribute, magnitude, subcoding, simultaneous coding |
| Elemental | Foundational summaries of content | Structural, descriptive, in vivo, process, initial, concept coding |
| Affective | Feelings, values and conflict | Emotion, values, versus, evaluation coding |
| Literary and language | Story, performance and talk | Dramaturgical, motif, narrative, verbal exchange coding |
| Exploratory | Preliminary or hypothesis-led passes | Holistic, provisional, hypothesis coding |
| Procedural | Pre-set, standardized schemes | Protocol, domain and taxonomic, causation coding |
In practice, most interview studies in education, health, management and UX research draw on six methods: descriptive, in vivo, process, values, emotion and initial coding. The rest of this guide focuses on those.
First-cycle coding methods with examples
To make the differences visible, here is one short, illustrative interview excerpt (written for this guide, not real participant data), coded six ways.
"When my supervisor went on leave, I just stopped writing for two months. I kept telling myself I'd start on Monday. Honestly, I think I needed someone to be disappointed in me."
Descriptive coding
Descriptive coding labels the topic of a passage with a noun or short noun phrase. It answers "what is this about?" Codes: SUPERVISOR ABSENCE, WRITING HABITS. It is the best starting point for large or mixed datasets and for teams new to coding, but on its own it rarely produces insight.
In vivo coding
In vivo coding uses the participant's own words as the code. Codes: "START ON MONDAY", "SOMEONE TO BE DISAPPOINTED IN ME". It keeps participants' voices visible and suits studies of youth, professional jargon or marginalized groups. Our in vivo coding guide covers it in depth.
Process coding
Process coding uses gerunds ("-ing" words) to capture action and change over time. Codes: STOPPING WRITING, POSTPONING THE START, SEEKING EXTERNAL ACCOUNTABILITY. Use it when your question is about what people do, how routines form, or how something unfolds. It is a natural feed into grounded theory.
Values coding
Values coding marks a participant's values (V), attitudes (A) and beliefs (B). Codes: B: I NEED OUTSIDE PRESSURE TO WORK, V: BEING ACCOUNTABLE TO OTHERS. It suits identity, ethics and decision-making research. It is also the most interpretive method here, so document your reasoning in an analytic memo.
Emotion coding
Emotion coding labels emotions the participant reports or that the researcher infers. Codes: GUILT, SELF-DOUBT. Mark which emotions are stated and which are inferred. In this excerpt, neither word is said aloud, so both are inferences that a second coder should check.
Initial coding
Initial coding, from Kathy Charmaz's constructivist grounded theory, is an open, line-by-line pass that stays close to the data and keeps codes provisional. Codes: LOSING STRUCTURE, DEFERRING, LOCATING MOTIVATION OUTSIDE ONESELF. It overlaps with process coding but is deliberately tentative. Our open, axial and selective coding guide explains how it connects to later grounded theory stages.
First-Cycle Coding Method Selector
This matrix is a Qualitati planning tool. Start from the wording of your research question, then pick a primary method and, if needed, one secondary method.
| If your research question asks… | Primary method | Good secondary method | AI first pass? |
|---|---|---|---|
| What topics come up? (exploratory, large dataset) | Descriptive | In vivo | Useful, with human review |
| How do people talk about X in their own terms? | In vivo | Descriptive | Useful for flagging phrases; human selects |
| What do people do, and how does it change? | Process | Initial | Partial; check sequence and context |
| What matters to people, and why? | Values | Emotion | Draft only; human codes |
| How does an experience feel? | Emotion | In vivo | Draft only; separate stated vs inferred |
| Which answers address each interview question? | Structural | Descriptive | Reliable for sorting, with spot checks |
| Does the data support an existing framework? | Provisional (deductive) | Magnitude | Useful with a clear codebook |
A five-step first-cycle workflow
- Write the research question at the top of your codebook and pick methods from the selector.
- Code two or three transcripts by hand first, ideally with a second coder, to calibrate code wording and boundaries.
- Draft code definitions, with an inclusion rule, an exclusion rule and one example per code.
- Run the rest of the dataset, by hand or with AI suggestions, and keep a memo of new codes and changed definitions.
- Stop and review before second-cycle coding. Merge duplicates, split overloaded codes and check agreement (see our intercoder reliability guide).
Can AI do first-cycle coding?
AI can produce a useful first-cycle draft, especially for descriptive and structural codes, but it should not replace researcher judgment on interpretive codes. Published results are mixed and depend heavily on the task.
Dunivin (2024, arXiv:2401.15170) found that GPT-4 with chain-of-thought prompting matched human coders on some hermeneutic codes but not others. A preprint by Kim and Mitchell (September 10, 2026, arXiv:2609.11109) tested multi-agent LLM coders that debate and reconcile disagreements. Accuracy depended on codebook length, how similar the data items were and how much the agents disagreed. The models also lacked adaptive responsiveness to context.
Our reading: AI is a good assistant for breadth (covering every transcript consistently with descriptive or structural codes) and a weak substitute for depth (values, emotion and initial codes that depend on context and theory). The selector above reflects that split.
Where Qualitati fits
Qualitati is a European, budget-friendly qualitative research platform for universities and research companies, with GDPR and data privacy as central priorities. For first-cycle coding, its QDA Workspace supports both routes:
- Manual coding with a code tree, multiple codes on the same passage and code descriptions.
- AI-assisted inductive and deductive coding, where the AI suggests codes for you to accept, edit or reject. Interviewer questions are excluded from AI coding, so only participant talk is coded.
- Codebook generation from your research questions, as a starting set of provisional codes.
When you move to second-cycle work, ThemeLens maps codes to research questions and synthesizes themes with participant-anchored quotes across up to 100 transcripts. Because Qualitati also runs AI-moderated interviews and conversational surveys in 10 languages, collection and coding can happen in one place. See transparent pricing or compare with NVivo.
Limitations and methodology concerns
- Method labels are not a quality guarantee. Saying "we used values coding" means little without code definitions and examples in an appendix.
- Too many methods dilute focus. Two methods per study is usually enough; more makes the codebook hard to apply consistently.
- Inferred codes need a second reader. Emotion and values codes are researcher interpretations. Flag them, and check them with a peer or participant where your design allows.
- AI suggestions anchor coders. Seeing a suggested code first can narrow what you notice. Hand-code a calibration set before you look at AI output.
- Reflexive thematic analysis works differently. Braun and Clarke treat coding as an interpretive, researcher-led practice, not a reliability exercise, so a cycle-based method menu may not fit that paradigm.
Methodology note: this guide summarizes published sources for planning purposes. Your supervisor, ethics board or journal may require specific methods; confirm with them before you code.
Who this is for, and when not to use it
Who this is for: master's and PhD students writing a methods chapter, research labs standardizing a coding procedure, and research firms briefing analysts on a shared codebook.
When not to use this approach: purely quantitative content analysis with a fixed scheme (use protocol coding and reliability statistics instead), or very small reflexive projects where a formal cycle structure adds paperwork without adding insight.
FAQ
What is the difference between first-cycle and second-cycle coding?
First-cycle coding assigns initial labels to segments of raw data. Second-cycle coding works on those codes, grouping, ranking or connecting them into categories, themes or theory. Pattern, focused, axial and theoretical coding are common second-cycle methods.
Which first-cycle coding method should I use?
Match the method to your research question. Use descriptive coding for "what topics", process coding for "what happens", values coding for "what matters", emotion coding for "how it feels", and in vivo coding when participants' own terms are central.
Can I use more than one first-cycle coding method?
Yes. Combining methods is common, for example descriptive plus in vivo, or process plus initial. Report each method, explain why it fits the research question, and keep separate definitions for each code type.
Is descriptive coding the same as thematic analysis?
No. Descriptive coding labels the topic of a passage. Thematic analysis builds broader patterns of shared meaning across many codes and participants. Descriptive codes can feed a thematic analysis but are not themes on their own.
Can AI do first-cycle coding for me?
AI can draft descriptive and structural codes across many transcripts quickly, but published studies show agreement with human coders varies by task. Review every suggestion, and code interpretive methods such as values and emotion yourself.
What book defines first-cycle coding methods?
Johnny Saldaña's The Coding Manual for Qualitative Researchers (SAGE). The fifth edition (2025) covers more than 35 methods and adds material on CAQDAS and AI tools.
Bottom line
First-cycle coding is where your analysis earns its credibility. Pick one or two methods that fit your research question, define your codes before you scale, and keep a human in charge of anything interpretive. If you want to try first cycle coding with AI suggestions you can accept or reject, start free with 30 credits on Qualitati, or read our guide to codes, categories and themes for the next step.