Coding is the first real analytical step in qualitative research: the process of attaching short labels to meaningful pieces of text, audio, or video so that patterns across a dataset become visible. Before coding, a dataset is just a pile of transcripts and notes. After coding, it becomes something you can query, compare, and build an argument from.
This guide covers what a code actually is, how coding differs from simply summarizing, and the two broad approaches researchers use to build a coding system. It’s written for anyone starting their first qualitative project, whether that’s a thesis, a UX research study, or an academic paper.
What is a code?
A code is a short word or phrase that captures the essence of a segment of data. If a participant says, “I stopped trusting the app after it changed my settings without asking,” a researcher might code that segment loss of trust or unexpected system behavior, depending on what the study is actually about.
A good code does two things at once: it stays close enough to the data that you can defend why you chose it, and it’s abstract enough to apply to other segments that express the same idea in different words. That second part is what makes coding useful — it’s how ten different participants describing ten different moments end up grouped under one recognizable pattern.
Coding is not summarizing
A common mistake for new researchers is treating codes like mini-summaries — writing “participant discusses onboarding” instead of naming what’s actually happening in that discussion. A summary describes; a code interprets. The difference matters because summaries don’t stack into patterns the way codes do. Twenty summaries are still twenty separate sentences. Twenty segments coded onboarding confusion are a finding.
Two ways to build a coding system
Researchers generally start from one of two directions, and most real projects end up blending both.
Deductive coding (top-down)
You start with a set of codes drawn from theory, prior research, or your interview guide, then apply them to the data. This works well when a study is grounded in an existing framework and the research questions are already fairly specific. The risk is missing something the framework didn’t anticipate.
Inductive coding (bottom-up)
You start with no predefined codes and let them emerge directly from the data as you read through it. This is closer to how grounded theory works (Glaser & Strauss, 1967) and tends to surface things a top-down framework would have filtered out. The tradeoff is that it takes longer and produces a messier first pass — codes get merged, split, and renamed as patterns become clearer.
In practice, most qualitative projects use a hybrid: a small set of deductive codes tied to the research questions, refined and expanded inductively as the data reveals things those original questions didn’t account for. Johnny Saldaña’s The Coding Manual for Qualitative Researchers (2021) is the standard reference most methods courses point students to for the mechanics of both approaches, including first-cycle and second-cycle coding methods.
What good coding sets up
Coding by itself isn’t the finding — it’s the structure that makes the next steps possible: grouping codes into broader themes, checking how often certain codes co-occur, comparing how coding patterns differ across participants or documents, and eventually writing up results that are traceable back to real evidence rather than impression.
The next guide in this series covers exactly that: turning a coded dataset into organized themes ready for analysis.
Coding in Byleron QDA: codes stay visible alongside your source text as you work, and every code you apply is one click away from the original quote it came from — so when you’re ready to write up a theme, you’re never more than a click from the evidence behind it.