{"site":{"name":"Koji","description":"AI-native customer research platform that helps teams conduct, analyze, and synthesize customer interviews at scale.","url":"https://www.koji.so","contentTypes":["blog","documentation"],"lastUpdated":"2026-11-16T00:28:33.476Z"},"content":[{"type":"documentation","id":"7f61898f-32a2-41ae-9ea6-10f1186ba82d","slug":"coding-qualitative-data","title":"How to Code Qualitative Data: Your First Coding Pass on 10 Interviews","url":"https://www.koji.so/docs/coding-qualitative-data","summary":"A beginner's path to coding qualitative data: one complete coding pass on ten interview transcripts. Read three transcripts, code them with short meaning-based codes and one-line definitions, code the other seven against that list, clean it to 10 to 20 codes, group them into 3 to 6 themes written as sentences, and check each theme against the data. Cites saturation research (Guest, Bunce and Johnson 2006; Hennink, Kaiser and Marconi 2017; Hennink and Kaiser 2022) on why ten interviews is a good first set, and points to first and second cycle coding for the full method. Koji runs AI-moderated interviews and gives a first coding pass with verbatim quotes on every answer.","content":"**Short answer:** To code qualitative data for the first time, take 10 interview transcripts, read three of them start to finish, then label every passage that answers your research question with a short code (a few words that say what the passage means). Write each code's definition the moment you create it, code the remaining seven interviews against that list while adding new codes as needed, then merge, split and drop codes until 10 to 20 remain, and group those into 3 to 6 themes. That is one complete coding pass. It takes a weekend by hand, and it is enough to answer most product and customer research questions.\n\nThis page is the short beginner path: one pass, ten interviews, step by step. If you want the full method behind it, with Saldaña's first and second cycle coding methods and when to use each, read [First and Second Cycle Coding](/docs/first-and-second-cycle-coding). If you are building a codebook for a team, read [How to Build a Qualitative Research Codebook](/docs/qualitative-research-codebook).\n\n## What a code is\n\nA code is a short label you attach to a passage of text so you can find, compare and count similar passages later. Johnny Saldaña, whose _Coding Manual for Qualitative Researchers_ is the standard reference, defines it like this:\n\n> \"A code in qualitative inquiry is most often a word or short phrase that symbolically assigns a summative, salient, essence-capturing, and/or evocative attribute for a portion of language-based or visual data.\" (Johnny Saldaña, _The Coding Manual for Qualitative Researchers_)\n\nIn practice, a code names one of four things:\n\n- **A problem** someone describes (\"rebuilds report every month\")\n- **A behaviour** (\"checks the dashboard before standup\")\n- **A reason or belief** (\"doesn't trust the numbers\")\n- **A feeling** (\"dreads month end\")\n\nSaldaña is also clear about what coding is not: \"Coding is not a precise science; it's primarily an interpretive act.\" Two careful people will code the same transcript a little differently. That is fine, as long as your codes are defined and you can show which passages sit behind each one.\n\n## Why 10 interviews is a good first set\n\nTen is small enough to code by hand and large enough to see patterns. The research on saturation, the point where new interviews stop adding new codes, backs this up:\n\n- In a study of 60 interviews, Guest, Bunce and Johnson (2006) found that saturation occurred within the first twelve interviews, and the basic elements of the main themes were already present at six.\n- Hennink, Kaiser and Marconi (2017) found code saturation, hearing the full range of issues, at nine interviews. Understanding what each issue meant to people took longer: 16 to 24 interviews.\n- A systematic review by Hennink and Kaiser (2022) found that across 16 empirical tests, the interviews needed for saturation ranged from 5 to 24.\n\nSo ten interviews from one fairly similar group will show you most of the topics. They won't always explain them in depth. Treat your first pass as a map, and plan more interviews where the map is thin. For the full evidence, read [Interview Sample Size Research](/docs/how-many-interviews-enough).\n\n## Before you start: three decisions\n\n1. **Write down your research question** in one sentence. \"Why do trial users stop using the product in week two?\" Every coding decision is easier when you can ask \"does this passage help answer that?\"\n2. **Choose inductive, deductive or both.** Inductive means codes come from the data. Deductive means you start with a list (from your research question, a framework or a past study) and apply it. Most first studies use both: three to five starting codes from the question, then new codes as you find them. See [Inductive vs Deductive Research](/docs/inductive-vs-deductive-research).\n3. **Pick a tool you already know.** A spreadsheet with columns for participant, passage, code and note works well for ten interviews. You don't need dedicated software for a first pass.\n\n## Your first coding pass, step by step\n\n![Process showing seven steps of a first coding pass on 10 interviews, from reading three transcripts to grouping codes into themes and checking them.](/visuals/first-qualitative-coding-pass-10-interviews/first-qualitative-coding-pass-10-interviews.png \"One pass, start to finish, for your first ten interviews\")\n\n### Step 1: Get clean, verbatim transcripts\n\nCode from transcripts, not from notes. Notes are already your summary, so coding them codes your own words back to you. Give each transcript a participant ID (P01 to P10) and keep the date and any segment you care about (plan, role, company size) at the top.\n\n### Step 2: Read three transcripts without coding\n\nRead the first three interviews straight through. Don't label anything yet. Jot down a few lines after each one: what stood out, what surprised you, what you want to check in the others. These notes are your first memos. See [Analytic Memos](/docs/analytic-memos-qualitative-research) for what to write.\n\n### Step 3: Code those three, passage by passage\n\nGo back to the start of the first transcript. For each passage that relates to your research question, write a short code next to it. Rules that keep a first pass clean:\n\n- **Code the meaning, not the topic.** \"Pricing\" is a topic. \"Unsure what the paid plan adds\" is a code you can act on.\n- **Keep codes short.** Two to five words.\n- **One passage can carry two or three codes.** People often say several things at once.\n- **Use the participant's own phrase when it says it better.** \"Human copy-paste machine\" is a code worth keeping as is. That is [in vivo coding](/docs/in-vivo-coding).\n- **Write a one-line definition the first time you use a code.** \"Unsure what paid plan adds: the participant can't name a benefit of upgrading.\"\n\nExpect 30 to 60 codes after three interviews. That is normal.\n\n### Step 4: Code the other seven against your list\n\nNow code interviews four to ten. Before creating a new code, check whether one on your list fits. If it does, use it. If it nearly fits, either widen the definition (and recheck the passages you already coded) or create a new code. Keep a tally of how many new codes each interview adds. If interviews eight, nine and ten add almost nothing, your first pass has found most of what this group has to say.\n\n### Step 5: Clean up the list\n\nWith all ten coded, sort your codes and tidy them:\n\n- **Merge** codes that name the same thing in different words.\n- **Split** codes that hide two ideas.\n- **Drop or park** codes that appear once and don't bear on the question. Park rather than delete when one person said something important. A single strong signal can matter.\n- **Rewrite definitions** so someone else could apply them.\n\nAim for 10 to 20 codes. This is your first codebook.\n\n### Step 6: Group codes into themes\n\nA theme is a pattern that answers part of your research question. It groups several codes into one sentence you could say to your team. Codes \"rebuilds report monthly\", \"pulls from four tools\" and \"fixes dates by hand\" might become the theme \"Month-end reporting is manual work people fit around their real job.\"\n\nWrite each theme as a full sentence, not a label. \"Reporting\" tells nobody anything. Three to six themes is typical for ten interviews.\n\n### Step 7: Check every theme against the data\n\nFor each theme, count how many of the ten participants it applies to and pick two or three quotes that show it. Then look for passages that contradict it. A theme that holds up in seven of ten interviews, with a clear reason why the other three differ, is a finding. A theme that rests on one vivid quote is a hunch. Write down both kinds, and say which is which.\n\n## A short worked example\n\nA founder interviews ten people who signed up for a trial and stopped using it. One participant, P04, says:\n\n> \"I set it up on Monday, then I had to ask IT for SSO and by the time that came through I'd already gone back to the spreadsheet.\"\n\n| Passage | Code |\n| --- | --- |\n| \"I had to ask IT for SSO\" | Blocked by IT access |\n| \"by the time that came through\" | Setup outlasts motivation |\n| \"gone back to the spreadsheet\" | Falls back to old tool |\n\nAfter coding all ten, \"Blocked by IT access\" shows up in six interviews and \"Falls back to old tool\" in seven. Together with two related codes they become the theme: \"Trials stall when setup needs someone else's approval, and the old tool fills the gap while people wait.\" That is specific enough to act on.\n\n## Common first-pass mistakes\n\n- **Coding from memory or notes** instead of transcripts.\n- **Codes that only name a topic** (\"onboarding\", \"price\"). You can't act on a topic.\n- **No definitions.** After three interviews you will no longer remember what \"friction\" meant.\n- **Stopping at a code list.** Codes are raw material. Themes are what you report.\n- **Only coding what confirms your idea.** Look for the passages that disagree with your favourite theme.\n\n## When two people code\n\nIf two people share the coding, have both code the same two interviews on their own, then compare. Talk through every disagreement and fix the definition that caused it. If you want a number, a common rule of thumb from Miles and Huberman (1994), as summarised by O'Connor and Joffe (2020), is 80% agreement on 95% of codes. O'Connor and Joffe also note there is no universally accepted threshold, so the conversation about disagreements matters more than the score. See [Inter-Rater Reliability](/docs/inter-rater-reliability-qualitative-research).\n\n![Two colleagues at a table comparing their notes on the same interview to agree on what each code means](https://images.unsplash.com/photo-1758876203420-1ed6db481d4f?crop=entropy&cs=tinysrgb&fit=max&fm=jpg&ixid=M3w5NDUzMDN8MHwxfHNlYXJjaHw3fHx0d28lMjBjb2xsZWFndWVzJTIwcmV2aWV3aW5nJTIwbm90ZXMlMjB0YWJsZXxlbnwxfDB8fHwxNzk0MjkzNDEwfDA&ixlib=rb-4.1.0&q=80&w=1080)\n\n_Photo by [Vitaly Gariev](https://unsplash.com/@silverkblack?utm_source=koji&utm_medium=referral) on [Unsplash](https://unsplash.com/?utm_source=koji&utm_medium=referral)_\n\n## How Koji helps\n\nCoding by hand is worth doing once, because it teaches you what good codes look like. Koji takes over the slow parts when you run research often:\n\n- **Transcripts you can code from.** Koji runs AI-moderated interviews by voice or text and asks follow-up questions on open answers, so you get full answers in the participant's words instead of a short line in a survey box.\n- **A first coding pass on every answer.** After each interview, Koji codes each answer with short labels about what it means, and attaches the participant's exact words to each code. Koji checks each quote against the transcript and drops a code it can't ground in what the person said. Quotes stay in the participant's own language.\n- **Codes grouped across interviews.** In the report, Koji groups similar codes from different interviews into themes for each question and shows how many interviews mention each one. With only a few interviews, the report says the themes may still change, much like the saturation research above.\n- **The evidence behind every finding.** Open any finding to see the quotes behind it and go back to the transcript, which makes step 7 faster.\n- **Numbers where you need them.** Add scale, choice, ranking or yes/no questions to the same interview, and Koji still codes the reasons people give.\n\n![Koji report summary with a side panel of supporting participant quotes behind one finding](/email/shots/shot-evidence-panel.jpg \"Every finding opens to the participants' own words behind it, so you can check a theme the way step 7 asks\")\n\nTreat Koji's codes and themes the way you would treat a colleague's first pass: read the quotes, rename what doesn't fit your question, and decide which themes matter. The interpretation is still yours.\n\n## Where to go next\n\n- For the full method, read [First and Second Cycle Coding](/docs/first-and-second-cycle-coding).\n- To turn your code list into a shared codebook, read [How to Build a Qualitative Research Codebook](/docs/qualitative-research-codebook).\n- To analyse your themes across segments in a matrix, read [Framework Analysis](/docs/framework-analysis-qualitative).\n\n## Related Resources\n\n- [First and Second Cycle Coding](/docs/first-and-second-cycle-coding): Saldaña's coding methods and how codes become themes.\n- [How to Build a Qualitative Research Codebook](/docs/qualitative-research-codebook)\n- [Framework Analysis](/docs/framework-analysis-qualitative)\n- [In Vivo Coding](/docs/in-vivo-coding): using participants' own words as codes.\n- [Thematic Analysis Guide](/docs/thematic-analysis-guide)\n- [Structured Questions Guide](/docs/structured-questions-guide)\n\n## FAQ\n\n**How long does a first coding pass take?**\nFor ten interviews of 30 to 45 minutes, plan on a weekend. Reading and coding the first three takes longest, because that is where you build the code list. The last seven go faster.\n\n**How many codes should I end up with?**\nExpect 30 to 60 codes after the first three interviews and 10 to 20 after you clean up. Those then group into three to six themes for a ten-interview study.\n\n**Should I code inductively or deductively?**\nFor a first study, both. Start with three to five codes from your research question, then add codes from the data as you find them.\n\n**Do I need software like NVivo or ATLAS.ti?**\nNot for ten interviews. A spreadsheet with participant, passage, code and note columns is enough. Dedicated tools help once you have many studies or several coders.\n\n**What is the difference between a code and a theme?**\nA code labels one passage. A theme is a pattern across many passages and participants that answers part of your research question, written as a full sentence.\n\n## Sources\n\n- Saldaña, J. _The Coding Manual for Qualitative Researchers_. SAGE.\n- Guest, G., Bunce, A., & Johnson, L. (2006). How many interviews are enough? An experiment with data saturation and variability. _Field Methods_, 18(1), 59–82.\n- Hennink, M. M., Kaiser, B. N., & Marconi, V. C. (2017). Code saturation versus meaning saturation: How many interviews are enough? _Qualitative Health Research_, 27(4), 591–608.\n- Hennink, M., & Kaiser, B. N. (2022). Sample sizes for saturation in qualitative research: A systematic review of empirical tests. _Social Science & Medicine_, 292, 114523.\n- O'Connor, C., & Joffe, H. (2020). Intercoder reliability in qualitative research: Debates and practical guidelines. _International Journal of Qualitative Methods_, 19.\n\n## Further reading on the blog\n\n- [How to Analyze Customer Interview Data: A Complete Guide](/blog/how-to-analyze-customer-interview-data) — You ran the interviews. Now what? Here is a step-by-step process for turning raw transcripts into clear, actionable insights your team will \n- [How to Analyze User Interview Data: A Complete Guide (2026)](/blog/how-to-analyze-user-interview-data) — You ran the interviews. Now what? This step-by-step guide covers how to turn raw interview data into clear, actionable insights — with and w\n- [How to Conduct Remote User Interviews: The Complete Guide (2026)](/blog/how-to-conduct-remote-user-interviews-2026) — Remote user interviews are now the default for most research teams. This guide covers everything — from recruiting and scheduling to running\n\n<!-- further-reading:blog -->\n","category":"Analysis & Synthesis","lastModified":"2026-11-13T05:57:11.229647+00:00","metaTitle":"How to Code Qualitative Data: A First Pass on 10 Interviews","metaDescription":"Code qualitative data for the first time: a step-by-step first pass on 10 interviews, with a worked example, the common mistakes and where to go next.","keywords":["how to code qualitative data","qualitative coding guide","thematic coding","qualitative data analysis","codebook research","inductive coding","deductive coding","research synthesis"],"aiSummary":"A beginner's path to coding qualitative data: one complete coding pass on ten interview transcripts. Read three transcripts, code them with short meaning-based codes and one-line definitions, code the other seven against that list, clean it to 10 to 20 codes, group them into 3 to 6 themes written as sentences, and check each theme against the data. Cites saturation research (Guest, Bunce and Johnson 2006; Hennink, Kaiser and Marconi 2017; Hennink and Kaiser 2022) on why ten interviews is a good first set, and points to first and second cycle coding for the full method. Koji runs AI-moderated interviews and gives a first coding pass with verbatim quotes on every answer.","aiPrerequisites":["user-interview-guide","thematic-analysis-guide"],"aiLearningOutcomes":["Build a qualitative codebook with clear definitions","Apply inductive and deductive coding to interview transcripts","Identify themes from coded data","Assess interrater reliability for team research","Present coded findings to stakeholders effectively"],"aiDifficulty":"intermediate","aiEstimatedTime":"9 min read"}],"pagination":{"total":1,"returned":1,"offset":0}}