{"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":"2026-05-18T13:32:43.555Z"},"content":[{"type":"documentation","id":"098e995c-dd7a-4f20-be0b-1a244bcc428a","slug":"understanding-themes-patterns","title":"Understanding Themes & Patterns","url":"https://www.koji.so/docs/understanding-themes-patterns","summary":"Koji automatically identifies 3-7 themes per interview using AI-powered thematic analysis. Themes are aggregated across interviews with frequency data and traceable citations linking back to source interviews. Sentiment is tracked per-interview, not per-theme. Theme names are AI-generated and cannot be manually created or edited.","content":"Themes are the backbone of qualitative research. When multiple participants independently raise the same issue, describe the same experience, or express the same need, that's a pattern worth paying attention to. Koji automatically identifies these recurring themes across all interviews in your study, saving you hours of manual coding and analysis.\n\n## How Koji Identifies Themes\n\nAfter each interview completes, Koji's AI automatically analyzes the conversation and tags it with 3 to 7 relevant themes. As more interviews are completed, the system identifies which themes appear repeatedly and tracks their frequency and supporting evidence.\n\nThis process mirrors what qualitative researchers call **thematic analysis** — a well-established research method where you systematically identify, organize, and interpret patterns in qualitative data. The difference is that Koji does the initial coding pass automatically, giving you a head start on synthesis.\n\nUnlike simple keyword matching, Koji's theme detection understands context. Two participants might use completely different words to describe the same underlying issue. For example, one might say \"the setup process was confusing\" while another says \"I couldn't figure out how to get started.\" Koji recognizes both as expressions of the same theme: onboarding difficulty.\n\n## What Theme Data Looks Like\n\nFor each identified theme across your study, you'll see:\n\n### Theme Name and Description\n\nA clear, concise label that captures what the theme is about, along with a brief description of what it encompasses. These labels are AI-generated and designed to be immediately understandable, even to someone who hasn't read the transcripts.\n\n### Frequency with Citations\n\nHow many interviews mentioned this theme, with traceable citations linking back to the source interviews. This is one of your strongest signals. A theme that appears in 8 out of 10 interviews carries far more weight than one that appeared once.\n\nTheme frequency helps you prioritize. In product research, for example, a usability issue mentioned by 70% of participants is almost certainly more impactful than one mentioned by 10%. Each frequency count is backed by specific citations — you can click through to the original interviews to verify context.\n\n### Supporting Quotes\n\nDirect quotes from participants that illustrate the theme. These aren't randomly selected — Koji picks the most vivid, specific, and representative quotes for each theme, with attribution to the source interview. Supporting quotes serve two purposes:\n\n- **Verification**: You can confirm that the theme accurately reflects what participants said\n- **Persuasion**: When presenting findings, real quotes from real users are far more compelling than abstract summaries\n\n## How Themes Are Aggregated\n\nKoji's theme aggregation works through a frequency-based approach with semantic citation matching:\n\n1. **Per-interview tagging**: Each completed interview is tagged with 3-7 themes based on the conversation content.\n2. **Cross-interview aggregation**: Themes are aggregated across all interviews, counting how frequently each theme appears.\n3. **Citation linking**: Each theme occurrence is linked back to the specific interview using keyword-overlap scoring, ensuring that citations are relevant and traceable.\n4. **Report integration**: When you [generate a research report](/docs/generating-research-reports), themes with their frequency data and citations form the core of the theme analysis section.\n\nThis approach ensures that themes reflect genuine patterns in your data rather than artifacts of a single interview.\n\n## Reading Theme Patterns\n\nThemes don't exist in isolation. The most valuable analysis comes from understanding how themes relate to each other:\n\n### Theme Clusters\n\nSome themes naturally group together. For example, \"difficulty finding features,\" \"unclear navigation labels,\" and \"too many clicks to complete a task\" might all be part of a larger usability cluster. When you see related themes appearing together, you're looking at a systemic issue rather than isolated complaints.\n\n### Theme Contradictions\n\nSometimes different participant groups express opposing views on the same topic. New users might find a feature confusing while power users love it. These contradictions are incredibly valuable because they reveal segmentation in your user base and suggest that a one-size-fits-all approach may not work.\n\n### Theme Evolution\n\nIf you're running ongoing research, themes can shift over time. A theme that dominated early interviews might fade as newer participants focus on different concerns. Tracking this evolution helps you stay current with user needs.\n\n## Using Themes for Decision-Making\n\nThemes become powerful when you connect them to action:\n\n### Product Prioritization\n\nMap themes to your product roadmap. If \"difficulty with onboarding\" is your most frequent theme, that's a clear signal to prioritize onboarding improvements. Themes give you evidence-based ammunition for prioritization discussions.\n\n### Stakeholder Communication\n\nThemes provide a natural structure for presenting research findings. Instead of sharing a wall of interview notes, you can present three to five key themes, each backed by frequency data and supporting quotes. This format is digestible for executives, designers, and engineers alike. [Research reports](/docs/generating-research-reports) present themes in this stakeholder-ready format automatically.\n\n### Hypothesis Validation\n\nIf you started your study with specific hypotheses — \"We think users struggle with our pricing page\" — themes let you validate or invalidate those assumptions with real data. The presence or absence of relevant themes tells you whether your hypothesis held up.\n\n### Identifying Opportunities\n\nThemes aren't always about problems. Positive themes — features people love, experiences that delight — are equally valuable. They tell you what to protect and amplify in your product, not just what to fix.\n\n## Tips & Best Practices\n\n- **Wait for saturation**: In qualitative research, \"saturation\" means you've heard enough to stop learning new things. If the same themes keep appearing in new interviews without any new themes emerging, you've likely reached saturation. Most studies reach this point between 8 and 15 interviews.\n\n- **Don't over-index on frequency alone**: A theme mentioned by 9 out of 10 participants is clearly important. But a theme mentioned by only 2 out of 10 might be equally valuable if those two participants represent a key user segment or if the theme reveals a critical edge case.\n\n- **Cross-reference with interview sentiment**: While themes themselves don't carry individual sentiment labels, you can cross-reference themes with the overall sentiment of the interviews where they appeared. A theme that shows up primarily in negatively-sentiment interviews signals a pain point, while one appearing in positive interviews highlights a strength.\n\n- **Look beyond your brief**: Sometimes the most interesting themes are ones you didn't ask about. Participants may raise topics outside your original research questions that turn out to be critically important.\n\n- **Trace back to transcripts**: When a theme feels important, go back to the source. Read the relevant sections of the transcripts to understand the full context. Themes are summaries — transcripts are the evidence.\n\n## Key Things to Know\n\n- **Themes update as interviews arrive**: Each new interview adds data to the theme analysis. Themes may shift in frequency and new themes may emerge as your sample grows.\n- **Theme names are AI-generated**: The labels are designed to be descriptive and clear. They cannot be manually renamed or edited — this preserves consistency and objectivity across the analysis.\n- **Themes are per-interview, not editable**: Koji generates 3-7 theme tags per interview automatically. You cannot manually create, rename, or delete themes.\n- **Themes feed into reports**: When you [generate a research report](/docs/generating-research-reports), the report's theme section is built from this same underlying analysis, presented in a stakeholder-ready format with traceable citations.\n- **No per-theme sentiment**: Sentiment is tracked at the interview level (positive, negative, neutral, mixed), not per individual theme. To understand sentiment around a theme, look at the sentiment of interviews where that theme appears.\n\n## Related Articles\n\n- [AI-Generated Insights](/docs/ai-generated-insights) — Per-interview themes that feed into cross-interview patterns\n- [Generating Research Reports](/docs/generating-research-reports) — Aggregate theme analysis in a shareable report format\n- [Structured Questions Guide](/docs/structured-questions-guide) — How structured questions complement thematic analysis\n- [Understanding Quality Scores](/docs/understanding-quality-scores) — How quality scores affect which interviews contribute to analysis\n\n## Frequently Asked Questions\n\n**Q: How many themes does Koji typically identify per study?**\nA: It depends on the breadth of your study and the number of interviews. Each interview is tagged with 3-7 themes. A focused study might surface 5-8 major cross-interview themes, while a broader exploratory study could produce 15-20.\n\n**Q: Can I create or rename themes manually?**\nA: No. Theme identification is fully automated to ensure consistency and objectivity. The AI-generated theme names are designed to be descriptive and immediately understandable. Manual theme creation or editing is not supported.\n\n**Q: How many interviews do I need before themes are reliable?**\nA: You'll start seeing theme patterns after 3-4 interviews, but reliability increases significantly with 6-8 or more. The more interviews contribute to a theme, the more confident you can be that it represents a genuine pattern.\n\n**Q: Are themes weighted by quality score?**\nA: Reports filter to interviews scoring 3 or above, so themes in reports only reflect qualifying interviews. The per-interview theme tags are generated for all completed interviews regardless of score.\n\n**Q: Can I compare themes across different studies?**\nA: Themes are generated per study. To compare themes across studies, you can review the reports from each study side by side and look for overlapping patterns.\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- [Best AI Thematic Analysis Tools in 2026: The Complete Buyer's Guide](/blog/best-ai-thematic-analysis-tools-2026) — A side-by-side review of the top AI thematic analysis platforms in 2026 — what each does well, where they fall short, and why AI-native rese\n\n<!-- further-reading:blog -->\n","category":"Reports & Analysis","lastModified":"2026-05-17T03:21:34.303214+00:00","metaTitle":"Themes & Patterns — Koji Docs","metaDescription":"Understand how Koji identifies recurring themes across interviews. Learn to read theme frequency, supporting quotes, and use patterns for decisions.","keywords":["research themes","thematic analysis","qualitative patterns","theme frequency","cross-interview analysis","research synthesis","Koji themes"],"aiSummary":"Koji automatically identifies 3-7 themes per interview using AI-powered thematic analysis. Themes are aggregated across interviews with frequency data and traceable citations linking back to source interviews. Sentiment is tracked per-interview, not per-theme. Theme names are AI-generated and cannot be manually created or edited.","aiPrerequisites":["ai-generated-insights","creating-your-first-study"],"aiLearningOutcomes":["Understand how Koji detects and labels themes across interviews","Read theme frequency, sentiment, and supporting quotes","Use theme patterns for product prioritization and stakeholder communication","Know when theme saturation has been reached"],"aiDifficulty":"intermediate","aiEstimatedTime":"8 min read"}],"pagination":{"total":1,"returned":1,"offset":0}}