Find themes across interview transcripts

The problem

You've got transcripts from 15 interviews. Reading through everything to identify themes and pull out quotes would take a week. Your evaluation report is due, and you need to find patterns across what people said.

The solution

Upload your transcripts to NotebookLM or Claude and ask it to identify themes. The AI reads all the interviews, spots recurring topics, and pulls out representative quotes. You can then ask specific questions about what participants said, and it cites which interview the answer comes from.

What you get

A structured analysis: key themes with supporting quotes and participant references, answers to your specific evaluation questions, and a summary of findings. You can drill into individual transcripts for more detail or verify that quotes are accurate.

Before you start

• Interview transcripts in text format (from manual transcription or a service like Otter.ai)
• Consent from interviewees that covers AI analysis of their responses
• Your evaluation questions or research framework
• NotebookLM (free) or Claude Pro (£18/month - required for Claude Projects feature) or ChatGPT Plus

When to use this

• You've got 5+ interviews and need to find patterns across them
• You have specific evaluation questions you need to answer
• You want to find representative quotes for your report
• Manual thematic analysis would take longer than you have

When not to use this

• You haven't got consent to use AI tools for analysis
• The content is highly sensitive and shouldn't be uploaded to third-party services
• You need formal qualitative research methodology (grounded theory, etc.) that requires human interpretation
• You only have 2-3 short interviews - might be quicker to just read them
• You haven't checked whether the AI provider uses your data for training - use NotebookLM (doesn't train on data), Claude Pro (doesn't train), or ChatGPT with 'Temporary Chat' enabled

Steps

1. 1

   Check consent covers AI analysis

   Review your consent forms or information sheets. Did you tell participants their responses might be analysed using AI tools? If not, you may need additional consent, or you'll need to anonymise thoroughly before uploading. Check with your ethics lead if unsure.

2. 2

   Anonymise the transcripts

   Replace real names with pseudonyms or participant numbers (P01, P02, etc.). Remove identifying details like workplace names or specific locations if they're not needed for analysis. This protects participants and makes you more comfortable using AI tools.

3. 3

   Upload to NotebookLM or Claude Projects

   NotebookLM is ideal: upload all transcripts as sources and it creates a research assistant that can cite which interview answers come from. Claude Projects works similarly. For ChatGPT, you'll need to paste transcripts into the conversation, which is clunkier for many interviews.

4. 4

   Ask for themes across all interviews

   Start broad: 'What are the main themes that come up across these interviews? For each theme, give me 2-3 supporting quotes with participant IDs.' The AI will identify patterns and provide evidence. Check whether the themes make sense given what you know.

5. 5

   Ask about your specific evaluation questions

   Now get specific: 'What do participants say about [topic]?' or 'How do participants describe their experience of [service]?' The AI pulls relevant sections from across the interviews. This is much faster than reading through everything looking for mentions.

6. 6

   Verify quotes in the original transcripts

   Before using any quote in a report, go back to the original transcript to check it's accurate and in context. WARNING: AI can hallucinate - it may paraphrase, merge quotes from different participants, or generate plausible-sounding quotes that weren't actually said. This is a known limitation of LLMs. Never include a quote in a report without verifying it appears in the original transcript.

7. 7

   Generate structured outputs(optional)

   Ask for outputs in your report format: 'Create a summary of findings organised by our three evaluation questions', 'Give me a table of themes with supporting evidence', 'Write a paragraph about barriers to access with quotes'. The AI adapts to your structure.

Tools

Resources