Before you use any prompt
- Protect participants. Remove names, places and other identifying details before pasting any transcript into an AI tool, and check your ethics approval and consent form allow it.
- Use AI as an assistant. It can suggest codes and spot patterns, but you must read the data, confirm every code against the text and own the final themes.
- Check quotes. AI tools sometimes paraphrase or invent text. Always match a quote to your transcript word for word.
- Be open about it. If your institution or journal asks how AI was used, describe it in your methods.
1. First-pass coding
When to use it: to get candidate codes from a few anonymised excerpts so you can compare them with your own.
You are helping with qualitative analysis. Read the interview excerpts below, which are anonymised. Suggest 8 to 12 descriptive codes. For each code give: a short name, a one-line definition and one exact quote from the text that fits it. Do not interpret beyond what participants said. Excerpts: [paste text]
Then: compare with your own codes. Keep the ones you can defend and ignore the rest.
2. Build a codebook
Here is my list of codes with example quotes: [paste]. Turn them into a codebook table with columns: code name, definition, include when, exclude when, example quote. Point out codes that overlap and suggest how to merge or separate them.
3. Group codes into candidate themes
Here are my codes and how often each appears: [paste]. Suggest 4 to 6 candidate themes that group related codes. For each theme give a one-sentence description and list the codes it contains. Then list any codes that do not fit and may be a separate theme or noise.
Treat the output as a starting map. Themes should tell a story about your research question, not just sort topics. Our step-by-step thematic analysis guide explains how to review and name themes properly.
4. Challenge your themes
My themes are: [paste names and definitions]. Act as a sceptical reviewer. For each theme say whether it is too broad, too thin or overlaps with another, and suggest what evidence in the data would support or weaken it. Do not invent data.
5. Look for negative cases
Theme: [name and definition]. Below are excerpts from participants. Identify any excerpts that contradict or complicate this theme and explain why. Quote only text that appears below. Excerpts: [paste]
6. Compare groups
Below are anonymised excerpts from two groups, A and B. Describe differences and similarities in how each group talks about [topic]. For each point, give the supporting quote and say which group it came from.
7. Write a theme description
Using the definition and quotes below, draft a 150-word results paragraph for the theme [name]. Start with the claim, then show it with two short quotes, then say what it means for the research question. Use only the quotes provided. Definition and quotes: [paste]
8. Summarise a long transcript
Summarise this anonymised transcript in 200 words. List the five most important points participants made, in their own terms, and note anything that surprised you. Do not add information that is not in the text.
9. Check coding consistency
Here is my codebook and 10 excerpts that I coded: [paste]. Code the excerpts yourself using only the codebook, then show where your codes differ from mine. I will use the differences to refine the code definitions.
This is not a replacement for a second human coder, but it helps you find unclear definitions.
10. Draft a methods statement
I used reflexive thematic analysis following Braun and Clarke. Steps: [list your steps]. Write a clear methods paragraph of about 150 words. Do not claim steps I did not list.
Get better results from any prompt
- State your research question and who the participants are.
- Ask for exact quotes and tell the tool to use only the text you supplied.
- Work in small batches rather than pasting a whole dataset.
- Ask for a table with fixed columns so you can paste it into your codebook.
- Run the same prompt twice. If the answers differ a lot, the pattern is weak.
What AI cannot do for you
It cannot decide what matters for your research question, judge the context in which something was said or take responsibility for your claims. If you want an experienced analyst to code your data or review your themes, see our thematic analysis service or NVivo analysis.
Frequently asked questions
Can I use AI to analyse my interviews?
You can use it to support coding and summarising if your ethics approval and data protection rules allow it, but you must check every output against the original text and make the final decisions yourself.
Is it safe to paste transcripts into an AI tool?
Only if they are anonymised and your consent form, ethics approval and the tool’s data terms allow it. When unsure, do not paste sensitive data.
Will AI-generated themes be accepted in a thesis?
Themes must be supported by your data and justified in your methods. Examiners care about your reasoning, not the tool, so show how you developed and checked each theme.
Which prompt should I start with?
Start with the first-pass coding prompt on a small anonymised sample, then compare the suggestions with your own reading.
Related guides and services
Want an expert to check or run your analysis?
Send us your question and data description. We reply with a clear plan and a quote.
Discuss your project on WhatsApp