Start with your research question and chosen method. Qualitative approaches differ in how they treat interpretation, coding and the researcher's role. If you are using reflexive thematic analysis, do not turn a request for suggested codes into a claim that the software discovered objective themes or demonstrated agreement between independent human coders.
Read the interview before extracting a theme
Suppose you are studying how staff experience a shared task tracker. The following three excerpts are original fictional teaching material, not participant data or a ChatGPT transcript:
| Interview and passage | Unchanged excerpt | Possible analytic note |
|---|---|---|
| A, passage 4 | “I stopped sending reminder emails because I could see who had picked it up.” | Visibility appears to reduce the need to chase ownership. |
| B, passage 7 | “I still keep a paper list. The board says done before the handover is actually finished.” | A completed status does not necessarily establish a completed handover. |
| C, passage 3 | “The reminders save me time, but now everyone can see when I fall behind.” | Convenience and exposure coexist in this account. |
“The tracker makes work easier” is a tempting first interpretation. It loses B's disagreement about what “done” means and C's concern about visibility. A stronger question for further reading is whether shared visibility shifts both coordination work and pressure between people.
That is a provisional interpretation of three short invented passages. It does not establish a final theme, represent all staff, or reveal participants' unstated motives. In a real study, return to the surrounding interview, the question that elicited the answer, and other relevant passages before retaining the interpretation.
Keep codes, themes and claims distinct
A code can identify a feature of a passage, such as “checking ownership” or “status hides unfinished handover.” A theme needs a developed pattern of meaning that addresses the research question. “Task tracker” names a topic, but it does not explain what the interviews suggest about that topic.
Braun and Clarke's reflexive thematic analysis guidance describes familiarization, coding, developing and reviewing themes, refining them and writing the analysis. The process moves back and forth; reviewing a candidate theme means checking it against coded material and the wider dataset.
You can ask ChatGPT to challenge a candidate interpretation during that work. OpenAI's prompting documentation recommends providing context and clear task boundaries. Supply the method and question, so “analyze this” does not become an unexplained blend of topic counts, sentiment and invented explanation.
Use a prompt that preserves the source trail
Only supply material permitted by your research approval, participant consent and organizational rules. Removing names may leave identifiable stories. Keep full source records in their approved research location and use stable excerpt references in the working memo.
Help me question my interpretation of these interview excerpts.
Research question: [PASTE]
Method and analytic orientation: [ENTER]
Relevant study context and my analytic position: [ENTER]
Permitted excerpts with interview IDs and passage references: [PASTE]
My current codes or candidate interpretation: [PASTE]
Offer two plausible readings grounded in the supplied words. For each,
quote exact supporting passages with their references. Identify passages
that complicate the reading and questions to check in the full interviews.
Separate participant words, your proposed code, and an interpretive claim.
Do not invent quotations, context, motives, participants or prevalence.
Do not equate frequent words with a theme, declare saturation, or claim
this constitutes a completed analysis. Label suggestions provisional.
Ask me to decide what to retain and record my reason for that decision.
For the tracker exercise, useful follow-up questions include what B means by “handover,” who can see C's status, and whether A describes fewer reminders elsewhere in the interview. An excerpt may leave those questions open. The model should not fill them with a convenient story.
Write a memo that shows the decision
An authored practice memo might read: “The initial convenience reading misses how shared visibility changes accountability. A describes easier ownership checking; B questions the status's accuracy; C combines time saved with exposure. I will review full accounts of handovers and visibility before deciding whether these passages support one pattern or several.”
The memo gives a reason to return to the data. It does not convert three passages into a percentage of staff or promise that a new feature would solve the concern. Preserve the alternative reading and why you changed it; those decisions help you explain your analysis later.
If your actual task is counting predefined categories in survey answers, the survey-analysis guide describes codebook rules and respondent denominators. If you are labeling expressed favorable or unfavorable opinions, the sentiment-analysis guide addresses that different task.
Before writing the finding, read the relevant interviews again and check the quotations against the source. Keep the method, analytic memo and retained interpretation together, including evidence that limits or complicates what you can say.