Chapter 9. Collecting qualitative data — accounts, observation and discourse, with AI

Key points

• The main qualitative methods are interviews, participant observation and discourse analysis
• Choose structured or unstructured interviews by purpose: exploring or testing
• AI accelerates transcription, organising and analysis, dropping the barrier to qualitative research dramatically

Drawing meaning out of accounts

Qualitative data digs into the "why" that numbers cannot capture. Three main methods.

Method 1: interviews — drawing out values through conversation

An interview seeks to understand thinking, feeling and the meaning behind behaviour through what someone says. Two broad types.

Unstructured interviews. No detailed questions decided in advance; the person speaks freely. Open questions such as "tell me about something you have used recently and were glad of", "why do you think you chose it?" → Useful in the exploratory stage before a hypothesis exists, and when you want unexpected findings.

Structured interviews. Items decided in advance, and everyone asked the same questions. Yields data that compares and analyses easily. → Useful when testing a specific hypothesis, and when widening the sample.

Think of a therapist's history-taking. "What brings you in?" and letting the person speak is unstructured; working through fixed assessment items is structured. You already use both, every day.

Method 2: participant observation and ethnography — going into the field

The researcher actually enters the group or the setting and deepens understanding by being alongside.

• reaches beyond what people say to the unspoken rules, habits and values behind everyday behaviour
• observes what an interview cannot elicit: things too obvious to mention, and unconscious behaviour

Examples: observing customer movement in a shop to understand purchasing; watching how users interact at a day service to read what kind of engagement is actually wanted. It takes time and effort, but yields extremely deep insight. For a therapist with a clinician's eye for observation, this may be the method that comes most naturally.

Method 3: discourse analysis — reading how words are used

Analysing large volumes of online language — social posts, blogs, reviews — and reading the cultural and social background out of how people talk.

Example: analysing posts about returning to work after having a child, to extract contemporary values around combining childcare and career.

What words are used, and what is treated as important. From this you see how your target audience talks about your service or its category.

AI drops the barrier to qualitative research

As Chapter 7 noted, qualitative data is where AI contributes most.

• Transcription — turning interview recordings into text (take the greatest care with consent and privacy before any audio leaves your machine)
• Coding support — a first draft of shared themes and codes extracted from large volumes of accounts
• Discourse analysis — organising sentiment tendencies and recurring topics across a body of posts

The iron rule does not change. The themes AI extracts are a first draft. The real meaning of an account, its context, what lies between the lines — a human has to read those. AI helps with organising; the responsibility for interpretation is the researcher's. And above all: raw interview data and audio must stay strictly within the anonymisation and consent principles of Chapter 2.

A prompt to try (structuring qualitative data)

You are a researcher who knows qualitative analysis well. Below are anonymised free-text responses (or interview extracts).

1. Organise the recurring themes into three to five, and give each a short name (a code).
2. Summarise the character of the accounts that represent each theme (summarise; avoid long verbatim quotation).
3. Propose two hypotheses worth testing next, generated from this.
4. Point out, self-critically, where your interpretation may be reading too much in.

[Data] (paste the anonymised accounts here)

Chapter 9 summary

• The main qualitative methods are interviews, participant observation and discourse analysis
• Structured for testing, unstructured for exploring — the same structure as history-taking
• AI accelerates transcription, organising and analysis; interpretation stays a human responsibility