How do you use AI for qualitative coding safely?

Oct 12 2026 · Method

AI can speed up first-pass coding of interviews and focus groups — if you redact first, keep the judgement, and log what you kept.

What can AI actually do in qualitative coding?

AI is useful for the mechanical parts of coding: reading long transcripts quickly, spotting repeated phrases, and proposing a starting code list. It can also group passages that look similar and draft a first thematic summary you then rewrite.

• Proposing an initial code list from a sample of transcripts.
• Grouping similar passages so you can review them together.
• Drafting a first thematic summary as a starting point.
• Flagging passages that may fit an existing code.

Where should AI stop in qualitative analysis?

AI cannot decide what a passage means in context, and it should not be the final word on a theme. It also should not see unredacted transcripts or the mapping between pseudonyms and real names.

• Deciding meaning or interpreting tone, silence or contradiction.
• Seeing names, places, employers or other identifiers.
• Being the only coder on a theme.
• Replacing your audit trail or reflexivity notes.

Which tasks should stay human?

How do you redact transcripts before using AI?

Redaction is the step that makes AI use defensible. Do it before any transcript leaves your machine or enters a prompt.

1. Replace names with consistent pseudonyms (P1, P2, Organisation A).
2. Remove places, employers, dates of birth and other identifiers.
3. Keep the pseudonym key in a separate file that never goes into the prompt.
4. Check quotes you plan to publish for re-identification risk.
5. Record what you redacted and when.

What does a safe AI coding workflow look like?

1. Redact a sample of transcripts.
2. Ask the AI for a candidate code list only.
3. Review the list against your research question and framework.
4. Apply the codes yourself, transcript by transcript.
5. Use AI to group passages under codes you have already applied.
6. Draft your own thematic summary, then compare with the AI draft.
7. Log what you accepted, changed or rejected.

How do you keep an audit trail of AI suggestions?

Funders and ethics committees increasingly ask how AI was used. A short, honest log is enough.

• Which tool you used and on which redacted files.
• The prompts you used, in plain language.
• What the AI suggested.
• What you kept, changed or rejected, and why.
• Where the pseudonym key is stored.

Which templates and tools help?

You do not need specialist software to start. A structured logframe, indicator set and evaluation plan keep the coding tied to your questions.

• Logframe template for Google Sheets: /blog/logframe-template-google-sheets
• Indicator builder: /store/indicator
• Evaluation plan builder: /store/evalplan
• Grant writing toolkit: /store/grant
• M&E career resources: /store/career
• Which AI tool to use: /which-ai
• Membership: /membership
• Tools: /tools
• Free handbook: /book

What mistakes should you avoid?

• Pasting unredacted transcripts into a chatbot.
• Treating an AI theme as a finding without checking the data.
• Letting AI assign final codes without your review.
• Forgetting to log what the AI suggested.
• Using AI to replace reflexivity or team coding.

Is AI-assisted coding acceptable to ethics committees?

Usually yes, if you can show that participants' data was redacted, that you kept control of interpretation, and that you logged the process. Check your institution's policy and your participant information sheet before you start. If in doubt, describe the AI use in your ethics application and keep it to the mechanical steps.