Ethics of AI-Assisted Qualitative Analysis

Description

A governance protocol for using AI in qualitative analysis, covering disclosure, data privacy, reliability checks against human coders, and human authority over interpretation.

Generative AI can now segment, code and summarise interview transcripts in minutes, but it does not interpret. This protocol treats AI as a junior coder whose work must be disclosed, validated and overruled where necessary. It specifies what to report (model, version, prompts, tasks delegated), how to protect participant data (local or contractually bounded models, de-identification before upload, consent wording that covers machine processing), and how to establish reliability: AI-applied codes are compared with an independent human coder using Cohen's kappa before the AI output is trusted at scale. Interpretation, theme construction and claims about meaning remain human work. Its rigour lies in treating AI output as a measurement instrument with error that must be quantified, rather than as an oracle, and in an audit trail that lets reviewers see exactly where machine judgement entered the analysis.

Attributes

Loss type
Latent
Stage
Definition
Effort
Days

Steps

  1. Disclosure in reporting: Report which AI system, for which coding tasks, with what IRR validation. Non-disclosure is an ethical violation.
  2. Data privacy: Cloud-based AI processes data on external servers. Use local models for sensitive participant data where consent does not cover external processing.
  3. IRR requirement: AI-assisted coding without IRR validation is not rigorous. κ must be calculated and reported.
  4. Human final authority: AI produces coded segments. Humans produce analysis, themes, and theoretical propositions.

Sources

  • AI for Humans (Heaton)
  • Sage Handbook of Qualitative Research Ethics

Also known as

  • AI-assisted coding governance
  • LLM-assisted qualitative analysis protocol
  • Responsible AI in qualitative research

Categories

  • Research Operations
  • Research Design & Rigor