RESEARCH PRACTICE · 4 min read
What to check in AI-assisted open-text coding
Five checks for AI-assisted survey verbatim coding, with attention to codeframes, minority views and research methods.
AI can help organise large sets of verbatims. The value depends on how easily a researcher can inspect the proposed coding and decide whether it fits the study.
Keep response-level evidence
A theme summary should link back to the actual responses. Preserve identifiers and verbatim text so reviewers can check why a code was assigned, inspect overlap and correct errors.
Test an agreed codeframe
If a project has an existing codeframe, assess whether the workflow applies it consistently. Use a reviewed sample that reflects the study and inspect disagreements. Report what was tested and where performance is weaker.
Inspect minority views
The most common theme is not the whole story. Check small groups, exceptions and contradictory comments. Semantic clustering can help exploration, but group size alone does not determine the importance of a finding.
Treat sentiment as context-dependent
A response may praise one aspect and criticise another. Decide whether sentiment is measured at response, theme or aspect level, then test examples containing mixed views, irony or ambiguous wording.
Match the method
Structured content analysis and reflexive thematic analysis have different aims and assumptions. Agree what role AI will play in the selected approach. Suggested codes and summaries do not remove the researcher's responsibility for interpretation.