Prepare participant research data for AI
Participant data can remain identifiable through a distinctive quotation, rare experience or combination of attributes. Replacing participant names is not enough to establish anonymity or permission to use another processor. Start with the study's approved purpose and receiving environment.
For Research operations and authorized study teams
Synthetic example: testing a coding framework
A fictional study team wants to test a thematic-coding table. The responses are invented to include overlapping themes and ambiguity; they are not participant quotations, study findings or evidence about any population.
What you are working with
- A DOCX interview extract containing names, places and personal experiences.
- An XLSX participant table with demographics and recruitment references.
- A proposed codebook defining themes and examples.
A safer approach
- Use invented response excerpts to test the codebook structure first.
- Keep participant mappings and recruitment data outside the AI-bound material.
- Preserve uncertainty and allow multiple candidate codes with supporting text.
Expected outcome: The assistant suggests how the proposed codebook applies to synthetic excerpts. Researchers review disagreements and revise the method without presenting generated categories as actual study results.
Work through the procedure
Verify the study boundary
Identify the responsible owner and the documented permissions for processing and sharing. Specify the AI task, input and destination. Do not infer permission from the fact that the team already holds the data; method testing can often proceed with invented responses while approval is clarified.
Separate analysis from participant linkage
Remove unnecessary identifiers and recruitment references from any approved extract. Consider rare combinations of age range, location, occupation and experience. Keep the linkage key in the authorized study store, and do not reuse externally shared labels across unrelated studies.
Protect meaning while minimizing
Record where an approved excerpt has been generalized so readers understand its limitations. Avoid rewriting a real quotation and then presenting it as verbatim. Check document comments, hidden table columns and properties, and use wholly synthetic material where safe generalization would distort the research question.
Review coding disagreements
Ask for supporting passages and plausible alternative codes. Have researchers inspect omissions, ambiguity and divergent readings. Retain a record of approved methodological changes and do not convert a model's output into participant findings without the study's established analysis and review process.
What to check before proceeding
1. Authorized analysis
- Ready when
- The study owner confirms that the proposed material and destination fit the documented scope.
- If the check fails
- Use invented examples or keep the activity within approved study tools.
2. Indirect identification
- Ready when
- The team has reviewed distinctive quotations and combinations of attributes.
- If the check fails
- Reduce the excerpt, aggregate appropriately or exclude it from the workflow.
3. Interpretive traceability
- Ready when
- Suggested codes cite supplied text and preserve alternative interpretations.
- If the check fails
- Return the output for researcher review rather than accepting a confident label.
Common mistakes to avoid
- A searchable quotation can identify a participant even after names and direct contact details are removed.
- Generalizing a response may change its meaning; do not hide that transformation or present edited text as a verbatim quotation.
Evaluate this workflow with Aona
Where Aona can help
Use synthetic study documents to test Aona's configured prompt and file policies. Supported DOCX, XLSX and PDF formats require verification on the actual provider and browser/native path.
What to confirm
Aona does not establish research consent, guarantee anonymity or validate thematic findings. Managed policy evaluation uses regional processing, which must be included in the study's approved data-handling review.
Managing this document workflow across a team?
Review your AI tool, document format and data-handling requirements. Use a synthetic example to discuss supported controls and the checks your team still needs to perform.