Code customer feedback with an explicit rubric

Feedback categories are useful when another reviewer could apply the same rules. Define inclusion and exclusion criteria before asking for sentiment or themes.

Define the evidence boundary

Give the research teammate a bounded set of documents or an explicit source scope. Request references next to claims and keep observation separate from interpretation. An AI summary is not independent verification of a source. Where live web access is unavailable, work from the supplied snapshot and label its date rather than describing the findings as current.

Inputs for this workflow: Sanitized feedback records with stable IDs.

Work through the task

1. Draft a small coding rubric.
2. Apply it with an uncertainty field and retain original IDs.

A brief you can adapt

Write coded-feedback.csv with record_id, category, evidence phrase and confidence note. Allow multiple categories and an Unclear bucket.

The filenames above are examples. Replace them with your actual inputs and destination before submitting the task.

Review the deliverable

Have a second reviewer code a sample and discuss disagreements before processing the full set.

Where this approach can fail

A sentiment label is an interpretation, not a measured customer outcome.

For the related desktop setup, see write task brief.