Clerk staff want to stabilize tone on ballot-explanation answers using a single gold sample. Which technique fits?
Select an answer to reveal the explanation.
Short Explanation
One excellent sample is a tuning fork for tone. Paste that gold ballot explanation as a one-shot example and the model echoes the clerk’s voice. Banning examples—or flooding the prompt with contradictions—will not stabilize style.
Full Explanation
Clerk staff stabilizing tone on ballot-explanation answers with a single gold sample should use one-shot prompting: paste that approved ballot explanation so the model anchors voice, structure, and neutrality to the clerk’s standard. One high-quality exemplar is enough to tune tone without a large library when the goal is stylistic consistency rather than broad topical coverage across unrelated domains. Zero-shot with an explicit ban on examples forgoes the available tone anchor and leaves neutrality to chance on sensitive election language. Unsupervised clustering of ballots does not supply a prompt exemplar for answer tone and addresses a different analytical problem entirely. Flooding the prompt with thousands of contradictory examples cancels stable tone and confuses the model about which clerk voice to follow. Exam caveat: the gold sample must be legally reviewed for impartiality; a biased one-shot propagates bias at scale across ballot explanations. Operational check: generate multiple ballot explanations with the one-shot present, score tone consistency with clerk rubrics, refresh the gold sample when ballot styles or statutory language change, and keep the exemplar under elections-office version control with change history.