A parks botanic-garden chatbot keeps mangling Latin plant names in resident answers. Staff have one gold reply that shows the required Genus species formatting. Which prompting change best fixes this narrow formatting failure?

One-shot prompting fits this parks botanic-garden chatbot because the failure is narrow and format-specific: Latin binomials must follow Genus species style, and staff already have a gold reply that demonstrates it. Adding that example to the prompt conditions the model on the required surface form without rebuilding the corpus. It shows the capitalization, italics-or-plain convention, and genus-species order through a concrete municipal answer. Switching the garden corpus to masked language modeling with random plant-name blanks is the wrong lever: masked language modeling is a pretraining or continued-pretraining objective for bidirectional encoders or encoder-style tasks, not a prompt-time remedy for a live chat formatting bug, and blanking names does not teach the gold layout. Raising temperature makes spelling and structure less stable, inviting creative but incorrect botanical spellings rather than preserving Genus species. Removing plant names from the system prompt and relying on zero-shot guessing discards civic naming cues and leaves formatting to chance. Exam caveat: one-shot prompting helps when one clear example captures the rule, but a single example may not cover scientific-name edge cases such as hybrids, cultivars, or authorities. Practical check: after adding the gold reply, sample resident questions that previously mangled names, verify Genus species against the garden checklist, and keep the exemplar in versioned prompt configuration so staff can refresh it when nomenclature standards change.