Produce Reliable Business Prompts with Google Gemini

Table of Contents

1. Challenge

   Evaluate the original prompt against a stated standard

   • Run the prompt "Summarize this email." in Studio on three sample emails.
   • Check each summary against the five criteria in your manager's quality standard and note where it fails.

   At the end of this objective, you'll know which kinds of failures a vague prompt produces on realistic emails.

2. Challenge

   Rewrite the prompt to define role, task, constraints, and acceptance criteria

   • Replace the vague prompt with a structured one, matching each constraint to a criterion in the quality standard.
   • Run the new prompt on Email 1 to confirm that Gemini follows the structure.

   At the end of this objective, you'll have a structured prompt that addresses the failures you found.

3. Challenge

   Compare the effect of model settings on the same inputs

   • Use the Compare view to run the same prompt and email at a minimal and a high thinking level.
   • Run it again with a very low and a sufficient output token limit, and compare the responses.

   At the end of this objective, you'll know which settings give more consistent, complete summaries, and why the length limit belongs in the prompt rather than in the token limit.

4. Challenge

   Retest the revised configuration and verify it meets the standard

   • Run the final prompt and settings on all three emails.
   • Score each output against the same five criteria and compare the results with your first run.

   At the end of this objective, you'll know whether the prompt reliably meets the standard, and how to fix one thing at a time when an output still fails.