A prompt produces inconsistent reports because it omits the intended audience, source context, and output structure. Which approach best fits?

1. Evaluate grounding with Google Search for the supported model and task.
2. Set appropriate output-token limits and request the desired concise format.
3. Specify the task, relevant context, audience, and required format.
4. Evaluate a higher supported sampling temperature.