Answers Are Too Vague, How to Get More Useful Results?

Specify What Information Is Missing

Saying "the answer is not good enough" is difficult to guide the next step. You can clearly state that reproduction steps are missing, original references are not cited, the environment is confused, or there is no distinction between verified facts and speculation. Pointing out specific problems is more effective than resubmitting the entire original task.

First confirm that the model has received the necessary materials; without version info, error messages, or scope of the goal, longer replies may just be guesses.

Constrain Results with Output Standards

Provide an easy-to-check format, for example, four columns: "Issue, Evidence, Impact, Next Steps," and require marking unknown information.

For example, instead of just asking for "more details," say "Please add the log snippet supporting the second conclusion and specify which condition has not yet been verified." This kind of feedback can directly change verifiable results and help you discover whether the issue is missing materials or missing reasoning.

1. Add materials directly related to the conclusions.
2. List constraints that must be met and conditions that cannot be changed.
3. Request the key conclusions first, then expand on the parts you really need.
4. Check results against evidence and correct omissions one by one.

When to Switch Models

Only after materials and task descriptions are sufficient is it worthwhile to compare other available models. Keep input and evaluation criteria consistent to judge differences. Switching models is not a way to fix all errors; when dealing with real environments, let tools verify key facts or clearly state which content cannot yet be verified. Before final delivery, verify key evidence and operational impacts yourself.