I don’t prompt it. I talk to it.

A blank prompt box can make people think there must be a perfect sentence that gets the right answer. I approach it more like a conversation about the work. I explain the situation, ask what is missing, and push back when the answer does not fit.

That does not mean treating a tool’s confidence as understanding. It means giving the exchange enough context to be useful while keeping responsibility for the result. The conversation is a working method. It is not a reason to stop checking.

Start with the situation

If I want help drafting a business reply, I explain who is reading it, what happened, and what the reply needs to accomplish. I include the facts the reader needs and the things I am not authorized to promise.

That is more useful than asking for something to sound professional. Professional can mean almost anything. A clear explanation of the job gives the draft a direction I can evaluate.

I also say what I do not know. If the date is not confirmed, I want the tool to keep it unconfirmed. If a policy is missing, I want a question about the policy rather than an invented answer.

Ask it to expose the gaps

A useful next question is: what would you need to know before this could be sent? The answer may point to a missing decision, an unclear recipient, or a claim that needs a source.

I read that list with the same skepticism as the first draft. The tool can misunderstand the task or suggest irrelevant questions. But the exercise changes the exchange from making sentences to examining the work those sentences are meant to do.

When an answer seems too smooth, I ask which facts support it. If the support is not in the material I provided, it needs verification or removal. A plausible explanation is not evidence.

Revise the actual problem

Sometimes the draft is bad because the request is vague. Sometimes the request is clear and the draft is still bad. I try to name the difference before asking for another version.

Perhaps the reader needs the next action earlier. Perhaps the language sounds like an agency when the work is being done by one person. Perhaps the draft has turned an intended outcome into a promise. Each is a specific correction. Asking for more polish would not fix the underlying issue.

The same approach works when thinking through a workflow. Describe the input, the person responsible, the exception and the finish. Then look for the place where the proposed system assumes something a person has not decided.

Keep the responsibility

I am still the person choosing the facts, approving the words and deciding whether the output is useful. A conversational style does not make private information safe to share with an ordinary tool. It does not remove the need for agreements, review or professional judgment.

Try the approach on a low-risk task with public information. Explain the situation in ordinary language. Ask what is unclear. Correct the assumptions. Read the result as the person who has to stand behind it. The point is not to perform a special kind of prompting. It is to think clearly enough to know what you are asking the tool to help you do.