Asking better questions

When people are disappointed by the answers they get from a chatbot, the problem is often how the request was framed. Language models are broad generalists. If you provide a vague, open-ended question, the system defaults to a generic, middle-of-the-road answer that satisfies the statistical average of everything it has read.

You do not need complicated technical jargon or complex prompting frameworks to get better results. A few straightforward habits will dramatically improve the quality of what you get back.

Specify the audience and length

A language model does not know who will read the final text unless you state it clearly. An explanation written for an experienced accountant looks entirely different from one written for an eight-year-old child, even if the underlying subject is the same.

Always include two basic constraints in your request:

1. State who the text is for.
2. State approximately how long it should be.

Telling the model, “Explain this in three concise paragraphs for someone with no legal training,” prevents it from producing eight pages of dense legalistic jargon. These constraints give the model clear boundaries before it starts predicting the first sentence.

Provide an example of good output

If you want a specific tone or format, the fastest way to get it is to provide a brief sample. If you are drafting a weekly team update, paste an update from last month that you liked and instruct the model: “Match the tone, formatting, and brevity of this example.”

Describing a tone abstractly, using words like “professional yet approachable”, leaves room for interpretation. Giving the model a two-sentence example provides an exact stylistic pattern to follow.

Supply the primary material

As discussed in earlier guides, asking a model to recall information from memory produces inconsistent results. When you are asking about a document, a policy, an email, or a set of notes, paste the raw text directly into the prompt. The model shifts from guessing what might be true to summarizing and editing text that is already visible.

Request the reasoning behind the answer

When dealing with decisions, comparisons, or multi-step calculations, ask the model to outline its reasoning before stating its conclusion. You can add a simple phrase like: “Explain your reasoning step by step before summarizing your recommendation.”

This practice helps in two ways. First, writing out the intermediate steps helps the model maintain logical consistency throughout the response. Second, it allows you to inspect the argument. If the model relies on a false assumption in step two, you can catch it immediately, even if the final recommendation sounds persuasive.

Ask where uncertainty remains

Language models tend to sound uniformly confident. You can counter this tendency by explicitly asking the system to identify weak spots in its response. Add a closing instruction: “Note any assumptions you had to make, and identify any parts of this answer where you have lower confidence.”

While the model cannot assess its own knowledge perfectly, this prompt often flags where it had to interpolate missing details or where standard practices diverge.

A realistic before-and-after example

Consider a common task: replying to a concerned email from your child’s mathematics teacher about missed homework assignments.

The vague attempt

“Write a reply to this email from my son’s teacher about missed homework.”

The result of this prompt is almost always an overly formal, verbose letter that sounds like a legal apology. It will likely apologize profusely, make sweeping promises about future study schedules, and sound artificial.

The refined prompt

“Here is an email from my son’s teacher concerning two missed homework assignments this week. Draft a response that is polite and calm, not defensive. Mention that our family had a family emergency early in the week, but that he will complete both assignments by Friday afternoon. Keep the email to four or five sentences and sign off with my first name.”

The second prompt gives the system clear boundaries:

• The exact situation and context.
• The desired tone (polite, calm, non-defensive).
• The specific facts to include (family emergency, completion by Friday).
• The exact length limit (four to five sentences).
• The appropriate closing format.

The output will be close to something you can actually send, requiring only a quick read before you hit send. By setting clear boundaries, you turn the tool into an effective drafting assistant.