An AI assistant can produce a polished answer to a vague question. That is convenient, but it creates a small trap: the quality of the writing can make an unclear task feel resolved.

Better questions do not guarantee correct answers. They make it easier to see what the answer is supposed to accomplish, which information it should use, and where a human needs to check the result.

Give the task a shape

“Explain this company” leaves many decisions open. Is the reader considering a job, studying a business model, or comparing suppliers? Should the answer explain the product or examine financial performance? Which time period matters?

A useful prompt makes a few of these decisions explicit. Specify the task, intended reader, available evidence, and desired output. For example:

Using the attached annual report, explain the company's main revenue drivers for a reader new to financial statements. Separate reported facts from your interpretation. Cite the relevant sections and flag questions the report does not answer.

This gives the response a purpose and a boundary. It also creates things to inspect: are the cited sections real, do they support the claims, and has interpretation been presented honestly?

More instructions are not automatically better. A short, coherent request often works better than a long list of competing requirements. Include details that change what a good answer would look like.

Ask for evidence you can inspect

When accuracy matters, supply the source material or ask for sources that can be opened. Request a passage, a calculation, or a clear statement of an assumption. These are useful because they can be checked independently.

An explanation of reasoning can help reveal assumptions, but a plausible explanation is still generated text. It should not substitute for evidence. Likewise, a citation-shaped link is not proof that the linked page supports the sentence beside it.

For numerical work, identify the inputs and their units. Ask the assistant to show the formula, then verify the calculation with a calculator or spreadsheet. For current information, check the source date and whether a more recent document changes the conclusion.

Turn one large request into a short exchange

Complex work often benefits from stages. First ask for a structure or list of open questions. Then choose the relevant parts, provide missing information, and request a draft. Finally, review the draft against the original task.

This creates opportunities to catch a misunderstanding before it spreads through a long response. Follow-up questions can be simple:

  • Which statements are directly supported by the supplied material?
  • What information would change this conclusion?
  • Which terms need a clearer definition?
  • What remains unresolved?

Good prompting is less about finding a secret phrase than about making the work inspectable. Define the job, keep the evidence close, and leave uncertainty visible. The result may still need correction, but the conversation becomes a more useful place to think.