Specify the task and audience, provide context and format, then refine the request after reviewing the first answer, La Nación advises.
The quality of a language model’s answer depends in part on how clearly a user describes the task and the desired result. La Nación’s guide highlights three practical steps: add specificity and context, define the role and output format, then refine the request after reviewing the first answer.
Instead of asking “make a presentation,” the article suggests naming the audience, topic, length and style. One example is a request for five simple slides explaining Argentine inflation to secondary-school students, using everyday examples. That level of detail gives the system direction and narrows the possible responses.
Context includes the purpose of the material and information needed to make the request unambiguous. A user can ask the model to respond as a journalist, teacher or subject-matter expert, then specify a tone such as formal, conversational or technical. The requested format can also be stated, whether that means bullet points, three paragraphs or subheadings. OpenAI’s guidance likewise recommends clear instructions, sufficient context and an explicit description of the desired style.
Even a detailed prompt may not produce a perfect first answer. The guide recommends treating work with AI as an exchange: review the output, ask for targeted changes, add conditions or provide a short example. Choosing a precise verb also helps. “Compare” asks for a different task than “summarize” or “analyze.”
Common mistakes include putting too many unrelated tasks in one message, failing to specify the desired length or audience, and assuming the model will always retain details from earlier conversations. A simple task such as translating a sentence may need only a short prompt. A long report or a programming task generally needs more context, constraints and examples. Spending a few seconds to make the request clear can reduce the number of revisions.