How to create effective prompts for artificial intelligence

An unsatisfactory response from an AI doesn’t always mean the tool is incapable. Often, the problem lies in how the request is phrased. When instructions are vague, context is missing, or the expected outcome is unclear, the machine has to fill in the gaps on its own, which can lead to errors.

A simple strategy to improve communication with AI systems is to structure prompts into three components: the task to be performed, the relevant context, and the desired output format. While this approach doesn’t guarantee perfect results, it reduces ambiguity and increases the likelihood of achieving the desired outcome.

1. Define the task clearly

The first step is to make your expectations explicit. “Write about artificial intelligence” leaves many possibilities open, whereas “write a 2,700-character article on the use of AI in customer service” sets a clear goal. Including an action verb—such as write, summarize, compare, review, organize, explain, or analyze—helps the AI understand the nature of the task. For complex tasks, breaking them into steps can be helpful: first identify key data, then draft the text, for example.

2. Provide necessary context

The same command can produce very different results depending on who will receive the material, the purpose of the communication, and which information should be considered. When asking for an email to be written, for instance, specify who will receive it, the relationship with that person, the reason for the contact, and the desired tone. When reviewing a resume, indicate the target position and the experience that should be prioritized. Restrictions should also be mentioned: whether certain data cannot be invented, whether the text must be a specific length, or whether it must follow a particular style—make this clear in the prompt.

3. Specify the output format

Even when the AI understands the task, it’s still necessary to define how the response should be presented. The user can request a list, a table, a continuous text, bullet points, or a step-by-step answer. Additional details, such as character limits, number of options, target audience, or level of depth, help shape the result. A prompt like “compare these three tools in a table, considering price, integration, and ease of use, and conclude with the main differences” provides much more guidance than simply “compare the tools.”

Iteration improves prompts

You don’t need to get everything right on the first try. If the response is generic, the user can point out exactly what was missing and request a new version. Practice also helps identify which information tends to make a difference in each type of task. For recurring requests, it’s worth creating a prompt template with fixed instructions and changing only the specific data for each request.

By following these guidelines, users of generative AI can obtain responses that better align with their expectations, reducing refinement time and increasing productivity.