What is a prompt and how to structure a good AI command in practice

A prompt is the command, written in natural language, that you use to ask a generative artificial intelligence tool, such as ChatGPT or Gemini, to do something. It tells the AI what to do, for whom, with what information and in what format to deliver the answer.

To structure a good prompt, include four elements: the role the AI should take on (persona), the task, the context and the format of the answer. The clearer and more specific the request, the closer the answer gets to what you needed.

The audience already using these commands is large: 32% of internet users in Brazil, about 50 million people aged 10 or older, used generative AI in 2025, according to the ICT Households 2025 survey by Cetic.br (in Portuguese).

What is a prompt?

In English, prompt means something like a nudge or a cue, and in computing the term has always named the signal that the system is waiting for user input. With the arrival of generative artificial intelligence, the word came to name the request you type itself, whether it is a question, an instruction or a text for the AI to work on.

Microsoft defines the concept like this, in the AI Builder documentation:

“A prompt is mainly made up of a natural-language instruction that tells a generative AI model to perform a task. The model follows the request to determine the structure and content of the text it needs to generate.”

The same text calls prompt engineering the process of creating and refining the prompt the model uses. In practice, it means testing, adjusting and rewriting the request until you get the answer you need.

It helps to separate a prompt from two similar things. It is not programming code, because it does not require special syntax. It is also not a keyword search: instead of returning a list of links, the AI generates a new answer based on what you asked for.

What is a prompt used for?

A prompt turns what you need into an instruction the AI can carry out, whether that is writing, summarizing, translating, organizing ideas, creating images or analyzing data. The quality of the answer depends heavily on how the request was made.

For anyone working as a digital creator, a prompt helps mainly with the stages that tend to stall production:

• generating topic ideas when you hit a creative block;
• building outlines for posts, scripts, emails and lessons;
• rewriting, summarizing or adapting a text to another format or channel;
• proofreading spelling and shortening long sentences;
• creating supporting images from a description.

How to structure a prompt in practice

To structure a good prompt, tell the AI who it should be, what it needs to do, with what information and in what format to deliver the answer. The Google Workspace prompt guide calls these four elements persona, task, context and format.

When a prompt is clear and detailed, it works like a script for the AI to follow. Compare two requests on the same topic:

• vague prompt: “create content about marketing”;
• specific prompt: “create a step-by-step guide on marketing strategies for small businesses in the technology sector”.

The first leaves the AI to guess the audience and the goal of the text. The second already states the format and who the content is for. We explain each element below.

Persona: who the AI should be

Start by saying which role the AI should take on, such as “act as a marketing consultant” or “you are an English teacher for beginners”. The role adjusts the vocabulary and depth of the answer. Without it, the AI tends to respond generically, as if talking to anyone.

Task: what the AI needs to do

Use a clear verb: write, summarize, list, compare, review, translate. Ask for one main task at a time. A request like “write a post, create an ebook and suggest prices” usually produces shallow answers on every item, and it is harder to tell what to adjust.

Context: the information the AI does not have

The AI does not know your business, your audience or the result you expect. Say who the content is for, what the goal is and what should be avoided.

If you have supporting material, such as an old text or the product description, paste it into the prompt. The OpenAI prompt engineering guide recommends including this kind of information, especially when it is your own data that the model could not know.

Format: how the answer should arrive

State the length (up to 150 words, 5 items), the structure (list, table, step by step) and the tone (formal or casual). If you do not define it, the AI chooses on its own, and the time you saved on the prompt ends up going into later adjustments.

Putting the four elements together, a complete prompt looks like this:

Act as a content strategist for Instagram. Create 10 post ideas for a profile that teaches personal finance to people aged 20 to 30 who are starting to invest. Present them in a table, with the topic, the suggested format and an opening line for each post.

Juan Lombana, a digital marketing specialist, used a similar comparison in a talk at Hotmart FIRE 2023:

“Think of prompts as Lego pieces, where we need to use different blocks to build something. First, we need to provide context. How do we want the artificial intelligence to act? Second, we need to provide the information we are going to make available. That will allow the artificial intelligence to understand our expectations. In addition, we must set limits, because that way the artificial intelligence will not do whatever it wants. The better we manage to define the prompt, the better the answer we get.”

Clarity, conciseness and specificity: how to make your prompt more precise

With the structure ready, review the prompt before sending it: it needs to be clear, short enough and specific. These criteria help the AI understand the request the first time and reduce the back and forth in the conversation.

Clarity

Use simple, direct words, as if you were explaining the request to a friend. Avoid terms that each person reads differently, such as “a good text” or “something creative”. If a word can have two meanings, replace it with a description of what you want.

Conciseness

Short, well-defined sentences take the AI straight to the point, with no detours. Being concise does not mean writing little; it means cutting what does not help the AI understand the request. Google’s guide sums the idea up in one recommendation: be brief, but specific.

Every word counts, as Juan Lombana points out:

“Because of one small different word in the prompt, you can change the result completely.”

Specificity

Give context to the request with details of what you want, such as audience, examples, constraints and references. Details pull the AI out of the generic and give it material for more creative answers. A quick test: would another person understand exactly what you want just by reading the prompt?

What are the types of prompts?

The most common prompt types in everyday use are “act as” commands, multiple choice, fill-in-the-blank, yes or no, with examples, image and step-by-step prompts. Each one solves a different need, and you can combine two or more in the same request.

“Act as” commands

Here, you ask ChatGPT or another AI to take on the role of a specific professional or personality. This can include commands such as “act as a senior marketing consultant” or “act as if you were the CEO of Apple”.

These prompts are good for gaining new perspectives and ideas, and they work best when the role comes with a task and a context.

Multiple-choice commands

These prompts give a set of options for the AI to choose from, such as “Which of these messages do you prefer?”. They are useful for making quick decisions between limited options, such as two titles for a post or three versions of a headline.

Also ask for the reason behind the choice, to assess whether the AI’s criterion makes sense for your audience.

Fill-in-the-blank commands

Here, you insert specific information into a sentence or question, which makes the command more elaborate and targeted. For example: “My company offers [product/service] to help customers [achieve goal/benefit]. Which mental triggers should I use in my social media ads?”.

Because the structure repeats, this type of prompt is easy to save and reuse.

Yes-or-no commands

Simple and direct, these prompts ask the AI to answer “yes” or “no”, which works for objective questions. Since the AI can be wrong, it is worth asking for a short justification along with the answer, to check that it makes sense.

Prompts with examples

In this type, you show the AI two or three examples of the result you expect before making the request. OpenAI calls the technique few-shot learning and explains that it steers the model toward a new task with a few input and output examples inside the prompt itself.

The technique works well for keeping your brand’s tone of voice in captions, emails or product descriptions.

Image prompts

In tools that generate images, such as Midjourney and the ChatGPT image generator, the prompt describes the scene you want to see. State the subject, the style (photo, watercolor, 3D illustration), the colors, the lighting and the format, such as square or vertical.

Test, adjust and ask again until you get the right artwork: changing a single word in the prompt can completely change the image.

Step-by-step prompts

For long tasks, split the request into parts and move forward one at a time in the same conversation. Instead of asking for an entire ebook, ask first for the table of contents, then the introduction and then each chapter. That way you review the path before the AI writes everything.

Prompt examples for creating content

The examples below follow the persona, task, context and format structure, and you can adapt them to your niche by swapping the parts in brackets.

If you plan to turn these ideas into a course, see how to sell online courses, and to promote what you create, check these social media tools and how to use email marketing.

Common mistakes when writing a prompt

The most common mistake is treating the prompt like a quick search, with few words and no context. Other frequent slips:

• asking for several different tasks in the same prompt;
• not saying who the content is for or what the goal is;
• leaving the format open, with no length, structure or tone;
• using vague words, such as “good” or “professional”, without explaining what they mean;
• giving up on the first answer instead of asking for adjustments in the same conversation;
• pasting customers’ personal data or confidential business information;
• publishing the answer without checking numbers, names and dates.

The last point deserves attention: the AI can present inaccurate information with the same confidence as correct information. Microsoft itself advises reviewing the generated content before publishing it, sending it to customers or using it in business decisions.

How to use prompts in your content production routine

Prompts can enter several stages of content production, from generating initial ideas to reviewing the final text. The secret is to use one request for each stage, rather than expecting a single prompt to deliver finished content.

Think of the AI as a talented assistant who sometimes needs guidance. It is great for giving you a head start or organizing your ideas, but human supervision is still necessary, because tools like ChatGPT can lose track and bring inaccurate information or connections that make no sense.

One possible routine, step by step:

1. Ideas: ask for a list of topics based on your audience and pick the ones that make sense to you.
2. Structure: ask for an outline with the topics and adjust the order before writing.
3. Draft: generate the text in parts, one section at a time.
4. Review: ask for corrections and cuts and check the data against reliable sources.
5. Adaptation: turn the same content into a post, an email or a video script.

Creativity is still yours. The AI speeds up the work, but the experience and the point of view that make content useful come from you. To go further, see the digital products you can sell once your content is ready.

AI prompt vs. command prompt: what is the difference?

An AI prompt is a natural-language request; the command prompt is the Windows window where you type technical instructions for the operating system. Both use the same word because they wait for user input, but they work in opposite ways.

In the Command Prompt, the cmd.exe interpreter only understands commands with exact syntax, such as “dir” or “ipconfig”, and a typo makes the instruction fail. In generative AI, you write the way you would talk to a person, and the model interprets the request even when the words vary.

Tools for using and creating prompts

The best-known tools for using prompts are text assistants, such as ChatGPT, Gemini, Claude and Microsoft Copilot, and image generators, such as Midjourney. There are also marketing-focused platforms, such as Jasper, with ready-made features to create ads, emails and posts.

If you have heard of DALL-E, you should know that OpenAI shut the model down on May 12, 2026, according to the company’s deprecations page. In its place, OpenAI recommends the GPT Image models.

These same tools help you create prompts. You can describe your goal and ask the AI to suggest a better prompt, or ask it to question you about the audience and the format before answering. It is a practical way to find out what information was missing from your request.

But, a word of caution: if you want your content to generate value for your audience, always review what the tool delivered and check that the information is correct. That way you avoid mistakes that can hurt your reputation.

Hotmart AI and Launchpad: prompts to create and sell your digital product

At Hotmart, the all-in-one platform for creators who want to create, sell and scale digital products, two features work from natural-language requests: Hotmart AI and Launchpad. Both apply the same logic you saw in this guide.

Hotmart AI is a free tool: you enter the topic of your digital product and get suggestions for strategies to structure, price and sell an online course. Inside the members area, the structure suggestion feature uses the topic you entered to propose modules and lessons, and you choose which ones to add to the course.

Launchpad, introduced at Hotmart FIRE 2026, brings the prompt to the creation of pages and apps. According to the official Hotmart announcement (in Portuguese), you describe by text or audio what you want to build, the AI generates a first version and you adjust it through conversation, without coding, with checkout already integrated. It is a quick way to build a landing page for your product.

In both cases, the more context you give about your audience and your goal, the more useful the suggestions get. Want to discover a new way to start your own digital business? At Hotmart, your knowledge has value. Create your free account and learn what Hotmart is and how it works.