28 ChatGPT prompting tips: how to get the results you actually want

Almost everyone knows ChatGPT, but only a handful of people really know how to use it. Wired has published 28 concrete techniques that take working with the model to a completely different level.

Assign a role and get an expert

One of the most effective tricks is a role instruction. Instead of "explain quantum physics to me," you write: "you are a physics teacher explaining quantum mechanics to a first-year student without formulas." The model immediately changes its style, depth, and vocabulary.

The same approach works for writing code ("you are an experienced Python developer who specializes in security"), data analysis, legal questions, or content creation. The more specific the persona, the more specialized the answer you get.

Examples, format, and constraints

Most users ignore the option of showing ChatGPT directly what the result should look like. Yet this is exactly what sharply cuts the number of iterations.

Basic formatting techniques:

• Add 2–3 examples of the format you want right in the request, and the model will pick up the style
• Specify the output structure: "output as a table," "bulleted list," "three paragraphs"
• Limit the length: "no more than 200 words," "explain in one sentence"
• Set the audience: "explain it to a ten-year-old" or "for CTOs"
• Ban the extras: "no introductory phrases," "no recommendation to consult a specialist"

Few-shot prompting, where you show several "question → desired answer" examples, is especially effective for classification, text styling, and template-based content generation.

Chains of reasoning and iteration

The phrase "let's think step by step" is one of the most powerful tools in prompt engineering. It forces the model to "reason out loud" and significantly lowers the likelihood of hallucinations on complex logic and math problems.

"Smart users don't just type in a request — they iterate: they ask follow-up questions, ask for rephrasing, and add context as they go,"

Wired notes.

Another strong technique is to ask ChatGPT to put clarifying questions to you first, before it answers. This is especially valuable when the user doesn't fully understand what they need.

Other tricks that work:

• "Play the role of a critic and find the weak spots in this argument" — for testing ideas
• "Suggest three different approaches" — instead of a single option
• "What else should I take into account?" — as a final question on complex tasks
• Breaking a large task into several short, sequential prompts

Context is everything

The model knows nothing about you, your project, or your constraints unless you tell it. Adding context at the start of a request ("I'm a B2B SaaS marketer, and my audience is CTOs in manufacturing") dramatically changes the quality of the answers.

The more specific the situation, the more this step matters. "Write an email" and "write a short email to a client who canceled their subscription renewal yesterday, citing budget" are tasks of completely different levels for the model.

What it means

Prompt engineering is neither magic nor programming. It's a skill of structured communication that you can learn in a few hours of practice.

ChatGPT stays a generator of boilerplate answers right up until the user learns to set precise context, role, format, and constraints. After that, it's a completely different tool.