Skip to content

AI & Automation · Guide

How to use ChatGPT effectively

Most weak answers come from weak prompts. Here is the four-part structure that fixes them, twenty prompts you can copy, and the three jobs to never hand over.

The Nextversity teamAI & Automation schoolUpdated August 10, 20266 min read

On this page
  1. The short answer
  2. The structure, applied
  3. Twenty prompts worth keeping
  4. The habits that improve output most
  5. Three jobs to keep for yourself
  6. Where prompting becomes a real skill

The short answer

Better answers come from better prompts, and better prompts have four parts:

  1. Context. Who you are, what the situation is, who the output is for.
  2. Task. One clear instruction, not three stacked together.
  3. Format. Length, structure, tone, and what to leave out.
  4. Constraint. What to avoid, what to assume, what it must not invent.

"Write me a marketing email" gets you a generic marketing email. The same request with those four parts gets you something you can edit rather than rewrite.

The structure, applied

Weak: Write a job ad for a designer.

Strong: We are a five-person studio hiring our first junior designer, mostly brand and packaging work, remote within Australia. Write a job ad of about 250 words for people with one to two years of experience. Warm and plain, no corporate cliches, no salary figures. Include what the first three months look like.

Same model, entirely different output. Everything else on this page is a variation of that difference.

Twenty prompts worth keeping

Writing

  1. "Rewrite this so a busy reader gets the point in the first sentence: [text]"
  2. "Cut this by 40% without losing any specifics: [text]"
  3. "Give me three subject lines for this email, each with a different angle: [text]"
  4. "Rewrite this in plain language for someone outside the industry: [text]"
  5. "What is unclear or missing in this paragraph? Do not rewrite it, just list the gaps."

Thinking

  1. "Argue the strongest case against this plan: [plan]"
  2. "What would have to be true for this to work?"
  3. "List the five questions I should ask before deciding this."
  4. "Summarize this into the three decisions it is actually asking for: [text]"
  5. "What am I likely to be assuming here without noticing?"

Learning

  1. "Explain [topic] in three levels: one sentence, one paragraph, one page."
  2. "I understand [thing I know]. Explain [new thing] by comparing it to that."
  3. "Quiz me on [topic] with five questions, one at a time, and tell me what I got wrong."
  4. "What is the most common misunderstanding beginners have about [topic]?"

Work

  1. "Turn these notes into an agenda with time estimates: [notes]"
  2. "Draft a polite reply that says no without closing the door: [message]"
  3. "Give me a checklist for [recurring task], ordered so nothing blocks anything else."
  4. "Read this spreadsheet description and suggest the formula: [description]"
  5. "Write the first draft of a slide outline for this argument: [argument]"
  6. "What did I forget in this plan? Be specific and blunt."

Number 20 is the most useful prompt on this list, and the one people use least.

The habits that improve output most

Iterate rather than restart. The second and third messages are where quality arrives. "Too formal, and drop the last paragraph" beats writing a brand new prompt.

Give it an example. One sample of what good looks like teaches it more than three paragraphs of description.

Ask for the reasoning when it matters. "Explain why you chose that structure" catches confident mistakes.

Split big asks. Outline first, then draft each section. One giant prompt produces one mediocre wall of text.

Use AI to speed up the parts you understand. Using it to skip the understanding is how you end up unable to check its work.

Three jobs to keep for yourself

  • Anything factual you will publish. It predicts plausible text, not verified facts. Check numbers, names, dates and quotes against a source. OpenAI's own help center is clear about the limits.
  • Anything confidential. Follow your organization's policy on what can be pasted in. When in doubt, anonymize.
  • The final judgment. It does not know your audience, your politics or what your manager asked for last quarter.

Where prompting becomes a real skill

The gap between casual use and skilled use is bigger than it looks: structured prompts, reusable templates, chaining steps, and knowing which tasks suit AI at all. OpenAI's prompt engineering guide is a good technical starting point.

The AI prompt engineering certificate covers that ground with practice, and there is a DeepSeek version if you use that model. One subscription opens the whole AI & Automation school.

Better in, better out. That really is the entire skill, and it takes a few hours to learn properly.

Questions people ask

What makes a good ChatGPT prompt?

Context, a specific task, a stated format and a constraint. Telling it who the answer is for and how long it should be does more for quality than any clever phrasing.

Why are my ChatGPT answers so generic?

Because the prompt was generic. Add the audience, the purpose, an example of what good looks like, and what to avoid. Specific input is the whole trick.

Can ChatGPT be wrong?

Yes, and confidently. It predicts plausible text rather than checking facts, so anything factual, numerical or legal needs verifying against a real source before you use it.

Should I tell it to act as an expert?

Role prompts help a little by setting tone and vocabulary. They matter far less than giving real context about your situation, which is the part people leave out.

Is it cheating to use AI for work?

That depends on your workplace and the task. Check your organization's policy, never paste confidential data into a tool that is not approved, and be transparent when the output is going out under your name.

Keep reading