How to start using ChatGPT from scratch

You do not need to learn a collection of secret words to start using ChatGPT. What changes an answer most is not telling the tool to be “the world’s greatest expert”, but explaining what you want to achieve, what information it has, and how you will know whether the result is any good.

The simplest way to use it is as an assistant that prepares a first version. You bring the judgement, you review what comes back, and you decide what gets used. That difference matters: a well-written answer is not automatically a correct one.

The idea in one sentence

Write down the specific task, add the context that is essential, ask for a format you can use, and review the result before you use it.

Why you do not have to talk like an engineer

The word “prompt” simply means the request you type. You do not need to build a complicated formula. An ordinary request can be enough if it answers four questions:

• What do I want to achieve?
• What does ChatGPT need to know in order to help me?
• Who is the result for?
• How do I want to receive it?

For example, “organise these ideas” is an open request. “Organise these ideas so I can prepare a 30-minute meeting, separate decisions from open questions, and finish with a list of three next steps” already contains a goal and a format.

That does not mean the second request will produce something perfect. It means it reduces the number of things the tool has to guess.

My four-sentence method

When I start a task, I write something like this:

1. “I want to achieve…”
2. “The important context is…”
3. “I need the answer as…”
4. “Before finishing, check…”

A practical example would be:

“I want to prepare an outline explaining a digital tool to someone who is not technical. The context is that they need to solve one specific task and have little time. Give me about 500 words, with numbered steps and a clear note about the limits. Flag anything you cannot verify instead of making it up.”

That request uses no jargon. Even so, it states the goal, the reader, the length, the structure and the standard of confidence.

Step 1: start with a small task

The first test should not be a contract, a medical decision or a confidential document. Pick something you can check in a few minutes:

• organising the ideas in an email;
• turning notes into a task list;
• summarising a public text;
• suggesting titles for a guide;
• preparing questions for a conversation.

A small task lets you learn how the tool responds without putting sensitive information at stake. It also forces you to define what “done well” means. If the result should be a list, a table or a short text, say so from the start.

Step 2: give context, but not the whole file

More information does not always produce a better answer. If you paste ten pages when only two paragraphs are needed, you add noise and increase the risk that what matters gets lost.

Before copying anything, ask yourself:

• Is it needed to solve the task?
• Does it contain names, phone numbers, addresses or identifiers?
• Does it include information about someone else who has not authorised me to share it?
• Can I replace the name with “client A” and a figure with “amount X”?

In my own project I learned this difference while working with strategic and technical documentation. Deciding the next step did not require handing over every detail: first I had to separate vision, priorities, status and execution. That sorting improved the answer before adding any more text.

Step 3: ask for an output you can check

A useful request does not end with “give me the best answer”. Ask for a shape of output that lets you verify it:

• “Separate facts, decisions and open questions.”
• “Say which claims need a source.”
• “Give me a summary first and then the steps.”
• “Mark anything you cannot confirm with [TO CHECK].”
• “Do not invent names, prices, dates or results.”

If you are still learning, you can also ask it to explain its assumptions. You do not have to accept those assumptions; they help you find what needs correcting.

Step 4: work in short rounds

Do not try to solve everything in a single request. Do one round, then fix one thing at a time:

1. Ask for an outline.
2. Remove or correct whatever does not fit.
3. Ask it to develop only the useful section.
4. Check the facts, the tone and the length.
5. Prepare the final version outside the chat.

This process is easier to control than asking for a finished document and copying it without reading. It also stops an early mistake spreading through the whole text.

What to do when the answer sounds convincing

Fluency is a sign of good writing, not proof of accuracy. Check in particular:

• dates and figures;
• feature names and the terms of services;
• links and sources;
• recommendations that could cost you money;
• claims about security, health or legal matters.

For a public guide, the original source should count for more than the confidence with which the answer is written. OpenAI’s own guidance recommends being clear and specific and improving the request iteratively; it does not promise that a well-written request removes the need for review.

Common mistakes

Asking it to “make it perfect”

“Perfect” does not explain the purpose, the reader or the constraints. Replace it with a standard you can check: “no invented data, numbered steps, and a warning about what still needs verifying”.

Confusing more words with more context

Context should help it decide, not bury the task. Start with what is needed and add more only if it changes the answer.

Handing over the whole decision

ChatGPT can compare alternatives or prepare questions. The final decision is still yours, especially where money, privacy, work or health are involved.

Testing with sensitive information

Before using personal material, review the service’s data controls. In a personal ChatGPT account you can turn off “Improve the model for everyone” in the data controls; temporary chats are handled differently, and OpenAI states they are removed from its systems after 30 days, with the security exceptions described in its documentation.

What I would do

I would start today with a task I already know well. I would write the result I want in one sentence, add only the essential context, and ask for an output that is easy to review. Then I would check the important claims one by one.

That method is what let me spot a problem while building my own site: an automation could turn a planning note into something that looked like an article. The fix was not writing a grander request, but separating the editorial work from the technical work, reviewing the piece here, and not passing it down the line until it was genuinely resolved.

Summary, to start now

1. Pick a small, low-risk task.
2. Explain the goal, the context, the format and the standard of quality.
3. Ask it to flag doubts and data that need checking.
4. Review the result and improve one thing at a time.
5. Do not share secrets, and do not delegate decisions you need to understand.

ChatGPT can save you preparation work. Control is not saved: it stays with the person who knows the goal and reviews the result.

Sources checked

• Best practices for prompt engineering in ChatGPT, OpenAI Help Center, checked on 2026-09-02.
• ChatGPT data controls, OpenAI Help Center, checked on 2026-09-02.

Related

• Ten genuinely useful things to do with ChatGPT
• How to spot a fraudulent text, WhatsApp or email

Review

Responsible for publication: Ibrahin. Date: 2026-09-02. Last checked: 2026-09-02.