brainax

How to use ChatGPT as a beginner: prompts that work, traps nobody mentions

Short answer: tell the chatbot four things (who to be, what to do, the format you want, an example), refine the answer in the same chat instead of starting over, and check every number, date, link and quote yourself, because the model can sound sure and still be wrong. The same rules work in ChatGPT, Claude and Gemini. Below is why each one works and two real stories of what happens when people skip them.

Free course on this topicAI basics: free courseStart the course →

What is ChatGPT, actually?

Think of nesting dolls. AI is any program that does "smart" things. Inside it sits machine learning: programs that learn from examples instead of rules a person wrote. Inside that are neural networks, and inside those are language models (LLMs), neural networks trained on text. ChatGPT, Claude and Gemini all live in that smallest doll.

Why this matters: a language model is not a search engine and not a database. It was trained on a huge pile of text to guess what comes next, and that one skill grows into answering questions, translating and writing code. Source: the AI, ML and neural nets lesson.

What it means for you: pick whichever chatbot you like. The skills below transfer, because they all work the same way under the hood.

How does ChatGPT come up with its answers?

It writes piece by piece. Each step it picks the best-fitting next token, a chunk of text that can be a short word, part of a word or a comma. After "Bitcoin is often called digital…" it is far more likely to pick "gold" than "pizza".

Why it matters: the model learned patterns, not exact pages. So its writing is smooth, but it can misquote a source. It also has a knowledge cutoff: training ends on some date, and after that it knows nothing new unless it searches the web, which it doesn't always do. Source: how a neural network thinks.

What it means for you: a confident tone says nothing about accuracy. Ask how many moons Mars has and you may get a confident wrong answer; the right one is two, Phobos and Deimos. Fresh news, prices and exchange rates: check yourself.

How do you write a prompt that actually works?

A prompt is just your message. A good one has four parts:

Why it works: the model can't read your mind, it only continues your text. "Tell me about marketing" gets you a lecture about nothing. "Give me 5 TikTok post ideas for a coffee shop near a campus, one line each" gets you something you can post today. Long prompts don't annoy it; details help. Source: the prompt that works lesson.

Prompt to copy: You are a career coach. Help me prepare for a barista job interview. Give me the 5 most likely questions, a short strong answer for each, and one question I should ask them. Keep it under 200 words.

What do you do when the first answer misses?

Don't open a new chat. Reply in the same one: "shorter", "simpler", "add a real-life example", "make it a table".

Why it works: inside one chat the model sees the whole conversation, so each follow-up builds on what it already did. A new chat starts from zero and often gives you the same thing again. Usually 2–3 follow-ups are enough. Source: the prompt lesson.

What it means for you: start a new chat only when you switch topics. Mixing your physics homework and your CV in one chat confuses the context.

Why does ChatGPT make things up, and how do you catch it?

Made-up facts are called hallucinations: invented numbers, quotes, studies and links that look exactly as convincing as real ones. They happen because the model picks a plausible continuation, not the truth. If it doesn't know, it will still write something that looks like an answer.

The most famous example: in 2023 a New York lawyer used ChatGPT to find court decisions for a client suing the airline Avianca. Six of the cases in the filing didn't exist, and when asked, the chatbot insisted they were real. On June 22, 2023 a federal judge fined two lawyers and their firm $5,000 (AP News).

What it means for you: always check numbers, dates, laws, health and money. Ask for a source, then open the link yourself and make sure it says the same thing. Link dead or saying something else? You can't use that fact. Asking the bot "are you sure?" isn't enough, it may confirm its own invention. Source: hallucinations and privacy.

What should you never paste into a chat?

Passwords, card numbers, seed phrases (the 12 or 24 words that open a crypto wallet) and other people's personal data. Whatever you send goes to the company's servers and may be stored there.

Real story: in April 2023 sensitive internal Samsung data ended up in ChatGPT by accident. From May 1 Samsung temporarily restricted generative AI on company devices, noting that data sent to outside AI services is hard to "retrieve and delete" (TechCrunch).

What it means for you: don't type into a chat what you wouldn't show a stranger. Already sent something? Change the password and turn on two-factor login, call the bank to reissue the card, or move crypto to a new wallet and treat the old one as lost. Your own study notes are fine. Source: hallucinations and privacy.

FAQ

Is ChatGPT free to use?

Yes, ChatGPT, Claude and Gemini all have free plans; paid plans mostly add higher limits and stronger models. Check the current limits on each company's official site, they change often.

Which is better for beginners: ChatGPT, Claude or Gemini?

They are the same kind of tool, so the prompt skills are the same. Run one prompt in two of them and keep the answer that fits you better, then check the facts either way.

Can I use ChatGPT for school or uni work?

Use it to explain topics, quiz you and give feedback on your draft. Check your school's rules before handing in AI-written text, and check every fact, since teachers spot hallucinated sources fast.

Try it now

The free lesson on the 4-part prompt takes about 10 minutes, and you leave with a template that works in any chatbot. Start the lesson

Free course on this topicAI basics: free courseStart the course →