How to Use ChatGPT: A Beginner's Guide to Prompts, Custom GPTs & API

When people ask me what ChatGPT actually is, I usually say it's like having a really well read intern who never sleeps, never complains, but needs extremely clear instructions to do anything useful.

That last part is important. The clear instructions part. Skip it and you'll wonder why everyone else seems to be getting better results than you.

I want to walk through the things that actually matter when you're starting out. Not the hype. Not the scary technical stuff. Just what you need to know to start getting useful output.

What Happens When You Type Something

You type a message and hit enter. Behind the scenes, ChatGPT breaks your text into tokens. A token is roughly three quarters of an English word. So a ten word sentence is about thirteen tokens.

Why does this matter? Because there's a limit to how much the model can process at once. The context window. For GPT-4o it's big enough to handle very long conversations or documents. But the useful limit for getting good responses is smaller than the technical maximum.

Practical implication: if you paste in a twenty page document and ask for a summary, the model might miss things near the end. Break long inputs into chunks. Start fresh conversations when switching topics. These habits keep the context clean and the responses accurate.

You don't need to understand tokens deeply. Just know that very long inputs can degrade quality and that's normal. It's not broken. It's just how the technology works.

Temperature Is the Knob Nobody Touches

There's a setting called temperature that controls how predictable or creative the output is. Think of it as a dial from zero to two. Lower means more predictable and factual. Higher means more varied and creative.

For writing code, extracting data, or answering factual questions, I set temperature low. Around 0.2. The responses become more consistent and accurate but also more repetitive.

For brainstorming, creative writing, or generating ideas, I set it higher. Around 0.8. You get more variety but also more randomness. Some outputs will be useless. Some will be surprisingly good.

The default setting works fine for general use. But if you're doing something specific and the results feel off, adjusting temperature is often the fix. It's available in the API and in some third party tools. Not in the regular ChatGPT web interface unfortunately.

The API Isn't as Scary as It Looks

I put off learning the API for months because I thought it required serious programming skills. It doesn't. If you can copy and paste a few lines of Python, you can use the API.

The basic idea is simple. Instead of typing into a web browser, your code sends the same kind of message to OpenAI's servers and gets a response back. The advantage is you can automate things. Send a hundred prompts programmatically. Integrate ChatGPT into your own app or website. Control every parameter precisely.

Cost wise it's cheap for light use. GPT-4o mini costs fractions of a cent per query. Even GPT-4o is affordable for personal projects. A typical conversation of a few back and forth messages costs less than a dollar.

Where it gets expensive is bulk processing. If you're summarizing thousands of documents, the costs add up. But for most people using the API for personal projects, we're talking about a few dollars a month. Maybe less than a coffee.

I'd say learn the API when you have a specific automation need. Until then, the web interface does everything you need.

Build a Custom GPT Before You Think You're Ready

This is my most controversial advice probably. Most people think custom GPTs are for advanced users. They're not. They're easier to build than they seem and they solve a real problem which is having to re-explain yourself every time you start a new conversation.

A custom GPT is just a saved set of instructions. You tell it who it is, how to behave, and what to avoid. Then every conversation with that GPT starts from those instructions. No repeating yourself.

The setup process walks you through it. You give it a name. You write some instructions in plain English. Optionally you upload files it can reference. That's basically it.

My most used custom GPT is for editing blog posts. Its instructions are something like: "You are a professional editor. When I paste in text, check for clarity, conciseness, and flow. Never change the author's voice. Suggest edits but always explain why. Keep suggestions brief." I use it multiple times a week.

Free Versus Paid: An Honest Take

The free version uses GPT-4o mini. It's fast and handles basic tasks like a champ. Writing emails. Summarizing text. Explaining concepts. Translating between languages. All fine on the free tier.

The paid version at twenty dollars a month gives you GPT-4o which is better at complex reasoning. If you're debugging tricky code, analyzing dense research, or doing creative work that requires nuance, the difference is noticeable. You also get image generation, web browsing, and custom GPTs.

So who should pay? If you use ChatGPT less than a few times a week, stick with free. If you're using it daily for work or serious projects, the twenty bucks is probably worth it. If you're not sure, use free for a month and see if you hit limitations.

I pay for it. But I also use it multiple hours every day. Your situation might be different.

FAQ

Q: Will ChatGPT replace my job?

Probably not. It's a tool that makes certain tasks faster. It writes drafts but not final copy. It suggests ideas but doesn't have taste. It explains concepts but doesn't have experience. The people who benefit most are the ones who treat it as an assistant, not a replacement for their own thinking.

Q: Why does ChatGPT sometimes refuse to answer?

OpenAI has built in safety restrictions. The model won't help with illegal activities, generating harmful content, or certain other prohibited uses. Sometimes it's overly cautious and refuses harmless requests. If that happens, rephrase your prompt. Adding more context about your legitimate use case often helps.

Q: How do I keep up with new features?

Honestly, you don't need to. The core functionality has been stable for a while. New features are nice but not essential. Focus on getting good at prompting. That skill transfers to every new model and every new feature they release.