How to Use ChatGPT: Tips, Prompts, Custom GPTs & API Guide for Beginners

The thing that annoys me about most ChatGPT tutorials is they assume you already understand the jargon. Tokens. Context windows. Temperature. System prompts. If you're brand new to this, that stuff is intimidating and mostly irrelevant to actually using the tool.

So I'm going to describe how I actually use ChatGPT in my daily life. Not the theory. Not the academic frameworks. Just what I type and what I get back and why it works.

Monday Morning: Writing a Difficult Email

Here's something that happened last week. I needed to tell a client we were going to miss a deadline. Not a fun email to write. I stared at a blank Gmail compose window for about ten minutes before opening ChatGPT.

My prompt was something like this: "I need to email a client to tell them their project will be delayed by about two weeks. The reason is that a key team member had a family emergency and was out unexpectedly. I want to sound professional but human. Not corporate. Acknowledge the inconvenience but don't grovel. Offer to hop on a call to discuss timeline adjustments. Keep it under 200 words."

What came back wasn't perfect. The first draft was a little too formal. I asked for a revision with "more casual, like you're talking to someone you've worked with for a year." The second draft was close. I edited maybe three sentences and sent it.

Total time including iterations was maybe four minutes. Compare that to the ten minutes I had already spent just staring at the blank email and the difference is significant.

Tuesday: Understanding Something Technical

I was trying to understand a pull request that touched code I wasn't familiar with. Specifically some database migration logic that I didn't write and couldn't fully follow.

I pasted the relevant code into ChatGPT and asked: "Explain what this code does. I'm a frontend developer who doesn't know much about database migrations. Walk me through it line by line and tell me if anything looks risky."

The explanation was clear enough that I understood the changes. More importantly, it flagged two things that looked questionable. One was a real issue that I brought up in the code review. The other was a false alarm but at least it made me think about it.

This is the kind of thing ChatGPT is genuinely good at. Not writing production code from scratch. But explaining existing code, flagging potential issues, and helping you understand things outside your area of expertise.

Wednesday: Meal Planning Without Thinking

Wednesday evening I had no idea what to make for dinner. I opened ChatGPT and typed: "I have chicken thighs, broccoli, eggs, parmesan cheese, pasta, and various condiments. Give me three dinner ideas that use at least two of these ingredients each. Under 30 minutes cook time. Nothing that requires a blender or food processor."

It suggested a pasta dish with chicken and broccoli, a frittata with parmesan, and chicken thighs with a parmesan crust and roasted broccoli. I made the pasta. It was fine. Nothing spectacular but it solved the decision paralysis that was the real problem.

This is the kind of low stakes task where ChatGPT shines. You don't need high accuracy or creative brilliance. You just need someone or something to break the logjam and give you options to choose from.

The Weekend Project: Building a Custom GPT

On Saturday I spent about twenty minutes building a custom GPT. The task it handles is boring and repetitive. Every time I publish a new blog post, I need to write social media promotion copy for three different platforms.

The custom GPT instructions are straightforward. It knows my blog's topic and tone. It knows the format requirements for each platform. It knows to ask me for the post title and a one sentence summary before generating anything.

Now instead of spending fifteen minutes writing social copy, I spend about two minutes giving it input and maybe three minutes editing what comes back. The quality is about the same but the time investment is dramatically lower.

This is the kind of thing custom GPTs are actually useful for. Not building a general purpose assistant. But automating one specific repetitive task that has a consistent format.

What I Still Get Wrong After Two Years

Even after all this time, I still make mistakes.

Sometimes I'm too vague. I type three words and hope for magic. It never works but I keep doing it when I'm in a hurry.

Sometimes I forget that ChatGPT doesn't know my context. I reference things from previous conversations that it can't access. I get frustrated by responses that miss the point without realizing the missing context is my fault.

Sometimes I trust output that I should verify. A statistic that sounds right. A technical explanation that seems plausible. A code snippet that looks correct at a glance. I've published things with errors because I trusted the model when I shouldn't have.

The lesson I keep relearning is that ChatGPT is a tool that amplifies what you put into it. If you're lazy with your prompts, you get lazy output. If you're careful and specific, you get useful output. There's no shortcut around that.

FAQ

Q: What's the minimum I need to know to start using ChatGPT?

You need to know how to type. That's genuinely it. Sign up at chat.openai.com, type something in the box, read the response. Everything else you can learn gradually as you go.

Q: How do I get better at writing prompts?

Use it every day for real tasks. Pay attention when the output is bad and try to figure out why. Was your prompt too vague? Did you not specify format? Was the task itself something ChatGPT isn't good at? You'll develop instincts over time that are better than any formal framework.

Q: Is ChatGPT going to keep getting better?

Probably. The technology is improving fast. But the fundamental skill of giving clear instructions to an AI is going to remain relevant regardless of how good the models get. Better models still need better prompts to produce their best work.