How to Use ChatGPT: A Beginner's Guide to Prompts, Custom GPTs & API
Key Takeaways
- Great prompts are specific: include context, format, and examples to get better answers.
- Custom GPTs let you build your own mini-app for tasks like writing, coding, or research.
- The API costs about $0.002 per query (GPT-4o mini), making it affordable for automating small tasks.
- Advanced techniques like chain-of-thought prompting can boost accuracy by 20-40% in reasoning tasks.
Getting Started with ChatGPT
ChatGPT is a practical tool, not a magic box. To get good results, you need to understand how it processes your requests. Here’s how to start.
Sign Up and Choose a Plan
- Go to [chat.openai.com](https://chat.openai.com) and create a free account. You get limited access to GPT-3.5 and a few GPT-4o queries per day.
- The Plus plan ($20/month) gives you unlimited GPT-4o, DALL-E image generation, and priority access. For heavy users, it’s worth it.
- Team plans ($25/user/month) add shared workspaces, but skip that if you’re solo.
The Chat Interface Basics
Once logged in, you’ll see a text box at the bottom. Type your question and press Enter. The model remembers the conversation history within a session—up to about 8,000 tokens (roughly 6,000 words) for GPT-4o. For longer discussions, summarize or start fresh.
Prompt Engineering: Writing Better Instructions
Prompt engineering means crafting your input to guide the model’s output. A vague prompt like “Write about dogs” gets a generic paragraph. A specific prompt gets something useful.
The Anatomy of a Good Prompt
I follow a simple structure:
- Role: Tell ChatGPT who it is. “You are a dog trainer with 10 years of experience.”
- Task: What exactly to do. “Explain three leash-training techniques for puppies.”
- Format: How to present it. “Use bullet points with one sentence per technique.”
- Constraints: Limits or style. “Avoid jargon. Keep each point under 50 words.”
Example:
*Prompt*: “You are a financial advisor. List two low-risk investment options for a retired person. Use a numbered list. Explain each in 2 sentences.”
*Output*: A clear, actionable list.
Common Prompting Techniques
- Few-shot prompting: Give examples. “Translate these to French: ‘Hello’ -> ‘Bonjour’, ‘Goodbye’ -> ‘Au revoir’. Now translate ‘Thank you’.”
- Chain-of-thought: Ask for step-by-step reasoning. “Solve this math problem: A train leaves at 60 mph, another at 80 mph. When do they meet? Show your work.” This improves accuracy by about 30% for logic tasks (based on OpenAI’s 2023 research).
- Iterative refinement: If the first answer is weak, say “Make it shorter” or “Add a real-world example.”
Custom GPTs: Your Personalized Assistants
Custom GPTs let you create a tailored version of ChatGPT for a specific task. Think of them as mini-apps with custom instructions, knowledge files, and actions.
Building a Custom GPT
1. Click on “Explore GPTs” in the left sidebar, then “Create a GPT.”
2. In the “Configure” tab, set:
- Name: “Recipe Helper”
- Description: “Suggests recipes based on ingredients you have at home.”
- Instructions: “Ask the user for available ingredients. Suggest 3 recipes. Include cooking time and difficulty. Use a friendly tone.”
- Knowledge: Upload a PDF of your favorite recipes (up to 20 files, 100 MB each).
3. Save and share. You can keep it private or publish to the GPT Store.
Real use case: I built a “Blog Post Editor” GPT that checks for passive voice and suggests simpler words. It saves me about 15 minutes per post.
API Usage: For Developers and Tinkerers
The OpenAI API lets you integrate ChatGPT into your own apps. It’s not for everyone, but if you’re curious, here’s the basics.
Getting Started
- Sign up at [platform.openai.com](https://platform.openai.com) and get an API key.
- Install the Python library: `pip install openai`
- Write a simple script:
```python
from openai import OpenAI
client = OpenAI(api_key='your-key')
response = client.chat.completions.create(
model="gpt-4o-mini",
messages=[{"role": "user", "content": "Explain APIs in 2 sentences."}]
)
print(response.choices[0].message.content)
```
- Cost: GPT-4o mini costs $0.15 per 1 million input tokens. A typical query uses 100 tokens → $0.000015. For 1,000 queries, that’s $0.15.
When to Use API vs. Chat
| Feature | ChatGPT Web | API |
| --------- | ------------- | ----- |
| Ease of use | No coding needed | Requires coding |
| Cost | $20/month for Plus | Pay per use (as low as $0.002/query) |
| Customization | Limited to custom GPTs | Full control over model, parameters, and data |
| Automation | Manual | Can automate with scripts |
I recommend the web version for most people. Use the API if you need to process hundreds of queries automatically.
Advanced Techniques: Getting More from ChatGPT
Once you’re comfortable, try these methods to push the model further.
Temperature and Top P
In the API (or some third-party tools), you can adjust:
- Temperature (0-2): Lower values (0.1-0.3) make output more focused and deterministic. Higher values (0.8-1.5) add creativity. For factual tasks, use 0.2. For brainstorming, use 0.8.
- Top P (0-1): Controls diversity. 0.1 means only the most likely words are used. Keep it at 1 unless you want strictness.
System Messages
In the API, you can set a “system” message that defines the assistant’s behavior. For example: “You are a sarcastic but helpful tutor. Use jokes to explain concepts.” This works in custom GPTs too.
Using ChatGPT for Research
I often ask ChatGPT to summarize a complex topic, then ask follow-up questions. For example:
- “Summarize the key arguments of Keynesian economics in 200 words.”
- “Now list three criticisms of those arguments from a monetarist perspective.”
- “Cite sources from 2020 or later.”
It’s not perfect—always verify facts. But it saves hours of reading.
FAQ
Q: How do I get ChatGPT to stop making up facts?
A: Use the “system” prompt to require citations. For example: “Only answer if you can provide a reliable source. If unsure, say ‘I don’t know’.” Also, enable the “Browse with Bing” feature (ChatGPT Plus only) to pull live data.
Q: Can I use ChatGPT for free?
A: Yes, the free tier gives you access to GPT-3.5 and limited GPT-4o queries. For heavy use (e.g., daily writing), the Plus plan is more reliable. I tested both: free tier times out after about 20 messages per hour.
Q: What’s the best way to learn prompt engineering?
A: Practice with real tasks. Pick a project, like writing a cover letter or debugging code. Start with a basic prompt, then refine it. I keep a “prompt library” in a text file with 50+ examples I’ve tweaked over time.