Hardly anyone can have missed the AI phenomenon that has taken the world by storm. Almost every major company has some kind of AI initiative now. Politicians talk about how important it is not to “fall behind in the AI race,” and hundreds of millions have started using AI chatbots.
The AI wave took off when OpenAI released its chatbot ChatGPT, which gives large language models a conversational interface. Many competitors have emerged since then, from both small and large companies, but ChatGPT remains the most popular.
ChatGPT isn’t necessarily the “best” AI service out there (recent tests show other tools can outperform it in certain areas), but it’s the most widely known and the benchmark that all other services are compared to. It’s the standard, plain and simple.
If you’ve never tried AI yourself, this guide is for you. Follow along, and I’ll show you how to get started.
Make an account and get started
You can try out ChatGPT’s simplest features without creating an account, but to access features like the ability to save conversations for later, you’ll need an account.
You can create your account in several ways, in the browser on your computer or in the ChatGPT app for iPhone or Android. For this guide, I have chosen the computer, but the app works the same way.
You can either create an old-fashioned account with an email address and password or log in with Google, Microsoft, or Apple to avoid having to remember another password.
Once you’re in, just start chatting. ChatGPT is conversationally trained, so you can ask questions, request help, or give prompts just like talking to a human. Responses also feel natural and ChatGPT “remembers” everything” (up to a point) that’s been typed in a conversation, so you can ask follow-up questions and reference things you’ve typed before without confusing the chatbot.
Should I subscribe?
One question many people ask is whether it’s worth paying for a ChatGPT Plus subscription or if the free version is enough. Here are the main advantages of a paid account:
- Access to the most advanced models (currently GPT-5)
- Better and faster image generation
- Other enhanced features
- More context (“memory”)
- Possibility to create customized chats, so-called GPTs
- Access to the Sora video generator
- Programming assistance with Codex
For real power users, there’s also a Pro subscription, but it costs $200 a month. You get minimal restrictions, the highest possible speed of response, and access to experimental new features.
My recommendation is that you start exploring ChatGPT without paying. Then, if you get curious about features that require a subscription, try it out for a few months. If you end up liking it, consider switching to an annual subscription, as it will be cheaper in the long run.
Managing chats, GPTs, and Projects
Foundry / Ashley Biancuzzo
The interface of ChatGPT is simple, with a column on the left where you can find the different features and a list of past chats, and a main window on the right, where the chats themselves are displayed.
At the top of the left column, there’s a quick button to open a new chat, a search bar for past conversations, and a Gallery feature, which collects images you’ve created with the chatbot. New users will also see a Projects button and Plus subscribers get an extra GPT button. Your previous chats are listed just below.
Projects is a new addition that allows you to collect multiple chats on a particular topic in one place and get a bit more organized. It basically works like folders and you can use it for all sorts of categories. Once you’ve created a project, you can give it its own icon in any color to make it easier to find if you have many.
Foundry / Ashley Biancuzzo
If you have a Plus subscription, you can also create GPTs. A GPT is a customized chat where each conversation starts with a set of instructions on how the AI should respond. You can upload files that the GPT has access to, choose a preferred model (for example, GPT-5 delivers stronger reasoning capabilities), and enable additional features like web search and image generation.
You can just talk to ChatGPT
If you prefer, you can talk to the ChatGPT instead of typing and reading the answers. The AI can recognize and respond in many languages, and you can choose your preferred spoken language in the settings.
Using ChatGPT as a search engine
Many people now use AI instead of traditional search engines like Google. Google even shows “AI summaries” at the top of the search results, and many users likely read those instead of scrolling past.
So it’s very common to use ChatGPT to find information of any kind–facts, news, recommendations, and so on.
Foundry / Ashley Biancuzzo
Doing so is easy and the results sound credible. The only problem is that AI is often wrong. Sometimes extremely wrong. According to a study commissioned by the BBC, ChatGPT is flawed in 81 percent of answers, with at least one serious error in a whopping 45 percent of all generated answers.
The less knowledgeable you are about a topic as a user, the less likely you are to spot errors and flaws in the AI-generated results yourself. That’s why you shouldn’t rely on what ChatGPT spits out without first checking more reliable sources. You can ask the chatbot to include links to sources and sometimes it will do so on its own, but it’s good to actually follow these and double-check.
ChatGPT and other AI chatbots don’t actually “know” anything. Their responses come either from external sources (like a search) or from the data the model was trained on. The more accurate information about a topic that’s in the training data, the more likely the results will be correct, but errors can still creep in. There’s no technical difference between a correct answer and a hallucinated one.
What ChatGPT is good (and bad) at
What ChatGPT and other AI chatbots using large language models are best at is generating answers that sound good–responses that feel believable and have a clear connection to what you asked for.
This in turn makes ChatGPT an excellent tool for improving and rewriting text, rewriting a text in a different style, translating, and much more where form is more important than content.
ChatGPT can also summarize long texts, give instructions, and answer complex questions on many topics, but you have to be careful with specific facts. Overall, it almost always gets it right, but individual numbers and other facts often get it wrong. With pictures and videos, it’s easy to see whether the results are good or not, but with text it’s more difficult.
The latest reasoning models are better at solving equations, though it’s still easy to trick them into making mistakes on purpose, and some problems they completely miss.
ChatGPT is good at generating and correcting errors in code, but it’s not great at building complete systems. You need to have the ideas and the big picture, otherwise it’s easy to pancake.
It costs nothing extra to try again
Was the result you got from ChatGPT when you asked for something not very good? Try again! For regular text chat, there’s no limit to the number of times you can ask it something. By rephrasing or adding additional details, you can often get the AI to produce significantly better results.
To understand why, you need to know a bit about how the AI generates its answers. Large language models are extremely advanced word guessing machines. They take an input in the form of text, turn it into a string of numbers, and then use a huge statistical model to work out what the continuation is likely to be, one piece at a time.
The AI has no knowledge, cannot distinguish between fact and fiction, and has no real understanding of anything. But it’s trained on vast amounts of existing text and conversations between humans. Since humans mostly speak coherently and stick to the topic at hand, the AI is likely to do the same. Adding more details increases the likelihood that the generated answer will be appropriate.
Prompt engineering can give better results
Systematically tweaking and refining prompts to ChatGPT has become a new skill called “prompt engineering.” There are thousands of videos on YouTube with tips on how to get better results, whether you’re writing code or generating images. Dozens of books have already been published on the subject.
The basic idea is that some prompts, on average, lead to better answers. For example, adding a phrase like “now it’s time to think hard and rethink all assumptions” when brainstorming a problem can make ChatGPT’s response more creative.
Since others have spent a lot of time finding such tricks, you can get a lot out of trying tips from videos, books, and articles.
Watch out for fart fatigue
Experimentation should be fun and inspiring, but it’s easy to get carried away with fine-tuning.
“This is good, but maybe it could be even better if I just change this word here and that word there…”
If you find that you’ve started to think like this and that your conversations with ChatGPT don’t feel fun and mentally relieving, you may be suffering from fart fatigue.
This is a new phenomenon. The goal of introducing AI is to save time, letting employees get more done, but in many cases they end up spending as much time fine-tuning their prompts as they previously spent doing the work themselves.
Privacy and security
Foundry / Ashley Biancuzzo
ChatGPT isn’t like a noticeboard in a town square where anyone passing by can see what you’ve written. But it’s also not a completely private service, so don’t share your deepest secrets on there. Your conversations aren’t encrypted, so there’s a chance OpenAI staff can see them. The likelihood of your chats being read is low, but it can happen.
In addition, if you don’t change the privacy settings, OpenAI can train its models on all your conversations with ChatGPT. You can turn this off in Settings -> Data Controls -> Improve modelling for everyone. But even if you do, you should avoid sharing personal data and other sensitive information, as you can never be sure it won’t leak.
