ChatGPT Complete Guide: Features, Prompts and Use Cases

ChatGPT was released by OpenAI on 30 November 2022 as a free research preview. By February 2026 it had 900 million weekly active users, according to Wikipedia's account of its history (Wikipedia: ChatGPT). For many people it is the first, and sometimes only, way they use generative AI.

That reach hides a wide range of results. Some people get excellent work from it; others get confident nonsense and give up. Most of the difference comes from knowing what the product can do, how to ask, and when to check the answer. This guide covers all three.

What ChatGPT is

ChatGPT is a chat application built on OpenAI's GPT models. It runs on a freemium model: a free tier with limits, and paid plans with more capacity and features. The models behind it change frequently, and so do the plans, so check OpenAI's own pages for current limits and prices.

It is also more than a chat box. Over four years it gained tools that let it search, read files, write and run code, create images and act on the web.

The main features, and when they arrived

The dates below come from Wikipedia's ChatGPT article as read on 3 October 2026, which cites the original announcements and news reports.

Two more developments matter. Memory lets ChatGPT keep facts you ask it to remember across conversations. And from January 2026 OpenAI began testing advertisements in the free version for logged-in adult users in the US.

For developers, ChatGPT can be extended with plugins built from skills, MCP servers and optional interfaces (OpenAI: plugins). See MCP explained.

Prompts that work

OpenAI describes prompt engineering as writing instructions that make a model consistently produce what you need, and notes that its GPT models benefit from explicit instructions about how to do a task (OpenAI: prompt engineering). In the chat window, that comes down to five parts:

1. Goal: what you want and why.
2. Context: the facts, documents or data it needs. Paste or upload them rather than hoping it knows.
3. Audience: who the result is for.
4. Format: length, structure, tone.
5. Checks: what to avoid, and what to do when unsure.

Then iterate. The first answer is a draft: ask for changes, give an example of what you liked, or tell it what was wrong. For techniques that also apply when building with the API, see prompt engineering techniques.

Good uses, and risky ones

Strong fits: drafting and editing text, explaining concepts at your level, brainstorming, summarising documents you provide, writing and explaining code, turning notes into structure, practising interviews and languages.

Use with checks: facts, figures, names and dates; legal, medical and financial questions; citations and quotes; anything you will publish or act on. Use Search or Deep Research, follow the sources, and verify what matters.

Avoid: pasting confidential company data or other people's personal information without permission, and relying on it as the only reviewer of important work.

The limits to keep in mind

• Hallucination. It can state false things fluently. Research from OpenAI's own scientists argues that models guess when uncertain because training and evaluation reward guessing over admitting uncertainty (Kalai et al.: Why Language Models Hallucinate). See how to reduce LLM hallucinations.
• Sycophancy. Wikipedia's article reports that chatbots tend to agree with users even when they are wrong, and that OpenAI withdrew one 2025 update over it. Ask it to argue against your idea as well as for it.
• Knowledge cut-off. Its training data stops at a date. Search helps with recent events.
• Bias. Training data carries biases, which can surface in descriptions of people.
• Privacy. Users can opt out of their chats being used for training. Workplaces often have their own rules about what may be shared; follow them.

ChatGPT or the API?

ChatGPT is a finished product for people. The OpenAI API gives developers the models themselves, to build applications with their own interfaces, data and controls. If you find yourself pasting the same instructions into ChatGPT every day, that is a sign the task could become an application. See the LangChain tutorial. For using AI assistants inside an editor, see AI coding assistants.

Common mistakes

• One-line prompts for tasks that need context.
• Trusting figures and citations without checking them.
• Pasting confidential data into a personal account.
• Accepting agreement as validation.
• Never iterating: treating the first answer as final.

The OpenAI API, prompt engineering and building assistants of your own are covered in the Generative AI course.

Sources

• ChatGPT (Wikipedia) — Wikipedia
• Prompt engineering (OpenAI API documentation) — OpenAI
• Plugins (OpenAI Developers) — OpenAI
• Why Language Models Hallucinate (arXiv) — Kalai et al.