What Is ChatGPT

Understand ChatGPT Before Using It

Before diving into the features, take the time to understand what ChatGPT really is and how it works. This isn’t an academic detour: most new users end up disappointed because they misunderstand the tool they’re talking to. They ask it to retrieve specific information when it excels at generating new content, they’re surprised when it remembers or forgets a detail from one conversation to the next, and they take made-up figures at face value. Understanding the mechanism tells you where to place your trust.

ChatGPT is OpenAI’s AI assistant. You write to it, or speak to it, in natural language, just like a colleague, and it responds. There’s no syntax to learn and no menus to explore: you formulate your request in plain English, and the conversation does the rest.

You use it on the web at chatgpt.com, in the iOS and Android apps, and in the desktop app for macOS and Windows. In supported countries, including those in Europe, you can even try it without an account for a single conversation. History, data export, and customization require an account: that’s the subject of lesson 2.

How ChatGPT Works

ChatGPT relies on large language models, or LLMs (Large Language Models), trained on vast volumes of text. When you ask a question, the model doesn’t fetch a ready-made answer from a database: it generates it, piece by piece, by predicting the most probable continuation based on your request and everything that came before in the conversation. Current models can also reason before answering and use tools: web search, code execution, file reading.

Three practical consequences follow from this, and they explain most beginner surprises. First, the model’s own knowledge stops at a cutoff date. For current events, ChatGPT launches a web search when the question requires it: this tool is enabled by default on all plans, and the response then cites its sources (lesson 11). Second, its memory has two levels. Within a conversation, it keeps the entire thread. From one conversation to the next, it relies only on what the Memory function retained, if it’s enabled for your account, and you can consult, correct, or turn it off in settings. Finally, it can be wrong with perfect confidence. OpenAI calls these “hallucinations”: false definitions or dates, invented citations and references, overconfident answers on ambiguous questions. This is the subject of lesson 16.

What ChatGPT Can Do

The scope of use is broad, and this course covers the essentials. Writing first: emails, reports, messages, from first draft to review, with drafts you can edit directly in the conversation. Analysis next: summarize a long document, extract key points, compare options, make a spreadsheet talk through data analysis. Translation, with tone and register adaptation. Research, on the web with citations, or in depth with deep research, which crosses many sources and delivers a sourced report. Programming, to write, explain, or debug code. Creating and editing images (lesson 15). Reading files and audio: PDFs, spreadsheets, presentations, photos, recordings to transcribe. Oral conversation finally, with voice mode (lesson 13).

With GPT-6, a response can also mix text, visuals, and interactive elements, like a chart, a comparison table, buttons, or a checklist, when that’s clearer than a long paragraph. OpenAI calls this operation Intelligent UI. A plain text response remains possible, and you don’t need to ask for anything special.

Three Ways to Work

Since summer 2026, ChatGPT brings together three experiences. Chat is for questions, research, and quick exchanges: that’s the subject of this course. Work is an agent designed for long tasks, in multiple steps, that result in a deliverable: document, spreadsheet, presentation, report, or site. Codex remains dedicated to software development. In the desktop app, a selector switches from ChatGPT to Codex, then from Chat to Work.

What ChatGPT Is Not

Knowing what it doesn’t do will save you just as much time. It’s not a classic search engine: even when it searches the web, it summarizes and reformulates, and a synthesis can distort its source. Open the citations that matter. It’s not an oracle: its responses are plausible, not guaranteed, and a confident tone is not a guarantee of reliability. It’s not an AI text detector: OpenAI specifies that it cannot reliably say whether a text was generated, or if it wrote it itself. It’s not a human, finally: it models language, doesn’t live your situation, and ignores everything you don’t tell it.

Your Turn

First contact. Open chatgpt.com or the mobile app, with or without an account, and send:

“Explain in three sentences what you are and what you can do.”

Follow up with a question, without recalling any context:

“What are your main limitations?”

It answers the second question taking the first into account: you just used conversation context, the tool’s most useful mechanism. Finally ask a current events question, for example about an event this week, and spot the citations that signal a web search. Remember the work rule: same subject, same conversation; new subject, new conversation.

Official sources: What is ChatGPT: FAQ (OpenAI help center: access, first conversation) · ChatGPT capabilities overview (tools and capabilities) · ChatGPT Work and Codex (the three experiences) · Intelligent UI in ChatGPT · Does ChatGPT tell the truth? (hallucinations, cutoff date, web access)

Key Takeaways

• ChatGPT is OpenAI’s AI assistant, on chatgpt.com, on mobile, and on desktop, with or without an account
• It generates its responses by predicting the most probable continuation; it doesn’t copy them from a database
• Web search, enabled by default, complements knowledge that stops at a cutoff date
• Chat for conversing, Work for long tasks and deliverables, Codex for code
• It can be wrong with confidence: verify anything that involves a decision