--- id: chatgpt-giving-chatgpt-context lang: en track: chatgpt title: "Giving ChatGPT context" seoTitle: "Giving ChatGPT context and managing memory" description: "How to give ChatGPT context with uploads, custom instructions and a short brief, and how to manage the memory it quietly builds up about you over time." part: getting-started partTitle: "Getting started with ChatGPT" partOrder: 2 order: 5 level: Beginner durationMinutes: 5 outcome: "Give ChatGPT the right context, and check and steer what its memory keeps about you." summary: "ChatGPT remembers more than most tools: past chats, saved memories, your instructions. Give it the right context on purpose, and check what it remembers so the context stays right." draft: false updated: "2026-10-14" furtherReading: - title: "Memory in ChatGPT" url: "https://help.openai.com/en/articles/8590148-memory-faq" note: "How memory works, the memory summary, and how to turn it off." - title: "Customizing your ChatGPT personality" url: "https://help.openai.com/en/articles/11899719-customizing-your-chatgpt-personality" note: "Base style and tone, next to your custom instructions." - title: "Uploading files and audio to ChatGPT" url: "https://help.openai.com/en/articles/8555545-file-uploads-faq" note: "File types and size limits." - title: "Projects in ChatGPT" url: "https://help.openai.com/en/articles/10169521-projects-in-chatgpt" note: "Keeping context together for ongoing work, covered in Part 4." keywords: - ChatGPT context - ChatGPT custom instructions - ChatGPT memory - ChatGPT file upload - how to give ChatGPT context - ChatGPT personalization quiz: intro: "Five quick questions. Ace them all and you've finished Part 2." passScore: 5 questions: - q: "How is ChatGPT different from a tool that starts every chat knowing nothing about you?" options: - "It doesn't; every chat starts from zero" - "It reads the files on your computer by default" - "It only remembers what you said earlier inside the same project" - "With memory on, it draws on past chats and saved details" answer: 3 explain: "When memory is on, ChatGPT decides what is relevant from past chats, saved memories, custom instructions, files, and connected apps. That is handy, and it is also why you should check what it remembers." - q: "You want your role and preferred tone to apply to every chat. Where do you put them?" options: - "Typed out at the top of each new chat, every single time you start" - "In custom instructions, under Settings, Personalization" - "In your profile name and account settings" - "In one big project that you open for every task" answer: 1 explain: "Custom instructions are direct guidance you write once and ChatGPT applies everywhere. Memory builds context over time, but instructions are the part you control word for word." - q: "An answer seems shaped by something from an old chat that is no longer true. What do you do?" options: - "Check the memory source, then fix or delete that memory" - "Delete the old chat; the memory goes with it" - "Switch to Temporary Chat for every conversation from now on" - "Raise the thinking level so it reasons past it" answer: 0 explain: "Memory sources show what influenced a response. You can correct or delete a memory, tell ChatGPT not to mention something again, or turn memory off. Deleting a chat alone does not delete a saved memory." - q: "You want a one-off chat that ignores your memory and does not add to it. What do you use?" options: - "A new project with project-only memory" - "Instant, with web search switched off" - "Temporary Chat, set to Unpersonalized first" - "A normal chat after deleting your memories" answer: 2 explain: "Temporary Chat does not save to your history or create memories, and the Unpersonalized option skips your existing memories and instructions. Choose it before the chat begins." - q: "Before you upload a confidential client file, what should you check?" options: - "That memory is off, so nothing is remembered" - "That you're in the company Business workspace" - "That the file is under the upload size limit" - "That you've started a Temporary Chat" answer: 1 explain: "The personal-versus-business line from lesson 1.4 still holds. Business workspace data is not used for training; personal plans train by default unless you opt out. Confidential work belongs in the workspace." pass: badge: "Part 2 complete" title: "You know your way around ChatGPT now." body: "Full marks, and that's Part 2 done. You can pick a plan, find your way around, choose a thinking level, keep your work in Pages and Sites, and steer the context ChatGPT uses. Next comes the big one: Part 3, prompting." retry: "Check the marked answers and give it another go. You've got this." --- > **Level:** Beginner · **Reading time:** ~5 min > > **What you'll be able to do:** give ChatGPT the context it needs, and check and steer what its memory keeps about you. Ask a cold assistant to "write a follow-up email" and you get something generic. Tell it *you are a logistics account manager, the client is annoyed about a late delivery, keep it warm but firm, three short paragraphs*, and you get something you could almost send. **Same request, night-and-day result, and the only thing that changed was the context.** ChatGPT adds a twist. Unlike a tool that starts every chat from zero, **ChatGPT remembers.** With memory on, it can draw on your past chats, things it saved about you, your custom instructions, your files, and connected apps. That is a head start, and it is also a responsibility: context you never meant to give can shape an answer too. This lesson covers both halves, giving context on purpose and keeping an eye on what it remembers. ## Give it the material The most direct context is the stuff you already have. **Upload it** with the **+** button in the message box: PDFs, Word and PowerPoint files, spreadsheets, text, and images. The limits are generous (files up to 512 MB, documents up to about 2 million tokens, spreadsheets around 50 MB, images 20 MB), so the real report, the actual spreadsheet, the brief you were sent all fit. If an answer depends on a document, **hand ChatGPT the document.** Do not retype it or describe it from memory. Give it the real thing and ask your question about it. On formats, lean wins. Plain text and **Markdown** carry the same words as a heavily styled Word file with none of the overhead, so when you control the format and will reuse the material, a `.md` or `.txt` file is the cleanest choice. When you do not control it, upload what you have. Any source beats no source. ## What ChatGPT already remembers With memory switched on, ChatGPT keeps two kinds of context. **Saved memories** are details you asked it to keep ("remember that I write for a non-technical audience") or that it saved as useful. **Chat history** lets it pull in relevant bits of earlier conversations, and that picture updates as you go. ChatGPT decides what is relevant for each answer. You stay in charge, and three habits keep it that way: **Look at the sources.** Under an answer, memory sources can show what influenced it. If something old or wrong crept in, you will see it there. **Correct or delete.** In **Settings, Personalization, Memory** you can read a summary of what ChatGPT keeps, correct it, delete items, or turn memory off. Note that deleting a chat does not delete a saved memory that came from it; delete the memory too. **Go clean when you need to.** A **Temporary Chat** is not saved and does not create memories. Choose **Unpersonalized** before you start and it also ignores your existing memories and instructions. Handy for a one-off task that should not colour the rest.
Two sources of context On the left, what ChatGPT already remembers: saved memories, chat history, custom instructions, and files and connected apps; check it under Settings, Personalization, Memory. On the right, what you give it on purpose: this message, this project, and your brief. The more often you would retype something, the wider you should set it. Two sources of context What it remembers Saved memories Chat history Custom instructions Files and connected apps Check it now and then: Settings › Personalization › Memory + What you give it This message A one-off: paste or attach. This project Ongoing work: Part 4. Your brief Custom instructions, every chat. Retype it often? Then set it wider.
## Context isn't unlimited All of this lands in ChatGPT's **context window**: what it can hold in mind while it answers. It is large but **finite**. The files you add, the conversation so far, and the reply being written all count. Long, wandering chats and heavy files fill it, and detail starts to slip. So the habits from [lesson 2.2](/learn/chatgpt/getting-started/first-steps-in-chatgpt) pay off twice: **one topic per chat**, a fresh chat when you switch tasks, and lean material. Part 4 opens with the full mechanics. ## Three places to put context Context can live at three levels, from "just for now" to "for good". Picking the right one saves you from repeating yourself. **This message.** Paste or upload context where you need it. It lasts for this chat. Perfect for a one-off: "here's the report, summarise it." **This project.** A **project** groups chats, files, and instructions for a piece of ongoing work, and its instructions apply to every chat inside it, overriding your general ones. That is a Part 4 topic. **Everything.** Your **custom instructions** (Settings, Personalization) apply to every chat. Next to them, **Base style and tone** sets the overall voice, from Professional to Efficient. This is where your durable facts live: your role, how you like answers, terms your sector uses. **Rule of thumb: the more often you would retype something, the wider you should set it.** ## Write a short brief once The best ten minutes you will spend in ChatGPT: write a **short brief about how you work** and put it in your custom instructions. A handful of lines that answer who, what, and how. Something like: *"I'm a marketing lead at a mid-size logistics firm. I mostly write client updates and internal briefs. My readers are busy and non-technical. I like a warm, plain tone, short paragraphs, no jargon. Never invent figures; ask me if you're unsure."* Paste that once and every answer starts closer to right. **One caution, straight from [lesson 1.4](/learn/foundations/ai-at-work-safety-and-privacy):** context is powerful, but the rules still apply. Confidential work belongs in your company's **Business workspace**, where it is not used for training. On a personal account it is, unless you switch training off. ## Where this goes next **Context tells ChatGPT who it is helping; the prompt tells it what to do.** Part 3 is all about the prompt. In Part 4 your brief grows into projects and skills, reusable ways to give ChatGPT context and behaviour on tap. For now the habit is enough: never make ChatGPT guess what you could simply tell it, and now and then, check what it thinks it knows. ## Try it Write your own brief, about your real role, in under a hundred words. Open **Settings, Personalization** and paste it into your custom instructions. Then open the **Memory** summary in the same place and read it. Is anything outdated or wrong? Correct or delete it. Finally, start a fresh chat and ask for something you actually need, like a short update to your team. Notice how much closer the first answer lands, and check the memory sources to see what shaped it. *(General exercise. Role and industry versions come once accounts are in.)* ## Key terms [Context](/glossary#context), [Context window](/glossary#context-window), [Custom instructions](/glossary#chatgpt-custom-instructions), [Memory](/glossary#chatgpt-memory), [Temporary Chat](/glossary#temporary-chat), [Markdown](/glossary#markdown), [Project](/glossary#chatgpt-project)