Before you start
You need a ChatGPT account, in a browser or the app. A prompt is the question or instruction you type into the chat window, and every step below starts with one.
On a work account, accept the invite link your employer sent and sign in with your company's single sign-on if asked. Keep the two apart: sign out of your personal account at work.
The menu names here come from OpenAI Academy's ChatGPT fundamentals guide and its companion guides, read on 5 November 2026. Plans differ in what they include, and OpenAI moves buttons often, so yours may sit elsewhere.
Your first chat in five steps
Start with a job you already do, so you can judge the answer. The prompt below is ours, built on the Academy's starter prompt that asks for use cases in your role as a table.
I run the front desk of a ten-person dental clinic.
Suggest five tasks from my week that ChatGPT could help with.
Show them as a table with three columns: the task, a prompt I could paste, and what I should check before I use the result.- Open ChatGPT and sign in. A new, empty chat is already waiting.
- Paste the prompt and send it. It names your job and the shape you want back. Leave both out and you get a general list that fits nobody.
- Change the answer in the same chat. Reply "make the prompts shorter" or "turn row two into an email". The chat keeps what came before, so you never repeat the setup.
- Pick a slower mode for harder questions. When the guide was last updated, on 29 May 2026, the default model decided by itself when to think longer. The model picker at the top also offered Fast for instant answers, Thinking for deeper reasoning and Pro for what it calls research-grade intelligence.
- Start fresh for a new topic. Click New Chat in the sidebar. Past chats stay in the left panel, grouped by date.
Custom instructions and memory
Type "What do you remember about me?" in any chat. ChatGPT lists what its memory holds: details it saved from earlier conversations so it can use them again. Type "Forget that" to drop something it has just saved.
For the full list, open Settings, then Personalization, then Memory, and choose Manage memories to review items, delete them or clear everything. Two switches sit there, Reference saved memories and Reference chat history. Turn one off and ChatGPT stops drawing on it.
Custom instructions work the other way round: you write them, and they stay until you change them. Open Settings, Personalization, Custom Instructions, turn on Enable for new chats, and fill in two fields. One says what ChatGPT should know about you, the other how it should respond. Select Save. The newer customizing guide gives a shorter way in: click your profile, then Personalize.
A stale memory is the one to watch. Change jobs in March, and a saved line about your old role keeps shaping answers until you delete it. Custom instructions do the job a system prompt does for developers: standing rules that apply before your first message.
| Level | Where you set it | Reaches | Good for |
|---|---|---|---|
| Custom instructions | Settings, Personalization | Your new chats | Your role, your tone, the format you like |
| Project instructions | Inside a Project | Chats in that Project | One client or one report, with its own style |
| A line in the prompt | The chat box | That chat only | A one-off, such as "write this as a team Slack update" |
Keep ongoing work in a Project
A Project is a space that gathers the chats, files and instructions for one piece of work. Choose Projects in the left-hand menu, create a new project and give it a name. Then add context: upload files, set instructions and link the resources the work needs.
A monthly sales summary is the classic fit. Put last month's report and your format rules in the Project once, and every new chat inside it starts from them. On a work account, Share invites teammates by email, and the project instructions tell them how ChatGPT answers in that shared space.
Business and Enterprise workspaces with memory on can choose project-only memory when they create a project. Chats then draw on other chats in the same project and ignore your saved memories. One client's details stay out of another client's answers.
Memory reads past chats
Memory can draw on all your past chats, not only the items it saved, to make answers more personal.
Source: OpenAI (@OpenAI) on X: memory references past chats (read November 5, 2026)
Shared projects
Teammates can work in the same project. Only its owner can invite or remove them.
Source: OpenAI Academy: Projects (read November 5, 2026)
A more capable memory
A new memory system starts rolling out in ChatGPT, the result of OpenAI's research into memory.
Source: OpenAI (@OpenAI) on X: a more capable memory system (read November 5, 2026)
Files, images and canvas
Click the paperclip icon, attach a file, then give it a task: "Summarize the main findings in this report" or "Visualize this sales data by region". Ask for a table or a chart, and download what ChatGPT makes.
For a picture, write "create image" in your prompt or choose it in the toolbar. Detail up front saves rounds later. Then steer it in plain words, such as "make it brighter" or "simplify the background", and highlight one area to change only that part. A picture of a real person needs a reference photo and that person's permission.
Canvas is an editing space beside the chat for longer text or code. Type /canvas or click the + icon and choose Canvas, paste your draft, highlight a paragraph and ask for a rewrite. Version history brings back an earlier draft.
Questions about today's facts need web search and a check of every source, which the research lesson walks through for ChatGPT, Claude and Gemini alike.
What Sam Altman and OpenAI posted about ChatGPT memory
Sam Altman is a co-founder of OpenAI and its CEO, as OpenAI's forum introduces him. On 10 April 2025 he posted on X that memory could now reference all your past conversations. He called it a surprisingly great feature, and a step toward AI that gets to know you over your life.
Fourteen months later, OpenAI's own account announced a more capable memory system, the result of research into how memory carries context from one chat to the next. Both posts sell memory as something that grows with you.
Decide per task. Keep memory on for your own drafting, where your style is worth remembering. A client's data is different. On Business and Enterprise, give that client a Project with project-only memory, so nothing it learns there reaches your other work.
We’ve been researching new ways for ChatGPT memory to carry context across conversations and keep it useful over time.
Screenshots of X's own embeds of the posts, taken on 5 November 2026, with the time in UTC. X cuts OpenAI's post short with Show more. Select one to open the post on X.
Common mistakes
- Pasting a client's private file because the chat feels private. Check your company's rules first, and read what AI apps keep before you do.
- Letting one long chat carry every topic. Earlier messages keep steering new answers, so a new subject gets a new chat.
- Asking for "a logo like Nike's". The image guide suggests asking for generic or ownable designs instead of copies of existing brands.
- Writing custom instructions as a wish list. Two or three plain rules, such as "UK spelling" and "short paragraphs", are easier to see working.
Try it yourself
- Fill in both custom instruction fields with three lines each. Open a new chat, ask for a short email and compare it with one from before.
- Open Manage memories and remove one entry that is out of date.
- Make a Project for one recurring task, add one file and two lines of instructions, and run two chats in it.
- Attach a spreadsheet, ask for one chart, then check one number in the chart against the file.
Next: Claude Projects
Claude has Projects too, each with a knowledge base, instructions and its own memory. How to use Claude Projects sets one up in five steps, and the habits from this lesson carry straight over.
Check yourself
You want ChatGPT to know your job and your tone in every new chat. Where do you set that?
Show answer
B. In custom instructions, under Settings and Personalization Custom instructions apply to new chats once Enable for new chats is on, so you write them once.
ChatGPT keeps using a job title you left months ago. What fixes it?
Show answer
A. Open Manage memories and delete that item Memory holds details from earlier chats, and Manage memories lets you review and delete each one.
You write the same monthly report with ChatGPT. What keeps last month's file and your format rules ready for every new chat?
Show answer
D. A Project with the file and instructions added A Project gathers the chats, files and instructions for one piece of work, so each new chat in it starts from them.
An image is right except for one cluttered corner. What is the quickest fix?
Show answer
C. Highlight that area and say what to change The image guide says you can edit a specific portion of an image by highlighting it.
Frequently asked questions
What is the difference between custom instructions and memory in ChatGPT?
Who writes them. Custom instructions are your words, fixed until you edit them. Memory is what ChatGPT picks up from your chats, so it grows as you talk, and each item can be reviewed or deleted.
Can I use ChatGPT without it remembering the chat?
Yes. Start a Temporary Chat, which ignores memory. If you want memory off everywhere, turn off both switches on the Memory page in Settings.
What files can I upload to ChatGPT?
Most office and image files. OpenAI's files guide names CSV, XLSX, PDF, DOCX, JPEG, PNG and TXT, among others, and gives no size figure. For a large file, split it or ask about one section at a time.
Who can see a ChatGPT chat I share?
On a work account, anyone in your workspace with the link, and they see messages you add later too. Enterprise workspace owners can limit sharing to the workspace or turn it off.
Do I have to credit OpenAI for an image ChatGPT made?
No. OpenAI's image guide says crediting it is optional. Your organization's guidelines and OpenAI's usage policies still apply to how you use the picture.
Sources
- OpenAI Academy: ChatGPT fundamentals https://academy.openai.com/public/clubs/work-users-ynjqu/resources/chatgpt-basics
- OpenAI Academy: Customizing ChatGPT https://academy.openai.com/public/clubs/work-users-ynjqu/resources/customizing-chatgpt
- OpenAI Academy: Projects https://academy.openai.com/public/clubs/work-users-ynjqu/resources/projects
- OpenAI Academy: Working with files https://academy.openai.com/public/clubs/work-users-ynjqu/resources/working-with-files
- OpenAI Academy: Creating images https://academy.openai.com/public/clubs/work-users-ynjqu/resources/creating-images
- OpenAI Academy: Canvas https://academy.openai.com/public/clubs/work-users-ynjqu/resources/canvas
- Sam Altman (@sama) on X: memory references all past conversations https://x.com/sama/status/1910380643772665873
- OpenAI Forum: Sam Altman on building the future of AI https://forum.openai.com/public/videos/event-replay-sam-altman-on-building-the-future-of-ai-2026-04-06
- OpenAI (@OpenAI) on X: a more capable memory system https://x.com/OpenAI/status/2062567556524003631
- OpenAI (@OpenAI) on X: memory references past chats https://x.com/OpenAI/status/1910378768172212636
- OpenAI on YouTube: Projects, 12 Days of OpenAI, Day 7 https://www.youtube.com/watch?v=FcB97h3vrzk