Make Claude remember you, so you stop explaining yourself in every chat
Claude starts most chats knowing nothing about you. Three layers fix that: a short "about me" note, a project for each ongoing job, and Claude's own memory, plus a 5-minute weekly tidy so it doesn't remember the wrong things. Gathered from Claude's help center and Anthropic's research on agent memory, rewritten in plain words.
Start here
How to use this canvas
- Top to bottom: why Claude forgets, then three layers, then how to keep them right.
- Do layer 1 today. It takes about 15 minutes and works on any plan.
- The same idea works in ChatGPT; see "Same idea in ChatGPT".
Why it forgets you
- Each new chat starts mostly fresh. Anything you said last week isn't there unless something carries it over.
- Three things can carry it over: a note you paste or attach, a project that gives every chat the same instructions and files, and memory that Claude writes as you talk.
- Anthropic's research team describes memory the same way: notes kept between sessions that the model reads later. See Sources.
What you'll end with
- A one-page "about me" note you can reuse anywhere.
- One project per ongoing job, each with its own instructions and files.
- Memory turned on, checked, and tidied once a week.
Layer 1: a short "about me" note
Let Claude interview you
Interview me so you can write a one-page "About me" note for future chats. Ask one question at a time, 8 questions at most: what I do, who my customers are, what I'm working on this month, how I like answers written, words I use for my products, and what I never want you to do. Then write the note in under 250 words, as short lines, no fluff.
Save the result somewhere you'll find it. You'll paste it into a project in layer 2.
Keep it short on purpose
- Anthropic's Claude Code docs warn that the longer an instructions file is, the less reliably the model follows it. So the layer that's read every time should stay small: Claude Code reads only the first 200 lines of its memory file and uses it as an index.
- Same for you: under 250 words. Put the long stuff (price lists, past work) in project files instead.
- Rewrite it when something big changes, not every week.
Layer 2: one project per ongoing job
Make a project
1. In Claude, click Projects in the left sidebar, then New Project.
2. Name it after the job, such as "Client: Northwind" or "Newsletter".
3. Click "Set project instructions" and paste your "about me" note plus 3 to 5 lines about this job.
4. Add files with the + button on the project's right side: a past example, a style guide, a price list.
5. Start every chat about this job inside the project.
Free accounts can make up to 5 projects (Claude help center, 2026-11).
Instructions or files: what goes where
| Put it in instructions | Put it in project files |
|---|---|
| Who you are, in a few lines | Long documents |
| How answers should look | Past examples to copy |
| Rules that always apply | Facts Claude should look up |
Instructions apply to every chat in that project only. Files are there to read when needed.
Layer 3: Claude's memory
Turn memory on
1. Open Settings, then Your Claude.
2. Turn on "Generate memory from chats". It's on by default for Free, Pro, Max and Ultra; on Team and Enterprise the owner decides.
3. On paid plans, also turn on "Search and reference chats" if you want Claude to look things up from old conversations when you ask.
4. Want Claude to keep something? Say "remember that I bill monthly, not hourly".
Each project keeps its own memory, separate from your general one (Claude help center, 2026-11).
Check what it remembers
- Settings, Your Claude, then Manage next to "Your memory". You can read, edit or delete each item.
- Or just say "forget that I work with Northwind".
- Pause memory keeps what's there but stops new saves.
- Reset memory deletes everything, including project memory, and can't be undone.
When you don't want it remembered
- Incognito chat: the ghost icon at the top right of a new chat outside a project. Not saved to memory or history.
- One normal chat without memory: switch Memory off in the + menu before your first message.
- Health, politics, religion and similar topics aren't saved unless you turn on "Include sensitive topics in memory".
- Never put passwords or card numbers in a note, project or chat.
Keep it right
A 5-minute weekly tidy
A wrong note gets trusted by later chats. Anthropic's fix is to have Claude check its notes against the original chat transcripts. You can do a manual, small version:
Look at what you remember about me and my work. List anything that looks out of date, duplicated or wrong, one per line, with why. Don't change anything yet.
Then delete or fix those items in Manage.
Test it with 3 questions
Open a new chat in the project and ask:
1. "What do I do, in one sentence?"
2. "How should your answers look for me?"
3. "What am I working on this month?"
If one answer is wrong, fix the note or the memory item it came from, not the chat.
Same idea in ChatGPT / Honest bit
The three layers in ChatGPT
- Layer 1: custom instructions, in Settings under Personalization.
- Layer 2: Projects, with their own instructions and files.
- Layer 3: Memory, with saved memories and reference to past chats; you can view and delete them in Settings.
Menu names change often. Check OpenAI's help center for the current steps before you follow them.
Limits worth knowing
- Free plans: up to 5 Claude projects, and searching past chats is for paid plans only (Claude help center, 2026-11).
- Memory can save something wrong. The weekly tidy is how you catch it.
- Reset memory can't be undone. Copy anything you want to keep first.
- Settings move between app versions. If a menu name doesn't match, search the help center for "memory".
Try it this week
Do layer 1 today
Run the interview prompt in "Let Claude interview you", save the note, and paste it into one project for your most frequent job. Ask the 3 test questions tomorrow.
Keep the note where you'll see it
Put your "about me" note, each project's instructions and your weekly tidy list as cards on one put·here canvas. When something changes, you change one card and paste it again.
Going further / Sources
Remember only what it can't read elsewhere
- Good memory: your preferences, decisions and why, words you use, people and projects. Things that live in your head.
- Bad memory: anything already in a file, a doc or a website. Point to it instead; copies go stale.
- Anthropic's research suggests telling the AI what memory is for, not how to use it. When they added a fixed rule ("record only failures"), their models lost much of the gain (2026-11).
Shape it: a short index plus a few topics
If you use Claude Code
- Claude Code reads a file called CLAUDE.md: one in your home folder (~/.claude/CLAUDE.md) for all projects, and one in each project folder for that project.
- Type /memory inside Claude Code to open and edit these files.
- It also keeps its own notes as it works; only the first 200 lines of its memory index load each session, so keep that index short (Claude Code docs, 2026-11).
Sources
- Claude help center, "What are projects?" and "How can I create and manage projects?" (official; checked 2026-11). https://support.claude.com/en/articles/9517075-what-are-projects
- Claude help center, "Use Claude's chat search and memory to build on previous context" (official; checked 2026-11). Memory settings, Manage, pause and reset, incognito, sensitive topics, project memory. https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context
- Anthropic, "The components of agent memory" (official research; 2026-11). Checking notes against chat transcripts to fix wrong ones; "tell Claude what memory is for, but not how to use it". https://www.anthropic.com/research/components-of-agent-memory
- Claude Code docs, memory (official; checked 2026-11). Long instruction files are followed less reliably; only the first 200 lines of the memory file are read each session. https://code.claude.com/docs/en/memory
- OpenAI help center, Memory FAQ (official; page didn't load for us in 2026-11, so the ChatGPT card is a summary to check against it). https://help.openai.com/en/articles/8590148-memory-faq
