The short answer
To make Anthropic's Claude remember your instructions across chats, add them in Settings, then Your Claude, then Instructions (Account, then Instructions for Claude in the older layout); Anthropic says these apply to all of your conversations. Use memory, on by default for Free, Pro, Max and Ultra, for facts about you, and edit its topics under Manage next to Your memory. Put rules for one job in a Project's instructions.
| Made by | Anthropic |
|---|---|
| Account-wide rules | Settings, Your Claude, Instructions (older layout: Account, Instructions for Claude) |
| Memory default | On for Free, Pro, Max and Ultra; Team and Enterprise members turn it on after an owner allows it |
| Edit memory | Settings, Your Claude, Manage next to Your memory, then Topics |
| Project rules | Set project instructions; apply only inside that project |
| Not remembered | Incognito chats; local Cowork sessions; sensitive topics unless switched on |
| Import memory | Settings, Your Claude, Start import (experimental) |
Why Claude seems to forget your instructions
You tell Claude to use British spelling, skip the bullet points and never open with "Great question". It does, for that chat. The next day, in a new chat, it is back to its old habits. That is not a bug. A chat on its own only knows what is in that conversation, so an instruction typed in one chat does not travel to the next unless you put it somewhere Claude reads every time.
Anthropic's Claude has three such places, and most people only know one of them. Anthropic's help centre describes them in its articles on personalisation, memory and Projects, all checked on 4 November 2026:
| Where | What it is for | Where it applies |
|---|---|---|
| Instructions (account-wide) | Rules you write once: tone, format, spelling, what to avoid | All of your conversations |
| Memory | Facts Claude saves about you and your work as topics | Chats outside projects; each project also has its own memory |
| Project instructions | Rules for one piece of work, such as a client or a report | Only chats inside that project |
The short version: rules go in Instructions, facts go in memory, and anything that only matters for one job goes in a project.
What you need first
- Any Claude plan. Anthropic says memory is on by default for Free, Pro, Max and Ultra plans on the web, Claude Desktop and Claude Mobile. Instructions and Projects are also available on the Free plan, which allows up to 5 projects.
- On Team or Enterprise, an owner who has allowed memory. Owners control whether memory is available in Organization settings, then Capabilities, and Anthropic says it is off for each member until they turn it on.
- The latest app. Anthropic's help pages describe two Settings layouts. In the new one the page is called Your Claude; in the older one the same options sit under Memory (or Capabilities) and Account. Use whichever you see.
- Five minutes to write your rules down. The quality of what Claude remembers depends on how clearly you state it.
Step 1: put standing rules in Instructions
This is the fix for "Claude keeps forgetting how I want things written". Anthropic's personalisation article says any instructions you add here "will be applied to all of your conversations".
- Click your initials in the lower left corner of Claude.
- Select Settings, then Your Claude (or Account in the older Settings layout).
- Find Instructions (called Instructions for Claude in the older layout).
- Write your rules as short, plain sentences, then save.
Keep this field for rules that are true in every chat. "Use British spelling" belongs here. "The client is called Northwind and hates jargon" does not, because it will leak into chats about other work. If you used Global instructions in Claude Cowork, Anthropic says they become part of your account instructions (the article calls them Instructions for Claude) once you have the new Claude experience, so check the field for old rules before you add new ones.
Step 2: let memory hold the facts about you
Memory is for context rather than rules: your role, your team, the projects you are working on, the tools you use. Anthropic's memory article says Claude saves memory as a set of individual topics as you chat, rather than summarising conversations after they end.
- Open Settings, then Your Claude (or Settings, then Memory in the older layout).
- Turn on Generate memory from chats if it is off.
- To see what Claude has saved, select Manage next to Your memory. Everything is listed under Topics; open one to read it, use the edit icon to change it, or delete it.
- To add something on purpose, tell Claude in a chat: "Remember that I lead a team of six in finance." You can also ask it to change or forget a memory the same way.
On paid plans there is a second switch on the same page, Search and reference chats, which lets Claude look up past conversations when you refer to them. Anthropic says chat search is available on Pro, Max, Ultra, Team and Enterprise.
Some things are kept out by design. By default Claude does not store sensitive topics such as your health, religious beliefs or politics; you can turn on Include sensitive topics in memory if you want them saved. Anthropic says government ID numbers, criminal history, financial account numbers and immigration status are never saved, even if you ask.
Step 3: use a project for instructions that belong to one job
When rules only make sense for one client, report or course, put them in a Claude Project. Anthropic's Projects article says project instructions are used for all the chats within the project, and its personalisation article says they only apply there.
- Create a project, then select Set project instructions.
- Write how Claude should behave for this work and select Save instructions.
- Add the files Claude needs to the project knowledge. Anthropic notes that context is not shared across chats in a project unless it is in the knowledge base, so a decision made in one chat should be added there if later chats need it.
Projects also change how memory works. Anthropic says each project has its own memory; inside a project Claude uses that project's memory plus your account memory from chats outside projects, but not memory from other projects. To keep a project completely separate, turn off Use account memory under Memory on the project's page. Our guide to setting up Claude Projects covers files, instructions and sharing in more detail.
A worked example: a marketer who is tired of repeating herself
This is an illustration, not a test we ran. Imagine Sam, a hypothetical marketing manager who uses Claude every day for emails, briefs and social posts, and who keeps typing the same three lines at the start of every chat.
- Instructions: Sam writes "Use British spelling. Write in plain, warm sentences. No bullet points unless I ask. Never open with praise for the question. If you are unsure of a fact, say so."
- Memory: in a normal chat Sam says "Remember that I am the marketing manager at a 40-person software company, my audience is finance directors and I report to the CEO." She then opens Manage next to Your memory and checks the topic says exactly that.
- Project: for a product launch she creates a project with the instructions "Every draft must mention the launch date and link to the pricing page. Keep posts under 120 words", and uploads the messaging document.
- A private chat: when she wants to ask about something personal without it shaping future answers, she starts an incognito chat with the ghost icon in the upper right corner of the screen, which Anthropic says is not saved to memory or chat history.
The next morning she opens a new chat and types "Draft a reply to this email". Her Instructions now apply to that chat and her memory tells Claude who her readers are, so she no longer has to repeat the three lines.
Common mistakes and how to fix them
- Putting rules in memory and facts in Instructions. Memory topics are written by Claude, not in your own words. Rules you need followed every time belong in Instructions, in your own words.
- Expecting a project to know your other projects. It does not use memory from other projects. Add the shared context to each project's knowledge or to your account-wide Instructions.
- Memory seems to do nothing on a work account. On Team and Enterprise, memory is off for each member until they turn it on, and an owner must allow it first.
- Memory switched off for one chat. You can turn memory off for a single chat from the + menu before the first message, and Anthropic says it cannot be changed after that. A crossed-out memory icon next to the title shows it is off.
- Cowork tasks ignore your memory. Anthropic says memory is shared with Claude Cowork when Cowork runs in the cloud, but Cowork sessions that run locally on your computer do not use memory. Put essential rules in Instructions instead.
- Deleting a chat to remove a memory. Memories created from a deleted chat are not removed. Delete the topic itself under Manage.
Moving your memory from ChatGPT, Gemini or another assistant
If you are switching to Claude, you do not have to start from nothing. Anthropic's import article covers moving memory between Claude and other AI providers, and says memory imports are available on Free, Pro, Max, Ultra, Team and Enterprise plans on the web and Claude Desktop: in Settings, then Your Claude, select Start import, paste what you exported from the other assistant and select Add to memory. Anthropic labels the feature experimental and says Claude may not keep imported personal details unrelated to work. For the reverse move, you can ask Claude to write out its memories of you verbatim. Our Claude alternatives page covers the other direction.
Which option should you use?
Use Instructions for anything you would otherwise type at the top of every chat; it is the most reliable of the three because it is your exact wording and it applies everywhere. Use memory for who you are and what you are working on, and check the topics once a month, because a stale memory is worse than none. Use projects when a rule or a file only matters for one job. If you mainly need Claude to stop forgetting how to write, Instructions alone solves it. To practise writing rules that Claude follows, try the Claude course in Learn, and see our Claude guide for plans and privacy settings.
Prompts to copy
Draft your Instructions
Interview me with up to 8 short questions about how I want you to write and work: tone, spelling, format, length, things to avoid and how to handle uncertainty. Then turn my answers into 5 to 8 plain sentences I can paste into Settings, then Your Claude, then Instructions.
You get: A ready-to-paste set of standing rules in your own preferences
Save the facts that matter
Remember the following about me for future chats: I am a [role] at [type of organisation], I work mostly on [main tasks], my readers are [audience] and I use [tools]. Then repeat back exactly what you saved.
You get: Clean, checkable memory topics instead of guesses from old chats
Audit your memory
List everything you currently remember about me, grouped by topic. Mark anything that looks out of date or that you are unsure about, so I can fix or delete it.
You get: A quick review that catches stale or wrong memories
Turn a chat into project instructions
From this conversation, write project instructions for [project name]: the goal, the audience, the rules every draft must follow and the files to rely on. Keep it under 150 words and leave out anything that applies to my other work.
You get: Project instructions that stay scoped to one job
Check before an important draft
Before you write, tell me which of my standing instructions and memories you will apply to this task and whether any of them conflict. Then write the draft.
You get: A visible check that Claude is using the rules you set