If you paste your job and your constraints at the start of every conversation, ten minutes of setup is enough for Claude to start already briefed.
Who it's for
Anyone who uses Claude more than once a week. Explaining your industry and your style again in every exchange is wasted energy. Everything is set in the app, without a line of code.
Instructions
You write once who you are, how you want it to answer and what it should never do. Claude applies these permanent rules to all your conversations and all your tasks.
Where to put them:
- Open Settings > General.
- Find the "Instructions for Claude" field.
- Paste your block, save.
Since the September 16, 2026 merge, this field also replaces the global instructions of the old Cowork. One place, valid for quick questions as well as long tasks.
A block to copy
Example for a freelance marketer:
I'm a freelance marketer in Bordeaux, working with B2B SMB clients.
Answer in English, direct tone, no em dashes.
For each action you suggest, give me the short version first.
If you're not sure about something, say so.
Before setup, the question "I need ideas for my newsletter" gets you a generic list of 10 formats. After setup, Claude suggests 3 B2B angles that fit an SMB, in English, with no fluff.
Keep this block short. Four or five lines that change the answers beat a page of rules Claude only half applies.
Styles
A style sets the shape of the answers: shorter, more explanatory, or in a voice you define yourself. There are preset styles, and you can create your own.
The difference with instructions: instructions say who you are and what you want, the style says how it's written. You switch styles from one conversation to the next, the instructions stay.
Memory
What Claude remembers about you from one conversation to the next: your industry, your current projects, your format preferences. It files this into Topics, short notes per subject that you can read, fix or delete. You no longer start from scratch.
What changed in 2026:
- Since July 10, 2026, memory is stored as entries sorted by category.
- Since August 25, 2026, there's only one memory, shared between quick questions and long tasks (back then, between chat and Cowork). Claude updates it during the conversation.
- In Settings > Memory, you see the Topics, the subjects Claude has remembered. You can edit them.
- Sensitive topics are left out of memory by default.
You can also be explicit: "remember that I'm working on the launch of X until the end of November". To fix something, go through the Topics: a corrected note applies to every following conversation. In Settings > Memory, you can also pause memory (it's kept but no longer used) or reset it (everything is erased, with no way back).
| Plan | Memory by default |
|---|---|
| Free, Pro, Max | On |
| Team, Enterprise | Off until you turn it on |
Coming from another assistant, like ChatGPT? You can import the memory it built up about you, or export Claude's memory to back it up. Anthropic presents this feature as experimental.
Reflect and incognito
Reflect (beta since July 9, 2026) gives you a monthly review of your usage in Settings > Reflect, with a shortcut to set break reminders. Only on Free, Pro and Max, on the web or the desktop app, with memory turned on. Optional, worth trying if you spend your days in Claude.
Incognito mode is for conversations you don't want feeding your memory. Since the merge, it opens the old interface: Claude doesn't create files or run code there. If you just want a conversation without memory in the new interface, turn memory off from the "+" menu before you send your first message.
Projects
For a client or a topic that keeps coming back, you create a project: a workspace with its own files and its own instructions. Every question you ask inside it starts with all that context already loaded.
Example: a "Client Acme" project where you drop the master contract, the brand guidelines and the last three meeting notes. In the project instructions: "Acme is an industrial SMB, contact is Julie, formal tone, always mention the approved budget". Every new brief starts with the right constraints.
In the unified interface, a project works for every conversation, quick questions and long tasks alike. Each project also has its own memory, kept apart from the rest: what Claude learns in "Client Acme" doesn't spill into your other conversations. The Free plan is limited to 5 projects.
When to make a project: as soon as you find yourself pasting the same context at the start of every conversation. A recurring client, a book in progress, a file that's been dragging on for six months.
Connectors
A link between Claude and one of your tools (Google Drive, Notion, Slack...). Claude reads the source directly, so you no longer upload files by hand. Connectors run on MCP, an open standard that lets an AI talk to a piece of software.
What moved:
- Since April 23, 2026, Claude suggests connectors during the conversation when they can help, with no paid placement. The same wave added everyday connectors: AllTrails, Audible, Booking.com, Spotify, Uber...
- On September 15, 2026, Salesforce arrived in Claude (beta).
- On September 23, 2026, Anthropic opened the Claude Marketplace, a catalog of connectors and plugins.
- Google Workspace, Office and Slack have their own integrations: see chapter 8.
To add one, accept Claude's suggestion when it shows up, or go through the settings. Each connector gives Claude access to your data in that tool: only connect what you use.
How much it costs
Instructions, styles, memory and connectors are included in every plan, Free included. Projects too, up to 5 on Free.
When to use it vs something else
| Level | Where | What for |
|---|---|---|
| Global | Settings > General, "Instructions for Claude" | Your profile, preferences that apply everywhere |
| Project | Project instructions and files | A client, a topic, a sensitive file |
| Code project | CLAUDE.md file in Claude Code | The rules of a repo, see the Claude Code guide |
Your instructions, the project's instructions and your connectors' tools are part of the context Claude rereads with every message. A 3-page wall of text or 15 connectors plugged in for nothing gets paid for out of your limits, message after message. Keep it short and unplug what you don't use (chapter 11 puts numbers on it).
Question
Your company just switched to Claude Team. A week later, Claude still remembers nothing from one conversation to the next. Why?
Pick an answer to see the explanation.
Key takeaways
- Start with 4 or 5 lines in Settings > General, "Instructions for Claude" field. Projects can come later, once you know what keeps repeating.
- One memory for everything since August 25, 2026. Reread the Topics now and then.
- Team and Enterprise: memory is off by default, turn it on yourself.
- One project per client or recurring topic.
- Connect only what you need, since each connector opens access to your data.