The Practice Room / The Desktop Apps / Episode 27
Set up Claude so it knows who you are
Every conversation you have with an AI starts from nothing. It does not know your name, it does not know what you do, and it does not know whether you want the short answer or the whole explanation. So it gives you the answer it would give anybody.
There is a panel that fixes that, and it is in the same place in both of the big apps: the bottom left corner, behind your own name.
Where it is in Claude
Open Claude in your browser or the desktop app. Look at the very bottom of the left panel, under your list of chats. Your initials and your plan are sitting there. Click that.
A small menu opens. Choose Settings. In the dialog that opens there is a list down the left: Account, Billing, Usage, then Appearance, Your Claude, Capabilities, Notifications, Privacy.
Two of those three steps are the part people miss. The door is your own name, and the room is called Your Claude, which does not sound like settings at all.
Account first, because it takes ten seconds
Before the instructions, open Account in that same list. Two fields there do real work and most accounts leave them empty.
What should Claude call you? Put the name you actually go by, not your legal name.
What best describes your work? This is a dropdown and it sits on Other until you change it. Changing it tells Claude what world your questions come from, which is most of what it needs to stop explaining the basics of your own field back to you.
The Instructions box, and what to put in it
Now open Your Claude. The first thing on the page is a large empty box called Instructions, with a note saying Claude will keep these in mind across your chats.
The mistake is writing a paragraph. Write it in sections instead, because you are going to come back and edit it.
Put this in
- About me. What you do, who you are writing or building for.
- My priorities, in order. Numbered. The order is the useful part.
- How to talk to me. How much detail, what to lead with, what to skip.
Leave this out
- Anything for one project only. That belongs in a project.
- Anything you would be unhappy to see in an answer.
- A wall of text. You will not maintain it.
A line worth stealing for the third section: explain it like you would to a fifth grader, and define any term you use before you use it.
Do it yourself
- Open Claude and click your initials at the bottom left.
- Choose Settings, then Account. Fill in what Claude should call you, and change the work dropdown off Other.
- Click Your Claude in the left list.
- In the Instructions box, write three headings: About me, My priorities in order, How to talk to me. Two or three lines under each.
- Close the dialog. There is no save button, it keeps what you typed.
While you are in there, look at Memory
Underneath Instructions is a section called Memory, and on a personal account two of its switches are already on: Claude can search your past chats, and it writes down what it learns about you as it goes.
Anthropic's own support page says memory is on by default on the Free, Pro, Max and Ultra plans, and that searching past chats is enabled by default once it reaches your account. So this is not something you turned on by accident.
Below those are the controls people do not know exist. Include sensitive topics in memory is off unless you turn it on. Your memory lets you read what it has written down, edit it, or delete any of it. There is an import if you are moving from another AI, and an export if you want your own copy.
Check yourself
Ask Claude something you have asked before, in a brand new chat. Then compare it with an answer from last week. The difference shows up fastest in what it leads with: a set-up account starts with the thing to do, rather than defining the term you just used.
When it goes wrong
Nothing seems to have changed. Instructions apply to new messages, not to the conversation you already had open. Start a fresh chat.
It is still too formal, or still too casual. That is a sign your third section is describing a personality rather than a behaviour. "Be friendly" does very little. "Lead with the step I should take first" changes every answer.
You cannot find Your Claude. You are probably in the Account tab. The list on the left of the Settings dialog is what you are looking for, and Your Claude is the fifth item down, under Appearance.
What she says in the film
You have been talking to a stranger every single time.
Claude. ChatGPT. Any of them. It does not know your name, your work, or how you want to be spoken to.
There is a panel that fixes that. I will show you where it is in Claude.
Bottom of the left panel, under your chats. Your name. Then Settings.
Account first. What it should call you, and what your work actually is. Most people leave that one sitting on Other.
Then Your Claude, and this is the part worth your time. Instructions.
Write it in sections. About me. Then your priorities, numbered, in order.
Mine says explain it like you would to a fifth grader, because the people I teach are not developers.
It keeps that in mind in every chat, not just the one you are in.
Underneath is Memory, and on a personal account the first two are already on.
It can search your old chats, and it writes down what it learns. Sensitive topics are off unless you say otherwise.
Your memory lets you read it, edit it, delete it. You can bring one in from another AI, or take yours out.
Ask it something before. Ask it the same thing after. The difference is not subtle.
Subscribe.
No permission required.
Book a session and we do it together on a call.
Book an hourAsk what did not make sense
Comments go through our own form and are read before they appear. Links are not allowed; describe it in words.
More from The Desktop Apps.

Claude Desktop: Chat vs Code, what actually changes
This is the Claude desktop app.

Where the settings are in Claude Desktop
You opened Claude, and you cannot find the settings.

Does Claude run on your computer?
Give the chat a job, and it does not run on your computer at all. It runs on a computer somewhere else, one th…
Get the next walkthrough.
One email when something goes up. The argument, the film and the term, nothing else. One click to leave.
Have a question now? Ask Mstimaj, she knows the archive.