Set Up Your Persistent Workspace

After this, you'll have a one-time setup in your AI tool of choice that tells it who you are and how you want responses, so you never have to repeat yourself again.

Before you start

Complete Find Your Daily Use Tipping Point first; this lesson builds on the recurring tasks you identified there so your workspace is configured around real, not hypothetical, use.

The idea

Most people retype the same intro at the start of every chat: 'I am a [job title], I work on [thing], please keep responses short.' That is wasted effort every single time. Each major AI tool has a one-time setup that fixes this permanently.

Preference tokens repeat at the start of multiple blank conversation paths. — The starting state for Set Up Your Persistent Workspace.

Set Up Your Persistent Workspace pathUse this model to move from the starting mistake to the lesson check.

ChatGPT calls it Custom Instructions. Claude calls it Projects. Gemini calls it Gems. They all work the same way: you describe your role and preferences once, and the AI reads it at the start of every conversation from that point forward. You stop re-explaining yourself. The AI stops defaulting to generic.

Here is the before and after: Before the setup: every Monday the same user types, 'I am a freelance copywriter specializing in B2B SaaS, please keep your writing direct and avoid filler phrases.' After the setup: Claude already knows that before the first message. A 45-second task that was happening dozens of times per month is now zero. The persistent workspace is the highest-leverage five minutes you will spend on AI this year.

What belongs in the setup: stable facts about who you are and how you work. Your role, your main focus right now, how long you want responses, whether you prefer bullet points or prose, any terms or approaches you want Claude to avoid. What does not belong: specific project details that change weekly, client names, or any sensitive information. The setup is not a vault. Keep it to stable preferences.

Review it every month or two. Your role, focus, and preferences evolve. The setup should too.

Try it (11 min)

Watch out for

Over-stuffing it with project-specific details that change week to week. The persistent workspace is for stable facts about you, not your current to-do list.
Setting it once and forgetting it. Review it every month or two. Your role, focus, and preferences evolve, and the setup should too.
Pasting in passwords, client names, or sensitive data. This is a saved setting, not a vault. Keep it to general preferences and role context only.
Making it so long that the important preferences get buried. Six sentences maximum. If you need more, you are probably mixing setup with task instructions.

Paste this into Claude

I want to set up a persistent workspace (Custom Instructions in ChatGPT, a Project in Claude, or a Gem in Gemini) so I stop re-explaining myself every session. Here is a draft of what I want it to know about me: I am [your job title or role, e.g. 'a freelance copywriter' or 'a small business owner in retail']. My current main focus is [one or two sentences about what you are working on right now]. When you respond to me, please [describe your preferences, e.g. 'keep answers under 200 words unless I ask for more, use plain language, no bullet points unless I ask, skip the preamble and get to the point']. Based on this, help me write a clean, concise setup I can paste directly into my tool. Tell me what I should cut, what I should add, and what to leave out entirely.

What a good response looks like

Here is a clean persistent workspace setup you can paste directly into Claude Projects or ChatGPT Custom Instructions:

'I am a freelance copywriter specializing in B2B SaaS companies. My current focus is writing email sequences and landing page copy for early-stage startups. Keep responses direct and under 200 words unless I ask for more. Use plain language, no jargon. When I ask for copy, default to a clear, conversational tone, not corporate or salesy. Skip preambles and get straight to the deliverable.'

That is six sentences. It covers your role, your current focus, and four specific preferences about how you want responses written. To save it: in Claude, click your initials in the bottom left, choose 'Profile,' and paste it in. In ChatGPT, click your name, choose 'Customize ChatGPT,' and paste into the 'How would you like ChatGPT to respond' field. Once saved, open a new conversation and test it with a real task. You should not need to explain who you are or what tone you want.

What good looks like

You wrote four to six sentences covering your role, your current focus, and how you want responses formatted
You opened the relevant settings (ChatGPT Custom Instructions, Claude Projects, or Gemini Gems) and saved your setup
Your next three chats started without you having to re-explain who you are
Your setup is under six sentences and covers role, focus, and at least two response preferences
You tested it in a fresh conversation and confirmed Claude responded differently than without a setup

Go deeper (15 min)

Paste this into Claude

I have my persistent workspace set up. Now I want to make sure it is actually working. Start three new conversations with my setup active, one for each of these tasks: (1) [a writing task you do regularly], (2) [a research or analysis task], (3) [a planning or decision task]. In each conversation, at the end, tell me: what did my setup tell you about me that shaped your response? What would you have done differently without it? Then suggest one thing I should add to or change in my setup based on what was missing.

What a good response looks like

Here is what I noticed across your three test conversations:

Writing task (product update email): Your setup told me you prefer under 200 words and direct tone. I skipped the standard 'I hope this finds you well' opener and went straight to the update. Without your setup, I would have written a standard formal email with a longer intro.

Research task (competitor pricing analysis): Your setup told me plain language, no jargon. I avoided terms like 'TAM' and 'pricing velocity' and described everything in plain sentences. Without your setup, I would have defaulted to business-report language.

Planning task (Q3 content calendar): Your setup did not mention anything about how you like planning documents formatted. I defaulted to a table, which may or may not be what you want. Suggested addition: 'For planning and calendar tasks, give me a simple bullet list by week, not a table.'

Add that one sentence to your setup and your planning outputs will match the format you actually use. Everything else looked well-calibrated.

What good looks like

You ran three different task types with your setup active
In each conversation, Claude named at least one thing from your setup that shaped its answer
You found at least one gap in your current setup from the three tests
You updated the setup with at least one improvement before ending the session

When this breaks

Breaks when you treat the setup as a project log instead of a stable profile, because constantly changing project details bury the durable preferences and the model loses signal about how you actually want to work.
Breaks when you set it once and never revisit it, because your role, focus, and tooling evolve while the setup keeps steering responses toward who you used to be.

AI can help with this

Open Claude and say: 'I want to set up a Project for [your role or focus]. What should I include and what should I leave out? Help me draft it in under six sentences.' Let Claude write the setup for you. You answer its questions, it writes the text.

The tokens settle into one persistent workspace anchor that feeds each path automatically.

You can now

Save a four-to-six-sentence setup in your AI tool that covers role, current focus, and response preferences, then verify in a fresh chat that responses arrive shaped by it.

Key takeaways

Configure once. Benefit every session. The persistent workspace is the highest-leverage five minutes you will spend on AI this year.

Configure your AI workspace once and stop retyping who you are at the start of every chat.
Keep the setup to stable facts: role, current focus, and a handful of response preferences. Six sentences max.
Leave out project-specific details, client names, and anything sensitive. The setup is not a vault.
Review it every month or two. Your role and preferences shift, and the setup needs to shift with them.

Go deeper

Take this with you

Was this helpful?

← Back to lessons

I want to set up a persistent workspace (Custom Instructions in ChatGPT, a Project in Claude, or a Gem in Gemini) so I stop re-explaining myself every session. Here is a draft of what I want it to know about me: I am [your job title or role, e.g. 'a freelance copywriter' or 'a small business owner in retail']. My current main focus is [one or two sentences about what you are working on right now]. When you respond to me, please [describe your preferences, e.g. 'keep answers under 200 words unless I ask for more, use plain language, no bullet points unless I ask, skip the preamble and get to the point']. Based on this, help me write a clean, concise setup I can paste directly into my tool. Tell me what I should cut, what I should add, and what to leave out entirely.

Here is a clean persistent workspace setup you can paste directly into Claude Projects or ChatGPT Custom Instructions: 'I am a freelance copywriter specializing in B2B SaaS companies. My current focus is writing email sequences and landing page copy for early-stage startups. Keep responses direct and under 200 words unless I ask for more. Use plain language, no jargon. When I ask for copy, default to a clear, conversational tone, not corporate or salesy. Skip preambles and get straight to the deliverable.' That is six sentences. It covers your role, your current focus, and four specific preferences about how you want responses written. To save it: in Claude, click your initials in the bottom left, choose 'Profile,' and paste it in. In ChatGPT, click your name, choose 'Customize ChatGPT,' and paste into the 'How would you like ChatGPT to respond' field. Once saved, open a new conversation and test it with a real task. You should not need to explain who you are or what tone you want.

I have my persistent workspace set up. Now I want to make sure it is actually working. Start three new conversations with my setup active, one for each of these tasks: (1) [a writing task you do regularly], (2) [a research or analysis task], (3) [a planning or decision task]. In each conversation, at the end, tell me: what did my setup tell you about me that shaped your response? What would you have done differently without it? Then suggest one thing I should add to or change in my setup based on what was missing.

Here is what I noticed across your three test conversations: Writing task (product update email): Your setup told me you prefer under 200 words and direct tone. I skipped the standard 'I hope this finds you well' opener and went straight to the update. Without your setup, I would have written a standard formal email with a longer intro. Research task (competitor pricing analysis): Your setup told me plain language, no jargon. I avoided terms like 'TAM' and 'pricing velocity' and described everything in plain sentences. Without your setup, I would have defaulted to business-report language. Planning task (Q3 content calendar): Your setup did not mention anything about how you like planning documents formatted. I defaulted to a table, which may or may not be what you want. Suggested addition: 'For planning and calendar tasks, give me a simple bullet list by week, not a table.' Add that one sentence to your setup and your planning outputs will match the format you actually use. Everything else looked well-calibrated.