Setup Guide

3 Claude settings
most users skip

By Dan Donovan, @askdandonovan

Turn these on before your next session. Ten minutes of setup and Claude starts feeling like it actually knows you, because it will.

3 Steps
~10 min Setup time
Free Cost
Any plan Works on

Claude's memory lets it remember details about you between conversations: your name, role, how you work, what you're building. Without it, you re-explain yourself every single session. With it, you don't.

What you'll have after this
Persistent context
Claude remembers your role, goals, and preferences across sessions
ChatGPT history carried over
Everything ChatGPT knows about you, imported into Claude's memory
Personalised from day one
Responses calibrated to your context, not generic defaults
A tailored learning path
Claude picks the right course for you from Anthropic's free library
Before you start: You need a Claude account at claude.ai. Memory is included on all plans: free, Pro, and Max. Paid plans give you significantly more daily usage and also unlock Search and reference chats, which lets Claude pull from your past conversations mid-chat.
1

Turn on Claude's memory

Memory is on by default for new accounts, but most people never check what's actually in there, or set it up properly. Here's where it lives:

→
Open claude.ai and go to Settings
→
Look for Memory in the sidebar. If it's not there, click Capabilities and find the Memory section (older accounts)
→
Make sure Generate memory from chats is on
→
Click through the memory entries to see what's stored
Why two paths: Anthropic is mid-rollout on a new memory experience. If you see Settings → Memory, you're on the new one. Your memories are listed there, and you can select any entry to read it, plus Pause and Reset controls. If Memory still sits under Capabilities, you're on the older version and there's a View and edit memory button instead. Same feature either way. On paid plans (Pro/Max), there's also a Search and reference chats toggle that lets Claude pull from past conversations mid-chat. Worth enabling too if you have it.

Now open a new chat and introduce yourself. Most people skip this step. Don't. This is what makes memory actually useful immediately instead of waiting weeks for Claude to piece you together.

Paste this prompt and fill in the brackets:

Paste into Claude 
Please save this to your memory about me:

My name is [your name]. I'm a [your role] at [company or context].

What I'm working on: [1–2 sentences about your current focus]

How I use Claude: [e.g. writing, research, coding, business decisions]

Preferences: [e.g. be direct, no fluff, explain things simply, use examples]

Goals: [what you're trying to accomplish over the next 3–6 months]

Claude will confirm what it saved. Go back to your memory settings any time to review it and remove anything that's wrong.

2

Import your ChatGPT data

If you've been using ChatGPT, it already knows a lot about you: your writing style, how you think, what you're working on. This step carries that across to Claude in about two minutes.

First, go to ChatGPT and paste this prompt:

Paste into ChatGPT 
Based on all our conversations, create a structured profile of everything you know about me. Include:

- My name and professional role
- What I'm currently working on or building
- My communication style and preferences (how I like responses structured)
- My goals, both short and long term
- Topics I frequently ask about
- Any recurring context I've given you (company size, tech stack, industry, etc.)
- How I think and make decisions

Format this as a clean, structured summary I can share with another AI assistant to get it up to speed on who I am.

Copy the output. Then go back to Claude and paste this:

Paste into Claude 
Please add this to your memory about me — this is a profile exported from my ChatGPT history:

[paste the ChatGPT output here]

Save whatever is useful. Ignore anything that doesn't apply.

Then verify it worked. Start a new chat and ask:

New Claude chat: verify 
What do you know about me? Give me a quick summary of what's in your memory.
If you've never used ChatGPT: Skip this step. Just use the prompt from Step 1 to introduce yourself directly. That's enough to get started.
If ChatGPT comes back thin: A vague, generic profile usually means ChatGPT's memory was switched off, or you've been working in temporary chats, so there was nothing much to summarise. Nothing's broken. Just use the Step 1 prompt and tell Claude about yourself directly.
3

Get your personalised learning path

Anthropic has a library of free courses, covering Claude itself, Claude Code, AI concepts, and applied use cases. Most people either don't know it exists, or they open the home page and don't know where to start.

Since Claude now knows your background and goals, let it pick the right starting point. Paste this into a new chat:

Paste into Claude 
Based on what you know about me, which of Anthropic's free Claude courses would be most useful for me to take first?

Give me:
1. Your top recommendation and why it fits my situation specifically
2. The 3 core concepts I'll learn from it, explained in plain language
3. One thing in my actual work I could apply it to in the first week

Give me the exact course name so I can find it. Don't guess at a URL. Keep it brief.

Because Claude now has your context from Steps 1 and 2, the recommendation will be specific to you, not a generic "start here" answer that ignores where you actually are.

What the courses cover: Claude Code, Claude Cowork (working with files on your computer), the Model Context Protocol, applied prompting, AI fundamentals, and building with the Claude API. Around 20 courses, all free, including the quizzes and the completion certificates. Self-paced: some are under an hour, others go several hours deep. Browse them at anthropic.com/learn and take them at anthropic.skilljar.com (email signup, no Claude account needed). One exception: Claude Code 101 needs a Pro/Max plan or an API key for the hands-on exercises.

What to try now that Claude knows you

Memory changes what's possible. These prompts work noticeably better now. Copy them directly.

High leverage
Find your highest-ROI automation
Claude knows your role and what you're working on. Ask it to identify the most time-consuming thing you do repeatedly, then help you automate it.
Prompt 
Based on my role and what you know about how I work, what's the single highest-leverage thing I could automate or systematise right now? Walk me through what it would look like in practice.
Blind spots
Ask what you're getting wrong
This one only works with memory on. It needs to compare what you say you want against what you actually spend time on. Ask a fresh Claude and you get nothing.
Prompt 
Based on everything you know about me, what am I getting wrong? Where does my stated goal not match what I actually spend my time on? Be blunt — I'd rather hear it than not.
Tool audit
Cut your tool stack
Most people are paying for tools they barely use. Claude can spot the overlap and tell you what to cut.
Prompt 
Here are the tools I'm currently paying for or using regularly:

[list your tools]

Given how I work and what I'm trying to accomplish, which of these would you cut? What overlap do you see? Are there any gaps where one tool could replace two?

Three things nobody tells you about memory

Worth two minutes now. They're the reason most people quietly give up on this in week three.

1. It remembers facts, not thinking. Memory stores who you are and how you like to work: your role, your preferences, your context. It does not store the strategy you landed on last Tuesday, the document you built together, or the reasoning behind a decision. Come back a week later expecting Claude to remember the conclusion and you'll be disappointed. If a specific piece of work matters, save it somewhere real.
2. Client work needs Projects, not memory. Memory is account-wide: one general picture of you, applied to every chat. That's exactly what you want for your own preferences, and exactly what you don't want when you're juggling three clients and don't want context bleeding between them. Projects hold their own separate memory, scoped to that project only. Use memory for you, Projects for each thing you work on. Full Projects guide here →
3. Memory goes stale, and stale memory is worse than none. This is the one that actually bites. A throwaway line from a single conversation (a tool you tried once, a client you no longer have, a goal you've moved on from) gets stored as a permanent fact about you, and then quietly skews every answer for months. You won't notice it happening. Audit it once a quarter:
Quarterly clean-up: paste into Claude 
Show me everything in your memory about me. Flag anything that looks like a one-off from a single conversation rather than a real, ongoing fact about how I work.

And you don't have to wait for the quarterly sweep. The moment you spot something wrong in any chat, just say it: "Update your memory: that's no longer true, I'm now [X]." It'll fix it on the spot.

The difference between a configured Claude and a fresh one isn't subtle. People who've done this report spending less time re-contextualising at the start of every chat. That friction compounds over weeks.

Business owners have used a properly-set-up Claude to run competitor analyses against their own positioning, rewrite entire pricing pages, build client onboarding workflows, and structure investor updates, without hiring anyone to do it. The tool is the same. The context is what changes what's possible.