Set Up Claude Properly

Everything you need

• Claude memory and chat search Anthropic's current guide to search, memory topics, account availability, incognito chats, and privacy controls.Open
• Import and export Claude memory The official import flow for the new and legacy memory interfaces, including current availability and limitations.Open
• Humanizer on GitHub The third-party blader/humanizer repository referenced in this guide. Review it before downloading the ZIP.Open
• Use skills in Claude Anthropic's official instructions for enabling code execution and uploading a custom skill ZIP.Open
• Claude Academy course library Free courses, tutorials, learning paths, and practical use cases from Anthropic's education team.Open

How to use it

1. 1

   Turn on chat search and memory

   Open Settings, then Memory. Turn on the options that let Claude search and reference past chats and generate memory from your chat history. If your account still uses the legacy interface, look under Settings, then Capabilities, then Memory. Free, Pro, and Max accounts normally have memory on by default; Team and Enterprise access can depend on an administrator.

   Show me what you currently remember about me. Separate saved memory from details you found by searching past chats. Flag anything outdated, uncertain, overly personal, or missing, then ask me what I want you to update or forget.

   ClaudeCopy
2. 2

   Review what Claude remembers

   Memory is useful only when it is accurate. Open the Topics shown in Settings, then Memory and edit or delete anything that is wrong. Use an incognito chat for conversations you do not want added to memory or chat history.

3. 3

   Download the Humanizer skill

   Open the blader/humanizer repository on GitHub, select Code, then Download ZIP. This is a third-party skill, not an Anthropic product. Review the repository, author, files, permissions, and recent changes before installing it. A writing skill can reduce common AI-style patterns, but it cannot guarantee that text will pass an AI detector or sound exactly like you.

4. 4

   Upload and enable the skill

   In Claude, open Customize, then Skills. Select the plus button, choose Create skill, select Upload a skill, and upload the ZIP. Skills require Code execution and file creation to be enabled under Settings, then Capabilities. Toggle Humanizer on after it appears in your skills list.

   Use the Humanizer skill to revise the draft below. Keep every supported fact, remove generic AI-style wording, and match the rhythm and vocabulary in my writing sample. Do not invent details or optimise for an AI-detector score. First explain the three biggest changes you made, then give me the clean final draft. MY WRITING SAMPLE: [Paste two or three paragraphs you wrote yourself] DRAFT TO REVISE: [Paste the draft]

   ClaudeCopy
5. 5

   Open the free Claude Academy

   Claude Academy is Anthropic's official free learning library. It now includes more than the original 13-course collection, with material on Claude, Cowork, Claude Code, AI fluency, MCP, agent skills, and the API. Start with Claude 101 if you are new, then choose only the courses that match what you actually want to do.

6. 6

   Build a personalised course plan

   Copy the Claude Academy course-library URL into Claude and use the prompt below. Claude can tailor the order using the context it knows about you, but it should not pretend to have read lessons it cannot access. Give it your goal and weekly study time for a better plan.

   Review the current course catalogue at https://academy.claude.com/courses and use the context you know about my work, skill level, goals, and preferred learning style to build a personalised Claude learning plan. First list the course pages you were able to access and say clearly if any course content was unavailable. Then recommend only the courses relevant to me, in priority order. For each one, explain why it matters, the outcome I should aim for, an estimated time commitment based on the catalogue, and one practical project that will prove I learned it. Finish with a four-week schedule that fits [NUMBER] hours per week. Do not claim to summarise a lesson you could not access.

   ClaudeCopy
7. 7

   Bonus: export useful context from ChatGPT

   Open Settings, then Memory in Claude and select Start import. Accounts still on the legacy interface may find it under Settings, Capabilities, Memory, then Start import. Copy Claude's suggested export prompt into ChatGPT, review the output carefully, remove anything sensitive or irrelevant, and paste only the context you want Claude to retain.

   I am moving my saved context to another AI service. Create one complete, structured export of the information you are allowed to disclose from your saved memory and relevant past-chat context. Include my response preferences, writing style, recurring instructions, work and projects, goals, tools, interests, corrections I have made, and important decisions. Distinguish saved memories from inferred or conversation-based context, include dates when available, and mark uncertain details. Do not invent anything. Put the export in one plain-text code block. Exclude passwords, authentication codes, payment details, government identifiers, private addresses, medical details, and any other sensitive information unless I explicitly request a specific item. After the code block, tell me whether anything else remains that you are able to export.

   ClaudeCopy
8. 8

   Review the imported topics

   After importing, inspect the individual memory topics Claude creates. Delete anything unnecessary, correct errors, and ask Claude what it learned. Anthropic labels memory imports experimental, so do not assume every detail was transferred correctly.

   I updated my memory. Tell me what you learned about my work, goals, preferences, and communication style. Flag anything uncertain or contradictory and ask me to confirm it before relying on it.

   ClaudeCopy

Tips

• Memory is not retroactive: Claude does not automatically remember everything from before the feature was enabled.
• Review memory topics regularly. Delete outdated details rather than letting them shape future answers.
• Do not import your entire ChatGPT history blindly. Remove sensitive, irrelevant, incorrect, and duplicated information first.
• Treat every downloaded skill as third-party software. Read its SKILL.md and inspect any scripts before enabling it.
• Give a humanizer two or three samples you genuinely wrote. Your own voice is more useful than generic rules about sounding human.
• Humanizer output still needs fact-checking and editing. No rewriting tool can reliably guarantee an AI-detector result.