Move your context safely

01Move the useful layer, not the archive

The valuable part of your ChatGPT history is usually compact:

• how you like answers structured
• the language and tone that work for you
• the questions the tool should ask before starting
• the checks that catch common mistakes
• a few safe examples of good work

That can fit on one page. A chat archive can be thousands of pages and still fail to explain what matters.

02Make a five-part inventory

Open a blank document outside both tools. Use these headings and write short notes under each.

1. My work. The kinds of tasks you regularly ask AI to help with.
2. My readers. The people who receive the work and what they care about.
3. My preferences. Structure, spelling, tone, length and formatting.
4. My checks. The facts, policy boundaries and quality tests that matter.
5. My recurring jobs. The two or three workflows worth rebuilding first.

Leave out secrets, personal data, customer material and anything your employer hasn't approved for Copilot. Useful context can still be safe context.

03Choose the cleanest route you have

If you still have ChatGPT access

Ask ChatGPT to help surface working preferences, then review every line yourself. Don't ask it to summarise confidential content. Ask for patterns in how you requested and corrected answers.

Help me make a portable summary of how I prefer to work with an AI assistant.

Focus on:
- the tasks I commonly ask for
- how I like answers structured and written
- the clarifying questions that help
- corrections I've repeated
- checks I usually ask for

Don't include names, company information, private facts, conversation summaries or confidential source material. Treat your answer as a draft for me to review, not a complete record.

You can also inspect ChatGPT's Memory and Custom Instructions under Settings → Personalization. OpenAI says the memory view is a summary and may not show every detail. Read it as a clue, then rewrite only what belongs in your work profile.

If you want your ChatGPT data export

OpenAI currently offers export through Settings → Data Controls → Export for Free, Plus and Pro accounts, with a Privacy Portal route if you can't sign in. OpenAI says the export can take up to seven days and the email link expires after 24 hours. Business and Enterprise workspaces don't offer export through those workspace settings.

Use an export only to find your own working patterns. Keep it in your personal storage, review it locally and move only a small approved summary into Copilot.

If ChatGPT is already blocked

Rebuild from memory. Pick three recent jobs where the tool was useful and answer:

• What did I give it?
• What did I ask it to produce?
• What did I correct?
• What made the final answer usable?

Your own edits are often a better record of your preferences than the original prompt.

04Interview yourself when the history is gone

You can use Copilot to run the interview without giving it private history.

Interview me to build a one-page working profile for AI-assisted work.

Ask one question at a time. Cover:
1. my role and recurring tasks
2. who reads my work
3. how I like information structured
4. tone, spelling and formatting preferences
5. what the assistant should never assume
6. checks required before work is ready
7. three recurring jobs I want to improve

After the interview, draft the profile. Mark any answer that may contain confidential or personal information so I can remove it before saving.

05Finish with a context salvage sheet

Your sheet isn't the final profile. It's the reviewed source you'll use to make one in the next chapter.

MY WORK
Role in plain language:
Regular tasks:
People I write for:

HOW I LIKE TO WORK
Useful answer structure:
Tone and language:
Length and formatting:
Questions to ask me first:

QUALITY BOUNDARIES
Facts to verify:
Things never to invent:
Policy or approval boundaries:

FIRST THREE WORKFLOWS
1.
2.
3.

SAFE EXAMPLES I CAN USE
1.
2.