Table of Contents
Introduction
When researchers move between universities or research institutes, we normally think about transferring email, research data, Git repositories, computing environments, and other digital resources.
ChatGPT is easier to overlook.
This is particularly relevant for postdocs, fixed-term researchers, students, and other researchers whose institution provides access to ChatGPT Edu, Enterprise, or Business.
When the appointment ends, access to that institutional workspace may also end.
After long-term use, however, ChatGPT contains much more than a collection of chat transcripts.
It may contain context about your research, ongoing projects, writing preferences, terminology, LaTeX conventions, programming environments, previous approaches you have tried, Custom Instructions, and Memory.
Taken together, these can make ChatGPT function as a kind of
personal research assistance environment.
That environment also needs portability.
This page summarizes what researchers should preserve before leaving an institution, based on OpenAI's documented features as of August 20, 2026.
The problem: official documentation exists, but not as an offboarding guide
OpenAI provides Help articles covering topics such as:
exporting ChatGPT data,
using exported conversations in another account,
data export from ChatGPT Edu,
Business and Enterprise workspaces,
and ChatGPT Memory.
However, the relevant information is distributed across several articles.
There is not currently a single
“do this before you leave your research institution”
offboarding / portability guide.
This is especially important because the available export functionality differs depending on whether your institution uses Edu, Enterprise, or Business.
As of this writing, data export through ChatGPT settings is available for Free, Plus, Pro, and eligible ChatGPT Edu workspaces. It is not available through ChatGPT settings for ChatGPT Business or Enterprise workspaces.
For that reason, migration planning should begin before you lose access to the institutional workspace, not afterward.
Pre-departure checklist
1. Determine whether your institution uses Edu, Enterprise, or Business
First determine which type of workspace you are using.
The interfaces may look similar, but the rules governing portability are not identical.
For ChatGPT Edu, data export is not automatically available in every workspace. The workspace must have no data-residency configuration, and an administrator must allow the relevant members to export their own data.
ChatGPT Business and Enterprise currently do not provide workspace data export through ChatGPT settings.
Therefore, identify the plan and check your institution's local policy first.
2. Do not casually merge your personal workspace into an institutional workspace
In some cases, a personal workspace can be merged into an institutional workspace.
This should be treated as a consequential decision.
For ChatGPT Business, merging a personal workspace into the Business workspace deletes the personal workspace after migration, and the merge cannot be undone.
For Enterprise, some organizations may require a personal-workspace merge. After such a merge, personal chats, files, and custom GPTs move under the organization's control, and the merge is permanent.
Researchers often move between institutions every few years.
For material that you expect to remain part of your long-term personal research environment, it is therefore worth considering whether your personal and institutional workspaces should remain separate.
3. Before leaving, ask whether your conversations can be exported
Do not wait until after your account has been deprovisioned.
For eligible ChatGPT Edu workspaces, an administrator can enable member data export. The resulting export may include conversations.json, files and other assets used in conversations, and related metadata.
For personal Free, Plus, and Pro accounts, data export can be requested from Settings → Data controls → Export data. OpenAI notes that an export may take up to seven days and that the download link expires, so this should not be left until the final day of an appointment.
If you use a Business or Enterprise workspace for which ordinary export is unavailable, check with your institution about what you are permitted to preserve before access ends.
Subject to institutional policy, it may still be useful to preserve selected conversations or your own research notes separately.
4. Save your Custom Instructions as plain text
Custom Instructions are an important asset independent of the individual conversations.
They may encode preferences such as:
desired level of detail,
mathematical notation,
LaTeX macros,
programming languages,
writing style,
field-specific terminology,
and stable information that ChatGPT should take into account.
Uploading exported conversations into another account does not transfer Custom Instructions automatically.
OpenAI's documented conversation-transfer process also does not transfer memories, GPTs, files, settings, or other account configuration.
Save your Custom Instructions separately as plain text or Markdown.
5. Ask ChatGPT to summarize important Memory and create a bootstrap.md
Memory is more difficult to migrate.
ChatGPT Memory is not simply identical to your chat history, and OpenAI notes that the visible memory summary does not necessarily contain every piece of information ChatGPT may know about you.
Before leaving, ask ChatGPT something like:
What do you currently know about me?
Organize the long-term information that would be useful for reconstructing my current research-support environment in a new ChatGPT account.
Then consolidate the useful result into a Markdown file such as:
bootstrap.md
It might contain:
research fields,
current research topics,
ongoing projects,
programming languages and software,
LaTeX and mathematical notation preferences,
writing preferences,
important URLs,
persistent user preferences,
and other context that would be useful in future conversations.
This is not an official OpenAI Memory-export feature.
It is a practical workaround created by the user.
Nevertheless, it provides a simple and robust way of carrying the essential answer to the question:
“What should a new ChatGPT account know about me?”
6. Do not keep important research files only inside ChatGPT
ChatGPT should not be the sole storage location for research outputs.
Manuscripts, source code, numerical data, figures, research notes, CVs, grant applications, bibliography databases, and similar material should also live in systems independent of your ChatGPT workspace, such as:
Git,
institutional storage,
personally managed storage,
appropriate cloud storage,
and backups.
Access rights to a ChatGPT workspace should be separated from the preservation of the research outputs themselves.
This is a useful principle for institution-provided cloud services in general, not only for ChatGPT.
7. In the new account, provide both conversations.json and bootstrap.md
For accounts where an export is available, the downloaded archive may contain conversations.json, or numbered JSON files for larger exports.
OpenAI officially documents a process in which those exported conversation files are uploaded into a new conversation in another personal ChatGPT account and used as reference material.
It is important to understand what this does not do.
Uploading the JSON does not:
recreate all old conversations in the sidebar,
restore Memory,
restore Custom Instructions,
transfer GPTs,
or transfer workspace membership.
It is a reference mechanism rather than a complete account restoration.
A practical reconstruction therefore uses three things together:
conversations.json
bootstrap.md
your saved Custom Instructions
In the first conversation of the new account, you can say something such as:
These files are migration materials from my previous ChatGPT account.
Use bootstrap.md as long-term background information and consult the exported conversation JSON when historical context is needed.
If you were using a Business or Enterprise workspace and cannot obtain conversations.json, consider preserving the most important conversations individually as Markdown or plain text before your access ends, where permitted by your institution's information-management policy.
8. Re-establish the important Memory in the new account
Finally, explicitly recreate the information that you want the new account to remember over the long term.
Simply uploading bootstrap.md does not necessarily mean that every statement in that file will automatically become persistent Saved Memory.
For particularly important information, explicitly tell ChatGPT:
Please remember this information for future conversations.
You should also check the relevant Personalization / Memory settings.
Memory and Custom Instructions serve somewhat different purposes.
Stable, explicit instructions are often best placed in Custom Instructions, while information that should provide continuing personalized context across conversations can be stored through Memory.
Minimum migration package
At minimum, I recommend preserving three categories of information.
1. Custom Instructions
A plain-text copy that you control.
2. bootstrap.md
A compact summary of your research background, preferences, ongoing work, and other persistent context.
3. Conversation history
Use the official conversations.json export when available.
If the workspace cannot be exported, preserve essential conversations separately where institutional policy permits.
Research files themselves should additionally be maintained in storage independent of ChatGPT.
Official migration versus the bootstrap.md workaround
This distinction is important.
Exporting conversation data and uploading exported conversation JSON to another personal account is an officially documented OpenAI workflow for eligible accounts.
By contrast,
asking ChatGPT to summarize important personalization and Memory information into a bootstrap.md file is not an official OpenAI migration feature.
It is a practical workaround for the absence of complete personalization portability.
The official functionality and the workaround should therefore be clearly distinguished.
Why this matters for researchers
Academic employment is often not permanent.
This is especially true for:
PhD students,
postdoctoral researchers,
fixed-term faculty,
fixed-term research staff,
and visiting researchers.
Institutional accounts and affiliations may change every few years.
Research questions, expertise, writing conventions, coding style, and accumulated intellectual context do not reset when the affiliation changes.
A personalized ChatGPT environment should therefore be regarded not only as an institutional software account, but also as part of a researcher's portable personal research environment.
We already think about offboarding for email, Git repositories, reference managers, LaTeX environments, and research data.
It may now be time to think in the same way about the research context accumulated in AI systems.
Official OpenAI Help
The following OpenAI Help pages were checked when preparing this article.
OpenAI Help — Exporting your ChatGPT history and data
Exporting your ChatGPT history and data
OpenAI Help — Transfer exported conversations between ChatGPT accounts
Transfer exported conversations between ChatGPT accounts
OpenAI Help — Exporting data from a ChatGPT Edu workspace
Exporting data from a ChatGPT Edu workspace
OpenAI Help — ChatGPT Business: General FAQ
ChatGPT Business: General FAQ
OpenAI Help — What is ChatGPT Enterprise?
What is ChatGPT Enterprise?
OpenAI Help — Memory FAQ
Memory FAQ
Updates and disclaimer
ChatGPT workspace, Memory, and export functionality changes frequently.
This article reflects information checked on August 20, 2026.
Before leaving an institution, verify the current OpenAI Help documentation and your institution's own policies.
This page is not official OpenAI documentation.