The direct answer is to ask whether you can export a complete, readable writing record and restore it without losing source evidence, revision history, permissions, or approval state. A downloadable text file is useful, but it is not a complete recovery plan if the workflow depends on prompts, source maps, templates, connectors, and publishing metadata.
What exactly is included
Ask the provider to identify which objects are exported: conversations, prompts, system instructions, uploaded files, source links, citations, drafts, comments, revision history, tags, templates, custom terminology, user roles, audit events, and publishing metadata. Request a sample export before committing critical work.
Check format and structure. Plain text or Markdown may preserve content but not comments or approvals. JSON may preserve relationships but require tooling. Proprietary packages may be complete yet unusable without the original service. Prefer documented, versioned formats and stable identifiers that let you map a draft to its sources and final publication.
Can the export be restored
Ask whether the same provider can import the export into a clean workspace and whether another tool can read it. Determine what must be rebuilt manually: users, permissions, connectors, secrets, scheduled jobs, and public URLs. Test a restore in a separate non-production workspace with a small representative project.
NIST SP 800-34, contingency planning guidance, emphasizes recovery procedures and validation before returning a system to operation. Define expected results, actual results, success status, and performer for the restore test.
How are sources and approvals preserved
Open restored citations and confirm the relationship between each claim and source survived. Check that opinion labels, factual-review notes, rejected drafts, approver identity, approval time, and publication status remain distinguishable. If an export flattens everything into final prose, it cannot by itself prove the editorial path.
The NIST AI Risk Management Framework applies governance and trustworthiness across AI design, use, and evaluation. Preserve enough context to understand how the output was produced and who accepted it, while avoiding unnecessary retention of sensitive input.
What is excluded or delayed
Ask about deleted items, temporary uploads, logs, media, linked cloud documents, external sources, analytics, billing records, and data held by subprocessors. Determine whether export is immediate or takes days, whether account closure starts a separate process, and how long download links remain active. Verify encryption in transit and at rest, access logging, and who can request a full export.
Do not assume an export deletes the provider's copy. Backup, portability, retention, and deletion are separate controls. Document each one.
How often and where will backups live
Set an interval based on publishing volume and recovery tolerance. Keep at least one copy outside the primary provider under owner-controlled access. Encrypt sensitive exports, limit administrators, record checksums, and avoid storing the decryption key beside the archive. Define retention so obsolete drafts and private source material do not accumulate forever.
Prove recovery end to end
Restore one article with its sources, metadata, review notes, and publication state. Render or open it, follow its links, compare a checksum where practical, and verify permissions. Then simulate one missing connector and confirm the runbook identifies the owner and replacement steps. Record the recovery time and every manual gap.
The workflow is portable enough only when an authorized person can retrieve the archive, understand its structure, restore an approved sample, verify the evidence, and resume work without relying on a departed employee or an undocumented provider feature.
Sources
- SP 800-34 Rev. 1, Contingency Planning Guide for Federal Information Systems — National Institute of Standards and Technology, May 1, 2010
- AI Risk Management Framework — National Institute of Standards and Technology
- Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile — National Institute of Standards and Technology, July 26, 2024