What is inside a ChatGPT export, and what should you review?

An export is a package of records, not a finished personal profile. Before uploading it to any tool, separate three questions: what files are present, what the conversations say and what you have permission to reuse. Those questions often have different answers. A file can be readable while still containing material that should remain private.

Begin with a file inventory

Open a working copy in your operating system’s archive viewer. Note the file types and approximate sizes without changing their contents. OpenAI describes its download as including chat history and other relevant account data. Exact structures can change, so inspect your actual archive rather than relying on a screenshot of someone else’s download.

Look for readable conversation material and structured records. Additional files may describe account information rather than conversations. Do not assume every file is necessary for your intended task. For example, recovering a list of books does not require uploading an entire account record or unrelated media.

Read a conversation as an exchange

Your messages and the assistant’s responses have different evidential roles. “I prefer quiet hotels” is a user statement. “You probably enjoy boutique hotels” is an assistant interpretation. A pasted paragraph beginning “my client prefers quiet hotels” concerns someone else. All three can occur in the same conversation.

Review the surrounding turns before extracting a preference. Requests may describe fictional characters, hypothetical budgets or a task performed for another person. A personal profile should not treat those as confirmed facts about you. When context remains ambiguous, keep the statement out or record it as an unresolved question.

Expect structure to matter

Some conversations contain edits, alternate responses or branching paths. A viewer that prints every message in a flat list can suggest a conversation that never happened in that order. Timestamps can also describe different events: creation, modification or an individual message. Do not silently turn every timestamp into the date of a life event.

For a small review, compare the visible conversation with its surrounding records. If you cannot determine which branch was used, avoid making a strong personal conclusion. Preserve the source reference so you can revisit it. Correct interpretation is more valuable than extracting the largest possible number of statements.

Separate valuable context from risky material

Identify content that could harm someone if disclosed: access keys, account recovery information, names attached to private disputes, health details, workplace documents and confidential customer records. Assistant replies can repeat those details even when the original prompt has been removed from a working copy.

Create a narrow derivative document for your project instead of editing the only archive. Keep paraphrases clearly labeled and distinguish them from quotations. If you plan to share the derivative, inspect its title, filename and supporting notes as well as its main text. Sensitive clues can survive outside the paragraph you intended to redact.

Use an archive without promising completeness

A snapshot does not prove that you have a complete history of your life or even every conversation you intended to preserve. Compare the archive with a few known projects and dates. Record gaps rather than asking a model to fill them with plausible stories.

The archive inspection described here is for your own source review. Persir uses a prompt summary rather than reading the archive: copy its ChatGPT prompt and confirm the returned summary once. Saved statements can then be inspected, corrected or deleted in Data. The provider may have only selective context, so do not treat the resulting summary as a complete account history or fill missing dates with guesses.