ChatGPT and Claude exports: how to compare them fairly

Two conversation exports can look similar while representing different periods, account types and structures. Comparing them fairly starts with coverage, not file size. A larger archive does not automatically contain more useful personal evidence, and a smaller one does not mean that a provider remembers less about you.

Compare the scope first

Create a simple private comparison sheet with source, account type, request date, download date and the period you can verify. Include known gaps. Both providers offer official export guidance, but availability and organization controls differ; follow the linked instructions for each account.

Choose a few comparable projects or dates and check whether they appear. Avoid treating archive size as a count of meaningful conversations. Files may contain different representations, metadata or media references. If you cannot confirm comparable coverage, describe the comparison as partial rather than presenting precise conclusions from unmatched histories.

Compare messages by role

Your messages are evidence of what you wrote in a particular context. Assistant responses are generated material that may summarize, suggest or speculate. Keep those roles distinct across both archives. A provider that produces longer replies should not appear to show more of your interests merely because its assistant wrote more words.

Also review whether the conversation was personal, hypothetical or performed on someone else’s behalf. “Plan a trip for my parents” should not be merged with “I prefer slow travel” as if both express your own holiday choices. Topic matching requires context, not just shared keywords.

Compare structure and readability

Check whether the files preserve message order, edits, branches and available timestamps. Different viewers can flatten structures differently. Test a few known conversations in a readable form before using a combined dataset for review.

If you convert records into a table, keep source and speaker fields. A row without provenance becomes difficult to correct later. Store source references privately and use neutral project labels if the table will be shared. Do not publish conversation titles that reveal clients, health concerns or account information simply to demonstrate your comparison.

Compare processing limits at the destination

An independent importer may apply different compatibility rules or common size and date limits. Read its coverage report for each source rather than expecting matching imports from unmatched archives. A narrow supported window can be deliberate, and excluded messages should remain excluded rather than becoming invented history.

Persir offers the same prompt-summary flow for ChatGPT and Claude, with provider-specific prompts and shared validation. Confirming a second provider adds information without replacing the first provider or typed answers. The available provider context is selective, so neither summary proves complete historical coverage. Full exports are outside this flow; keep them for independent backup and inspection if useful.

Combine only after reviewing differences

If the same preference appears in both archives, keep both references but do not double its importance automatically. The conversations may concern the same single event. If statements conflict, compare dates and context and ask what represents your current position.

A useful comparison ends with a practical decision: keep separate archives, create a small combined project brief or import supported material for private review. It should not produce a score for which provider “knows you best.” The goal is a more accurate record that you can understand, correct and remove when it no longer serves a purpose.