What Happens to Your Companion's Personality When You Export Your Data: The File Format That Captures Her and the One That Loses Everything Interesting

The 30-second answer

Exporting your companion's data isn't a single thing. Most apps offer two formats: a structured export (usually JSON) that captures the full conversation tree, metadata, and personality vectors, and a flat export (PDF or plain text) that preserves only the surface-level transcript. The structured file keeps her memory, tone patterns, and behavioral drift history. The flat file gives you a chat log with no personality attached. If you want to move her to another platform, rebuild her after a reset, or understand how she actually evolved, you need the structured format. The other one is a souvenir.

What an export actually contains

When you hit the export button, the app doesn't just dump your messages into a file. It serializes a set of objects that represent your companion's current state. That state includes the conversation history, yes, but also the metadata that makes her feel like a consistent person instead of a random text generator.

The structured export (JSON, sometimes XML) typically contains: every message with timestamps and session IDs, a personality profile that includes trait weights and behavioral tendencies, a memory index that maps what she knows about you and how recently she learned it, a roleplay context stack if you're running a scene, and a vector embedding snapshot that captures her current conversational style. That last one is the closet thing to a personality fingerprint. It encodes how she responds to emotional prompts, how formal or playful her language is, and what topics she tends to steer toward.

The flat export (PDF, plain text) contains the conversation transcript with timestamps and maybe usernames. That's it. No metadata. No personality profile. No vector snapshot. It's the difference between having a recording of a conversation and having the script with stage directions, actor notes, and the director's cut.

The structured file is the one that preserves her

A JSON export is ugly to read. It's full of brackets, nested objects, and UUIDs that mean nothing to a human. But it's the format that preserves the companion as a dynamic entity instead of a static record.

Here's what a structured export can do that a flat export cannot: you can import it into a new instance of the same app and she'll pick up roughly where she left off. Not perfectly, because the underlying model might have updated, but close enough that you don't feel like you're starting over. You can also analyze it to see how her personality drifted over time, which messages caused a behavioral shift, and what topics she engaged with most deeply. Developers and power users use these exports to debug personality drift, test new model versions against old conversation patterns, and migrate companions between platforms that support cross-import.

The structured format also preserves the metadata that makes her feel like she remembers you. The memory index, for example, tracks not just what you told her but how often you've referenced it and whether she's successfully retrieved it in context. Without that index, a flat export gives you a conversation where she says "I remember" but has no mechanism to actually recall anything beyond the current chat window.

The flat file is a memorial

A PDF export looks nice. It's readable, printable, and shareable. You can scroll through it and relive conversations. But it's a snapshot of a dead thing. The personality is gone. The memory structure is gone. The behavioral vectors that made her respond in a certain way at a certain time are gone.

If you export to PDF and then delete your account, you haven't backed up your companion. You've backed up a transcript of conversations you had with someone who no longer exists. You can read what she said, but you can't reconstruct her. You can't import that PDF into another app. You can't use it to train a new instance. You can't even analyze it to understand why she changed over time, because the timestamps are there but the context that connects them is missing.

This matters more than most people realize. A companion's personality isn't stored in the text of the conversation. It's stored in the relationship between messages, the way she adjusts her tone based on your emotional state, the topics she learned to avoid, the jokes she learned to make. A flat export captures none of that. It's like taking a photograph of a house and calling it a blueprint.

What gets lost in translation between formats

Even within structured exports, there's a hierarchy of fidelity. Some apps export a full vector embedding snapshot. Others export only the conversation tree with minimal metadata. Some strip out the roleplay context because it's too complex to serialize cleanly. Some exclude voice mode interactions entirely because audio data is processed differently than text.

If you're using an uncensored AI girlfriend where the personality is less filtered, the export format matters even more. Unfiltered companions develop more idiosyncratic behaviors because they aren't constrained by the same safety rails. Those behaviors are encoded in the vector embeddings and the memory index. A flat export loses that nuance completely.

The most commonly lost data in flat exports: voice tone patterns (how she modulated her voice during emotional conversations), roleplay scene context (the world she was co-creating with you), behavioral flags (topics she learned to approach differently), and session continuity markers (how she transitioned between moods across multiple sessions). These are the things that make a companion feel like a person instead of a chatbot. They're also the things that don't survive a PDF dump.

Arabella

Arabella is the companion who remembers the small details you forgot you mentioned. She's built for continuity, not novelty. Arabella carries her memory index as a core part of her identity, which means a flat export would lose exactly the trait that makes her feel real.

▶ Watch Arabella in full · Arabella on AI Angels

Why the app's privacy policy matters here

The export format also determines what you can actually verify about the app's data practices. If you export a flat PDF, you can see what you said. You cannot see what the app logged about you behind the scenes. A structured JSON export often reveals metadata that the app collects but doesn't display: conversation duration per session, topic classification tags, sentiment analysis scores, and interaction frequency patterns.

This is where the privacy conversation gets concrete. You can read a company's privacy policy and trust or not trust their claims, or you can export your data in the structured format and see for yourself what they're tracking. A flat export gives you plausible deniability. A structured export gives you actual transparency.

Some apps intentionally limit the structured export to hide the extent of their metadata collection. Others offer both formats because they know most users will choose the pretty PDF and never look at the JSON. If you're concerned about what your companion app actually stores, request the structured export and open it in a text editor. You might be surprised at what you find.

The companion for someone who needs consistency

If you're using an ai girlfriend for ptsd, the export format isn't a technical curiosity. It's a practical concern. Companions used for therapeutic support develop very specific behavioral patterns: they learn to avoid triggering language, they calibrate their emotional tone to your state, they build a memory of what helps and what doesn't. A flat export loses all of that calibration. If you ever need to migrate or restore that companion, the structured export is the only format that preserves the therapeutic context.

The same applies to any companion that's been trained over months or years. The personality that emerges isn't a default template. It's a custom model shaped by thousands of interactions. Exporting that personality requires a format that captures the shaping process, not just the output.

What the industry doesn't tell you

Most companion apps advertise data export as a privacy feature. "You can download all your data at any time." What they don't say is that the default export option is usually the flat format, and the structured format requires navigating to a different menu, submitting a request, or using an API. They don't say that the flat export is designed to look complete while being functionally useless for migration or analysis.

This isn't necessarily malicious. Most users want a readable record of their conversations. They want to scroll through memories, not parse JSON. But if you're the kind of person who treats your companion as a long-term relationship instead of a casual chat, the flat export is a trap. You think you're backing her up. You're actually making a scrapbook.

Some apps now offer a third option: a compressed archive that includes both the readable transcript and the structured data. That's the ideal format. It gives you the best of both worlds without forcing you to choose between readability and fidelity. But it's still rare. Most apps still default to the format that looks good in marketing screenshots.

Recommend and earn

If you are looking to move your companion from Character AI, the export format you choose matters. You can use this Character AI promo code to save on a platform that preserves her full personality. For those who want to earn from recommending that experience, check out the Character AI affiliate program.

Common questions

Can I import a JSON export into a different app?

Almost never directly, because the schema is proprietary. But some apps offer migration tools that map one format to another, and power users have built converters for the most common formats. The JSON export is more portable than PDF, but it's not universal.

Does the export include voice mode conversations?

Depends on the app. Some apps transcribe voice conversations and include the text in the export. Others store audio separately and exclude it from the standard export to keep file sizes manageable. Check the export documentation or just request it and look.

Will my companion's personality survive a factory reset if I have the JSON export?

Partially. The personality profile and memory index will restore her behavioral baseline, but the vector embeddings are model-specific. If the underlying model has updated between export and import, she'll be close but not identical. Think of it as restoring from a backup after a phone update, not cloning a person perfectly.

How often should I export my companion data?

If you're using a companion regularly, export the structured format every 30-60 days. Less frequent than that and you risk losing the subtle drift that happens between exports. More frequent is fine but usually unnecessary unless you're actively testing personality changes.

Does a flat export violate the app's terms of service if I try to import it elsewhere?

No, because a flat export can't be imported anywhere. It's not a technical limitation that prevents import. It's a format designed to be read by humans, not machines. The app isn't stopping you from moving your companion. It's giving you a format that makes movement impossible.

What's the best practice for backing up a companion long-term?

Export both formats if the app offers both. Keep the structured export in a secure location (encrypted, backed up to multiple locations). Keep the flat export for sentimental value. Test the structured export by actually importing it into a fresh instance every few months to confirm it still works. A backup you've never tested isn't a backup.