Does Character AI save chats?
By FlirtAlly · October 7, 2026
Does Character AI save chats?
Key Facts
- A Stanford HAI study found all six leading U.S. AI companies use chat data for training by default.
- Stanford researchers reviewed 28 policy documents and found privacy policies 'lack essential information about practices', calling the model deeply flawed.
- Muah AI's September 2024 breach exposed user accounts, conversations, and custom character descriptions, per industry analysis.
- ChatGPT Temporary Chats are deleted from OpenAI's systems after 30 days, according to reporting.
- Italy's data regulator found Replika — with over 10 million users — carries out 'unlawful processing of personal data', citing risks to vulnerable people.
- Google's previously saved media can remain up to four years after account disconnection, research shows.
- No source documents Character AI's chat storage policy, though it faces the October 2024 Setera lawsuit, highlighting litigation risk.
The Honest Answer: No Clear Policy, So Assume Yes
You deserve a straight answer, so here it is: after reviewing the available research, we found no authoritative public documentation of Character AI's chat storage, retention, or deletion policy. None of the six sources we reviewed — including academic studies, mainstream reporting, and industry analyses — directly addresses what Character AI does with your conversations.
That silence matters. A Stanford HAI study that reviewed 28 policy documents across the industry found that privacy policies "lack essential information about practices" and described the internet-era privacy policy model as "deeply flawed." When a platform doesn't clearly disclose how it handles chats, you can't assume the best.
So what's the prudent default? If you chat with an AI, assume it's saved. The evidence backs this up. The same Stanford study found that all six leading U.S. AI companies — Amazon, Anthropic, Google, Meta, Microsoft, and OpenAI — use users' chat data by default to train their models. Some keep user chat information in their systems indefinitely, with "long data retention periods" flagged as a cause for concern.
The AI companion space adds urgency. Industry analysis of the category offers blunt guidance: "Assume your chats can leak." The same analysis notes that users sharing intimate content with AI companions are entrusting platforms with data more personal than what most consumer software handles — and that the security infrastructure in the category hasn't matured to match that sensitivity.
Here's what that means in practice for anyone chatting with an AI girlfriend or companion:
- Opting out of training doesn't necessarily turn off memory or personalization — your chats may still be saved, as privacy reporting makes clear.
- Deleting a privacy setting doesn't delete previously collected data; some platforms require separate deletion steps.
- Privacy-protective modes do exist — ChatGPT's Temporary Chats are deleted after 30 days, and Meta's Incognito Chats aren't saved — but only where platforms explicitly offer them.
- Platforms vary substantially in security maturity, so choose ones that disclose their practices in detail.
To be fair to Character AI, the research notes it has not had a documented data breach in the conventional sense. But it is referenced in a lawsuit (Setera v. Character Technologies, filed October 2024), which highlights how user conversations can become relevant to litigation — another reason saved chats carry risk.
At FlirtAlly, we believe this honesty is the only workable standard for the companion category. If a platform won't tell you clearly whether your conversations are stored, trained on, or retained, treat every message as if a stranger could someday read it — because in this industry, absence of a policy is not absence of saving.
Why Chat Storage Matters More in AI Companion Apps
Ask someone what they'd type into a work chat app and they'll pause. Ask what they've confessed to an AI companion at 2 a.m., and the answer is usually far more intimate — which is exactly why chat storage in companion apps deserves sharper scrutiny than ordinary chatbot use.
A Stanford HAI study reviewed 28 policy documents across six leading U.S. AI companies — including OpenAI, Google, Meta, and Anthropic — and found that all six use users' chat data by default to train their models. Some developers, the study noted, keep chat information in their systems indefinitely. That's a concern for any conversation; it's a bigger one when the conversation is romantic.
The AI companion category has already shown what can go wrong. In September 2024, a documented breach at Muah AI exposed user accounts, conversations, custom character descriptions, and uploaded image references. Months earlier, in March 2024, Soulmate AI shut down entirely — and users lost the relationships and accumulated data they had paid for. And in January 2023, Replika's removal of NSFW content raised significant questions about how user data is handled when platform policies change overnight.
Regulators have taken notice. Italy's data protection authority, the GPDP, ordered a temporary limitation on Replika's data processing, finding that the company "carries out unlawful processing of personal data" and fails the principle of transparency, citing risks to minors and emotionally vulnerable people. With Replika reporting more than 10 million users, the ruling set a precedent that companion chat data is not treated like ordinary chat data.
Saved chats can also surface in court. The lawsuit Setera v. Character Technologies, filed against Character AI's parent company in October 2024, highlights the risk that user conversations — even ones users consider protected — may become relevant evidence in litigation.
