How Long AI Chat Apps Keep Your Conversations After You Leave

• Only one of the five names an actual retention period, and it is five years
• Two state that conversations are used for training; one says nothing either way
• Deleting an account is generally a request under conditions, not a switch
• No policy in this set promises that conversation text is erased outright
• One publishes nothing readable before you register

The question is simple and almost nobody answers it in one place: if you stop using one of these apps and delete the account, what happens to what you wrote? We read five policies to find out. The answers are not identical, and the differences are larger than the marketing suggests.

The five answers, side by side

Three things this table shows that reviews do not

First, silence is a category of its own. Kindroid does not say it trains on conversations, and it does not say it refrains. That is not the same as a no, and it cannot be broken as a promise later. Every other line in its document is specific, which is what makes the gap visible.

Second, the most forthcoming policy carries the least comfortable number. Chai is the only one that quantifies retention, and the figure is five years past deletion. A company that publishes nothing may be keeping data longer. You simply would not know, and could not find out.

Third, deletion almost never means erasure. Across these documents it is a right to request removal under listed conditions, with carve-outs for legal obligations, anonymised copies and training archives. If you send something you would want gone completely, none of these policies commits to that outcome.

What anonymised actually means here

Three of these policies lean on the same idea: the text stays, your name comes off. Chai describes removing personal identifiers so that only the conversational content itself is used. Nomi describes archives that survive deletion with the link to the user removed. CrushOn describes training without linkage to individual users. Kindroid describes anonymised, aggregated copies persisting after deletion.

It is a reasonable arrangement and it is not nothing. It is also worth understanding precisely what it protects. Removing an identifier protects you from someone reading the archive and seeing your email beside the text. It does not remove the text, and it does not remove details inside the text — a first name, a city, a job, an anniversary, the specific thing you said at three in the morning. Those are content, not identifiers, and they are exactly what these apps are built to elicit.

The practical consequence is small but real: what you type is more durable than the account it was typed into. That is not an argument against using any of them. It is an argument for knowing it before week eight rather than after, which is roughly when people start writing things they would not put in a text message. If you are still choosing between platforms, the same documents differ on plenty else — we went through Character.AI's position separately, and the Candy.AI comparison covers a different pair on the same axes.

How to check any app in ten minutes

• Open the privacy policy before creating an account. If you cannot, that is the answer to your first question
• Search for the words retention, retain and delete, and read the whole clause rather than the heading
• Search for train and model. An absent answer is information, not an oversight to fill in generously
• Look for a number. Days, months, years. Most policies avoid one, and the ones that give it are telling you something the others are not
• Note the date on the document and the date you read it. Both change without notice

Everything above is quoted or paraphrased from the companies' own published documents on the dates shown. Where a document was unreadable without an account, that is recorded as unreadable rather than guessed at. If any of these have since changed, this page is out of date, and the read dates are printed so you can tell.