Private AI chat: the best private AI chatbot apps compared on their own policies

What makes an AI chatbot actually private?

Four things, in descending order of how much they matter. Whether your conversations are used to train models, whether they are used to target advertising, how long they are kept after you delete them, and who inside the company can read them. Encryption matters too, but it is the easiest promise to make and the least differentiating: everyone encrypts in transit, and it says nothing about what happens to the plaintext at the other end.

The reason this ordering matters is that training and ad targeting are irreversible in a way deletion is not. Once a conversation has contributed to a model's weights, no delete button unwinds it. Once a preference inferred from your chat has been written into an ad profile, it has already been acted on. Retention windows are recoverable by comparison. So a service that keeps chats for a year but never trains on them is meaningfully more private than one that deletes in 30 days and trains on everything first.

Why free AI chat is rarely private

Running inference is expensive. Every message costs the operator real compute, and that cost does not go away because the product is free. So a free service is funding those tokens somewhere else: advertising, model improvement that makes the underlying product more valuable, data licensing, or venture money on the assumption that one of those arrives later. None of that is sinister, but it does mean your conversations have to be worth something to someone other than you.

This is the clearest single filter you can apply. If a chatbot is free and the company is not visibly selling something else at scale, assume your conversations are part of the business model until its policy says otherwise in specific words. Subscriptions are not automatically virtuous, but they align the incentive: we get paid whether or not your chat is interesting to an advertiser, so there is nothing to gain from it being interesting to one.

The mainstream assistants score worst, and it is not close

Google states that Gemini conversations picked for human review can be retained for up to three years, and that those copies are not deleted when you delete your activity. Meta began using AI chat interactions for ad personalization across its apps on December 16, 2025, with no clean opt-out for US users. Character.AI offers a model training opt-out only in the EEA and UK, which means US users have no opt-out at all. These are not accusations, they are each company's published position.

The uncomfortable part is that these are the products most people use for the most personal questions, precisely because they are free, polished, and already installed. A search assistant is a reasonable place to ask about a tax form. It is a poor place to work through a medical worry, a relationship, or anything you would not want inferred into an advertising profile, and that mismatch is what this whole comparison is really about.

How to check any AI chat app yourself in five minutes

Open the privacy policy and search it for four words: train, improve, retain, and advertising. Language about improving services using your content is usually training in softer clothing. Then look for a number attached to retention. Companies that intend to keep data briefly say how briefly; companies that intend to keep it indefinitely write only as long as necessary. Finally, check whether any opt-out described is available in your country, because several of the biggest ones are regional.

If a policy will not load, or the company publishes no pricing and no clear content rules, treat that as information rather than as a neutral absence. It is not proof of anything bad. It does mean you are relying on marketing rather than on a commitment, and for conversations you would not want read back to you, that is the wrong thing to rely on.

Where we fit, honestly

Depravity is a paid, uncensored AI chat product for US adults 18 and over, from $15 a month, or $12 a month billed yearly. Chats are encrypted in transit and at rest, never sold, never shared with data brokers or marketers, and never used to train advertising. You can export or delete any conversation from settings yourself. There is automated and human review, and we disclose it, because it is how the hard lines get enforced.

What we are not: free, and not the most private option that exists in absolute terms. A locally run model on your own hardware beats every hosted service including ours, because nothing leaves your machine. It costs you a capable GPU, setup time, and a noticeably weaker conversation. If that tradeoff appeals to you, take it, and we would rather say so than pretend hosted chat is something it is not.

Does deleting a chat actually delete it?

Not always, and there is a documented case worth knowing about. In May 2025, a federal magistrate ordered OpenAI to preserve and segregate all output log data that would otherwise have been deleted, as part of the New York Times litigation. That included chats users had already deleted. The obligation ran until late September 2025 and was formally terminated in October 2025, but data preserved during that window stayed in the litigation discovery set.

For roughly four months, deleted did not mean deleted, and users were not asked. That is not a scandal so much as a reminder: a delete button is a promise from a company, and a company operates inside a legal system that can override the promise. Google is more explicit about a version of this. Its own Gemini help page says chats selected for human review are retained for up to three years and are not removed when you delete your activity.

So the useful question is not whether a service has a delete button. Every service has one. It is how much data the company holds that the button never reaches: human review queues, backups, moderation records, and anything already folded into a model. Ask about those four and the answers separate the products very quickly.

Why does privacy matter more for an uncensored chat?

Because candor is the entire point. An assistant you have to self-censor in front of is not an uncensored assistant, no matter what the filter policy says. If you are second-guessing every message because some part of you knows it is being logged, scored, and possibly used as training data, you are having a managed conversation, not an honest one.

Privacy and candor are the same feature seen from two sides. Remove the filters but keep the surveillance and you have not actually given anyone the freedom to speak. It is also why the free adult apps are the worst possible place to be candid: the more personal the conversation, the more valuable it is to whoever is monetizing it, and adult chat platforms collect exactly the categories of data that are most damaging to have leaked.

That is the standard this product is built to. No training on your conversations for advertising, no selling, no sharing with brokers, encryption in transit and at rest, and an export or delete you control. There is automated and human review to enforce the hard lines on minors, non-consent and illegal activity, and we disclose that rather than quietly leaving it out.