Does Character AI read your chats? The honest answer

What the records do not answer is the version of the question most people mean: does a person at the company sit and read what you wrote? Nothing on file describes staff access to chats, how often it happens or under what rules. I won't fill that gap with guesses, in either direction.

Here is what is actually documented about your chats:

• Data you share is kept to help train and improve the AI models
• People in the EEA and UK have a switch to leave that training
• Flipping it doesn't cover everything: the company can still use data for search, recommendations and safety classifiers
• Deleting your account is "permanent and cannot be reversed," with no time window given

My take: I would treat every Character AI chat as not fully private. That is not a reason to panic, and the app is legit, run by Character Technologies, Inc. It is simply the honest default when a company says plainly that it learns from what you type and gives most of the world no switch to stop it.

If you want to reduce what sits in your history today, my guide on how to clear out your Character AI chats walks through that part. The table right after this section puts Character AI next to the companion apps I cover, so you can see what each one offers at a glance.

What is published points to machines more than people. The company says data may be used for "safety classifiers," which are automated systems, and for "search, recommendations." That tells you your words are processed in several ways beyond producing a reply. It does not tell you that someone is scrolling through your stories, and I would not claim that either.

My advice is to write as if a stranger could read it one day. Not because anyone is known to, but because the policy leaves the door open and gives you no way to check. The records also describe no tool for seeing what the company has stored about you, so you cannot audit it yourself. Given how personal role-play can get, that is a gap I would like closed.

• On the web: Settings > Account > Manage Account & Data > Model Improvement
• In the mobile app: Settings > Data & Privacy
• Untick "Improve the Model for Everyone" in either place
• Keep the limit in mind: search, recommendations and safety classifiers can still draw on your data

I like that the switch exists and that the company names it clearly. I don't like that it stops at the EEA and UK border. If you live in the US or anywhere else outside those regions, there is no opt-out on record at all, and that bothers me more than anything else about this app.

For readers outside those regions, the practical choices are simpler: share less, delete what you don't need to keep, or choose a different app. Nothing in the paid plans changes this, which I cover further down.

For the full picture of the app beyond privacy, including memory, voices and plans, read my Character AI review, privacy section included.

The bigger point is who else handles your words. From what is documented, that is the company itself, through model training, search, recommendations and safety classifiers. So when you ask whether people can read your chats, the realistic answer is that the company's systems can, and the policy says so.

Here is what you can control, as far as the records go:

• What you type in the first place, which is the one filter that always works
• The training opt-out, for anyone based in the EEA or UK
• Deleting chats you no longer want kept
• Deleting the whole account, on the web under Profile > Profile Settings > Account > Manage Account & Data > Delete Account, or on mobile under Profile > Settings icon > Remove Account

I'd lean on the first item hardest. Keep real names, your address, your workplace and photos of yourself out of any character chat. Role-play works just as well with invented details, and you give away nothing you cannot take back.

One more caution on account deletion. It is "permanent and cannot be reversed," and the help article gives no time window for when the data is actually gone. So deletion is a real exit for the account, but not a promise about timing, which is something I would like to see fixed.

Until then, I'd treat deletion as the last step of leaving, not the first. Cancel any paid plan, save what you want to keep, clear the chats you care about, and only then remove the account for good.

If Character AI's approach to data is a deal-breaker, other apps are worth a look, though each has its own policy to read. My ranked list of Character AI alternatives sorts them by what they do better. The box right below sums up OurDream AI, one of the apps people often compare it with.

The paid plans sell comfort and memory: (c.ai) lite costs $4.99 monthly for fewer ads and 30 memory pins, and c.ai+ costs $9.99, drops ads and slow mode, and adds unlimited premium chat styles and premium support. None of those plan descriptions mentions data use, training or privacy.

So the training statement applies to free and paying users alike, and the opt-out still depends on where you live, not on what you pay. I actually think that is the fair way round. Privacy should not be a paid upgrade, and nothing suggests it is one here.

Memory pins are worth a thought in this context. They exist so a character keeps the facts you choose, which means you are deliberately handing the app a list of things to hold on to. That is fine for plot points and character traits.

I would not pin anything about your real life, because a pinned fact is exactly the kind of detail you are asking the system to keep close.

Other reviewers report strong recall of buried facts, alongside invented details stated with the same confidence. That cuts both ways for privacy. A model that remembers well is great for stories, and that same memory is why I'd leave the real you out of them.

If you are paying and want to leave over this, the exit rules matter. Renewals are automatic, charged to your payment method until you cancel, and the company treats payments as non-refundable unless the law says otherwise. Paid with Apple or Google? Then any refund request goes to them, not to Character AI.

My habit is to cancel with a few days to spare before renewal, never on the last day.

For cancel steps, data questions and billing rules on other apps in the same plain format, my help pages for companion apps collect them in one place. Reading the one for any new app before you share anything personal is the habit I'd recommend most.

Health details you would not tell a new acquaintance. Photos of yourself or people you know. Anything that would be awkward if it turned up in a recommendation later.

Everything else is fair game for storytelling. Character chat is at its best when you build a scene, and a scene does not need your postcode. A made-up name and a made-up city give a character just as much to work with.

I also like a simple test before you hit send: would you be comfortable if this line sat in a dataset, with your account attached? If the answer is no, rephrase it or leave it out. That habit costs nothing and protects you on any app, not only this one.

Voice deserves a mention too. Character AI lists voice calls as a feature, and the records say nothing specific about how voice is stored or used. I'd apply the same rule out loud as in text: speak as a character in the story, not as yourself with your real details, and skip anything you would not want recorded.

Clear but strict beats vague and friendly in my book. An app that says "we train on your data, here is the switch for some regions" beats a silent one for me, because at least you know where you stand. What costs Character AI points with me is the regional limit on the opt-out and the missing time window on deletion.

If you want the full method behind my ratings, my notes on how I read app records explain what counts and why.

What the records don't give is how long the data takes to disappear on the company side. No window is stated.

For single chats, the records hold no retention detail at all. My working assumption is the cautious one: a deleted chat disappears from your screen, but anything already used to train or improve the models under the stated policy may not be pulled back out. I can't confirm that either way from the records, and I'd rather you plan for the stricter case.

That is why I put so much weight on what you type in the first place. Deletion is a tidy-up, not an undo. And if you are on a paid plan, cancel it before deleting anything, because the renewal runs on its own schedule and nothing on file says it stops when your chats or your account are gone.

It is not, from the records, the same as a person reading along, and I would not describe it that way.

What it does mean is that trying to push past the rules is a bad plan. Evading an app's policy breaks its terms and can get an account banned. If you want grown-up romance, the honest move is to pick an app that allows it.

On this page, SpicyChat, Nomi AI and OurDream AI allow adult content, and Candy AI and GirlfriendGPT offer it on paid plans only.

My guide on whether NSFW could ever come to Character AI covers the policy side in more depth, including why the rules are unlikely to loosen for paying users.