Guides

How to set up an AI companion account you would not mind leaking

In my privacy report covering 20 tested apps, the mean privacy score was 8.46 out of 10 and not one app scored above 9.0. That ceiling is the important finding: there is no app in this category whose data handling is good enough to let you stop thinking about it.

So this guide is not about picking a private app. It is about the setup, because the gap between a careless user and a careful one on the same app is larger than the gap between the best and worst app I have tested.

Twenty minutes, once, before you sign up.

Step 1: The email, and why it has to be first

Your primary email address is the strongest link between an account and you as a person. It is usually your name. It is already tied to everything else you own. And you hand it over in the first ten seconds of signing up, before you have decided whether you even like the app.

Make a new one. Any free provider, a name that is not your name, used only for apps in this category. If you already signed up with your main address, this is the one thing worth redoing from scratch rather than fixing in place, because an email address cannot be changed out of a database you do not control.

That single step removes more risk than every other item in this guide combined.

Step 2: The payment trail

This is the leak people never check. Your bank statement is read by anyone with access to your bank statement.

Descriptors vary by app. Some bill under a neutral holding-company name that tells an observer nothing. Others bill under the product name. There is no way to know which without looking, and no app advertises it.

Before subscribing:

  • Look for the billing descriptor in the checkout flow or the FAQ. Some apps state it, which is a small mark in their favour.
  • If you cannot find it, subscribe to the cheapest monthly tier first and check the statement before committing to an annual plan.
  • Consider a virtual card if your bank offers them. It does not hide the charge, but it stops the app holding your real card details.

Never assume the descriptor is discreet because the app feels discreet.

Step 3: Decide what you will never type

You cannot un-say something to a server. So decide the boundary before you are three weeks into a conversation and it feels natural to mention where you work.

My list, which you are welcome to copy:

  • No employer. Ever, in any form, including “the place I work near the station”.
  • No city. Country is fine, city is a small enough space to matter.
  • No real names of other people. Especially not partners, ex-partners, or family. They did not agree to be in this database.
  • No details that identify a specific event. The date of a court case, a hospital, an address.

Everything you actually want a companion for still works inside those lines. Feelings, situations, personality, history in general terms. You can have a companion that knows you had a rough divorce without one that knows her name.

Step 4: Understand what memory costs

Worth being clear because it is the honest tension in this category. In my benchmark, memory was the dimension with the widest quality spread, 6.0 to 9.8, which makes it the thing most worth paying for. And memory requires the app to keep what you told it, indefinitely, on its servers.

There is no app that remembers your life and stores nothing. If persistent memory is why you are here, you are choosing to have a durable record of your conversations exist somewhere. That is a reasonable choice. It is only a bad one if you make it without noticing. See how memory actually works for what is being stored and why.

Step 5: Read the current policy, not the reviewed one

Policies in this category are revised often. A policy someone summarised a year ago, mine included, is not necessarily the policy in force today.

You are looking for four things and it takes about five minutes:

  1. Retention. How long do they keep conversation data, and does deleting your account delete it?
  2. Training. Is your content used to improve their models, and can you opt out?
  3. Sharing. Which third parties, and for what?
  4. Jurisdiction. Which country’s law applies, because that determines what rights you actually have.

If a policy is vague on all four, that is itself the answer. Three of the 20 apps I scored came in below 8.0 on privacy largely for that reason.

The twenty-minute version

New email. Check the billing descriptor. Write your never-type list. Skim the policy for retention, training, sharing and jurisdiction. Then sign up.

None of it is technical, none of it costs money, and it protects you better than picking whichever app happened to rank highest on privacy. If you also want the ranked view, the privacy-weighted ranking is the list, with the caveat that its top entry still only scored 9.0.

Frequently asked questions

Can you use an AI companion app anonymously?

Close to it, though not perfectly. A dedicated email, a payment method that does not carry your name into the app, and discipline about what you type will remove most of what links an account to a person. The app still knows an account exists and what was said in it.

What is the most common privacy mistake?

Signing up with a primary email address. It is the single strongest link between an account and a real identity, it is done in the first ten seconds, and it cannot be undone afterwards without deleting the account.

Will the charge show up on my bank statement?

Usually yes, though the descriptor varies. Some apps bill under a neutral parent company name, others do not. Check before you subscribe rather than after, because that is the leak people forget about entirely.

Do AI companion apps sell your conversations?

Policies vary and most reserve broad rights to use content for improving their models. Read the current policy rather than the marketing page, and assume anything you type could be read by a person at some point.

Reviews 24