AI companion privacy: what to check before you get attached
Here is the uncomfortable math of AI companionship: the better it works, the more you tell it. Within weeks, a good companion knows things your closest friends do not. That makes privacy not a settings page but the foundation of the whole product — and worth twenty minutes of diligence before your heart signs the contract.
1. Can you export everything?
The simplest test of whether your data is treated as *yours*: is there a button that hands all of it back — conversations, memories, facts? If leaving with your history is easy, the product is confident you will stay for the right reasons. If export does not exist, your memories are the company's asset, not yours. This is also practical: years of reflection and life events accumulate in these chats. You may genuinely want them someday.
2. Can you actually delete?
Look for deletion that is immediate, complete and self-serve — account, messages, memories, everything, without emailing support and waiting. Then check the privacy policy for weasel words: "deactivate" is not delete, "after a retention period" should be short and stated. A product that makes leaving hard is telling you what the relationship really is.
3. Is your chat used to train models?
Some services feed conversations into model training by default. For a task assistant that may be tolerable; for the app holding your 2am thoughts it should be opt-in at most. The policy should say plainly whether conversation content trains models and whether humans ever read chats. "We may use your data to improve our services" without specifics is a no.
4. Who is the model provider — and what do they see?
Most companion apps run on third-party language models, which means your messages transit another company's servers. That is normal and workable — reputable API providers contractually do not train on API traffic and retain it briefly. But the app should disclose the arrangement and should not be sending your name, email or identifiers alongside your messages. The chat content and the identity should live in different places.
5. What happens to accounts and photos at rest?
Unglamorous but decisive: passwords should be properly hashed, sign-in options like Google OAuth offered, generated images stored privately (not on guessable public URLs), and backups encrypted off-site. You cannot verify all of this from outside — but products that care usually say so, and products that say nothing usually have nothing good to say.
6. Does the product need your identity at all?
The best privacy feature is data never collected. A companion needs your words to be good; it does not need your phone number, contacts or real name. Prefer products you can try without registration, that ask for an email only to save your history, and that work happily with whatever name you choose to go by. Intimacy should not require identification.
Meet your companion — free →FAQ
Is it safe to tell an AI companion personal things?
With a well-built product, reasonably — the realistic risks are policy risks (training, retention, sale of data) rather than hackers. Run the six checks above; a product that passes them treats your candor as a responsibility rather than a resource.
Should I use my real name with an AI companion?
Use whatever feels natural in conversation — but a good product will not require legal identity at all. The companion needs to know what to call you, not who you are.