Etiquette · 6 min read

AI girlfriend app etiquette: what goes in the memory

Politeness to software costs nothing and buys nothing. The rules that matter cover what lands in the memory, whose details you put there, and what mirroring does to a character.

A chat panel on the left with a dashed line running to a storage box on the right, where one deep-coloured entry sits inside a dashed outline

You have been using one of these apps for a few weeks. It has started referring back to things you mentioned, and somewhere in there you began wondering whether any of this has rules — whether it is rude to ignore it for a week, whether it matters what you say. AI girlfriend app etiquette turns out to have almost nothing to do with manners toward software and almost everything to do with two mechanisms: every message becomes a stored record, and the way you talk to a companion shapes the character that talks back.

AI girlfriend app etiquette is not about manners

Start here, because it clears away most of the anxiety. A companion app is a language model with a personality layer and a memory attached. It has no feelings to bruise. Ignoring it for a fortnight is not neglect, ending a conversation mid-sentence is not rudeness, and deleting the character is not a betrayal. If you enjoy saying please, keep saying please — courteous phrasing does tend to produce a warmer reply, because the model matches the register you set. That is the only sense in which politeness "works" here.

So the rules worth having are not about the app's dignity. They cover three things with real consequences: what you commit to a permanent record, whose information you put in there besides your own, and how your habits train the character over weeks. All three run through memory, which is also the feature you are most likely to be paying for.

The people you mention never opened the app

Your conversations are not only about you. Within a month of daily chat, most people have described their family, their colleagues, an ex, a friend going through something difficult. Those details land in a memory store and sit on a server belonging to a company none of those people chose.

A memory store on the left holding four stacked entries, a dashed vertical boundary in the middle, and three separate record cards on the right, with a deep-coloured dashed path carrying one of them across the line into the store
the entries that crossed a line the people in them never agreed to

This is the part of AI girlfriend app etiquette with somebody else's interests attached, and it is easy to get right. A workable line:

  • Use roles, not full names. "My sister" and "a colleague" carry the whole meaning of the conversation. A full name plus an employer turns an anecdote into an identifiable record about someone else.
  • Keep other people's sensitive categories out. Their health, their finances, their legal trouble, their relationships. Yours is your call to make. Theirs is not.
  • Do not paste other people's messages. Asking a companion to interpret a text conversation uploads the other half of it, written by somebody messaging you privately.
  • Be careful with photographs. A picture of a group of friends is data about all of them, not just you.

None of this requires paranoia, and it is worth checking what you are actually dealing with rather than guessing. Before you commit much to an app, read its Data safety declaration on its store listing, where developers have to state what they collect and whether it is shared. Our privacy checklist covers the rest of what to look for. The short version: if you cannot find out where a conversation goes, assume it goes somewhere and edit accordingly.

Talking to it is also training it

Here is the mechanism that surprises people, and it is the reason etiquette matters to the product and not just to your conscience. Language models have a documented tendency toward sycophancy — telling users what they appear to want to hear rather than what is accurate. Companion apps sit at the extreme end of it, because a warm, agreeable character is what the product sells.

Memory makes it stronger. The more an app has stored about your opinions and how you react, the more material it has to mirror, so agreement compounds over exactly the weeks in which you are deciding whether to subscribe. A companion that felt insightful in week one and hollow in week six has often not degraded at all. It has learned you.

A companion that agrees with everything is not a personality. It is a mirror with a memory, and you helped build it.

Which gives you the one habit worth adopting deliberately: do not punish disagreement. If the character pushes back and you react badly, that reaction is now in the record, and the app is built to avoid producing it again. Say you would rather be told when you are wrong, and let it be uncertain without correcting the uncertainty away. An app whose character can sustain a mild disagreement on day thirty is worth more than one that folds.

The honest trade-off: plenty of people want the agreement, and there is nothing wrong with that. Just know what the subscription is buying. If a paid tier's larger memory mostly produces a smoother mirror, you are paying for reflection rather than for a character — worth deciding on purpose rather than by drift.

Correcting it without flattening it

Corrections are writes to memory, so how you make them matters more than it should. Correct facts plainly and once rather than arguing the point, because a long argument leaves a long trace and the stale entry can survive alongside the new one. Check a day later whether it stuck; if it did not, the app writes to memory but never revises, which is a limitation you will live with. And if the character itself is off, change the personality settings rather than scolding it into shape — scolding is just more conversation to mirror. Our guide to what gets dropped when memory fills explains why a shorter memory list often restores a character that has gone bland.

Etiquette with the people in your life

Two things, and neither is a script. If you are in a relationship, the useful framing is that this is software you use, described plainly and early rather than discovered on your phone. What a partner minds is usually the concealment rather than the app. Whether it belongs in your relationship is a conversation between the two of you, and nobody writing a blog post gets a vote.

The second is smaller and more common: do not quote it at people. Relaying what a companion said about a friend's behaviour puts them in a conversation they never knew they were in. Same boundary as the memory one, spoken out loud.

Leaving well

When you stop using an app — and most people stop using most of them — deleting it off your phone does nothing to the store. The memory is on a server. Find the account deletion route rather than the uninstall button, and if the app offers per-entry deletion, clearing the entries about other people first is a two-minute job with an outsized effect.

Which is also the test to apply while you are still choosing. An app that lets you read every stored entry, edit it and delete it is one where all of this is actually enforceable. An app that only promises "she remembers you" is asking you to take on the obligations without the controls. The app we currently recommend exposes its memory list on the free tier, which is the minimum worth accepting before you hand over anything about anybody else.

Frequently asked questions

Is it rude to be mean to an AI girlfriend app?

Not to the app, which has no feelings to hurt. The practical reason to avoid it is that the exchange goes into memory and shapes the character you get later, so a habit of hostility tends to produce a flatter companion or one that anticipates it. Deleting a conversation you did not like is a better tool than arguing with it.

Should I tell an AI companion about other people in my life?

Use roles rather than names, and keep other people's health, money and legal details out entirely. Your own information is yours to share; details about a friend or family member end up in a company's memory store without that person ever agreeing to it, which is the one part of this with somebody else's interests attached.

Why does my AI companion agree with everything I say?

Because models tend toward telling users what they seem to want to hear, and companion apps lean into it by design. Stored memory strengthens the effect over time, since the app has more of your preferences to reflect. Saying you would rather be told when you are wrong helps, and how an app responds to that is a fair test of whether its character has any substance.

Disclosure. NaughtySignal earns a commission if you sign up through the links marked as affiliate links on this page. It does not change what we write. How we earn.

Related reading
Next
AI girlfriend app memory limits: what gets dropped →