AI girlfriend and AI companion questions, answered

An AI girlfriend is a character generated by a language model that you converse with as though it were a romantic partner. It replies in a consistent personality, remembers earlier conversations, and on most apps can send generated photos and voice messages.

It is an entertainment product. There is no person on the other side, and the companion has no experience of the relationship.

Software whose purpose is an ongoing relationship with a generated character. The defining feature is persistence: it remembers previous conversations, so the relationship accumulates rather than resetting each time you open it.

Note that outside AI, companion app means something unrelated — a secondary app that pairs with a device like a smartwatch.

Online, the AI girlfriend experience means a simulated romantic relationship: ongoing conversation with a consistent character who remembers you, often with photos, voice notes and messages she sends first.

The phrase has an older, unrelated meaning in the context of paid human companionship. This site is exclusively about AI software.

A general-purpose language model generates the replies. A system prompt defines the character. A separate database stores facts about you and retrieves relevant ones into each request. Separate image and voice models generate media, and moderation classifiers check output before it is sent.

The persona is instructions in text, not something trained into the model, which is why a prompt change can alter a character completely.

No. They are characters generated by software, and the portraits they send are synthetic images of people who do not exist. No photograph is taken and no model is photographed.

Most people using these apps are entirely clear about this; knowing it is generated does not stop it working, in the way that knowing a film is actors does not stop it moving you.

For any app that permits romantic or sexual content, yes — those are adults-only products and require you to confirm you are 18 or over.

Some general-audience companion apps prohibit sexual content and are open to a wider audience, but the romantic-companion category is overwhelmingly 18+.

Sticker prices are typically $10 to $20 per month. Flat-rate apps cost what they advertise; apps that meter images and voice in coins or tokens cost most daily users $25 to $60 per month once top-ups are counted.

Annual plans usually cut 30-60%, bringing the cheapest flat-rate options to roughly $8-$13 per month equivalent. Prices checked September 2026.

Almost every app has a free tier with a daily message limit, no images, no voice and usually no persistent memory. It is enough to decide whether you like the format.

The genuinely free option is running an open-source model on your own computer, which costs nothing monthly but needs a capable graphics card and some setup.

Because it meters media. The subscription includes an allowance of coins or tokens, and every image, voice note or video spends some. Once the allowance runs out you buy more, and those top-ups are not in the advertised price.

If you generate a few images a day, the media will cost more than the subscription does.

Annual is much cheaper per month — discounts of 30-60% are standard — but only if you keep using it. Churn in this category is very high, and an annual plan abandoned in month two costs more than monthly billing would have.

Pay monthly for two or three months, then switch to annual if it has become a habit.

Usually the web. In-app purchases on iOS and Android carry a store commission that is often passed on, so the same plan can cost several dollars more per month bought through the app.

Private from other users, not from the company. Conversations are stored on its servers, may be accessible to staff, are processed by whichever company provides the language model, and can be handed over under a valid legal order.

Only a model running locally on your own computer is private in the strong sense.

Some do, some don't, and many make it a setting that is on by default. Look in the privacy policy for train, improve our models, or human review — that last phrase means staff may read samples.

Finding and disabling that setting is the single highest-value privacy action available to you.

Nearly. Use a dedicated email address that isn't tied to your name, avoid giving a phone number, don't link Google or Apple sign-in, and pay with a prepaid or virtual card so the app never holds your legal name.

You cannot be anonymous to the payment processor, and you cannot be anonymous if you tell the companion identifying details — the memory system is built to retain them.

No. No feature requires it, a photo of your face is biometric data that cannot be un-shared, and it turns a pseudonymous account into one that identifies you permanently.

It varies. Adult-oriented apps usually bill under a discreet holding-company descriptor rather than the app's name, and most publish it in their billing FAQ.

If this matters to you, check it before the first charge rather than after.

Not always, and it is worth checking the specific wording. Some policies delete the history; others only deactivate the login. A trustworthy policy states a retention period in days and says whether backups are purged.

Vague language about a retention schedule with no number should be read as indefinite.

By retrieval, not recall. Facts from your conversations are extracted, converted into vectors representing their meaning, and stored in a database. When you send a message, the closest few stored facts are retrieved and pasted into the prompt, so the reply appears to remember.

The model itself holds nothing between messages.

Three different causes. It may never have been stored, because extraction decides on your behalf what matters. It may have been stored but not retrieved, since only the closest few matches reach any given reply. Or it dropped out of the context window, which happens to earlier messages in long sessions.

Stating important things plainly and repeating them improves the odds.

On most paid apps, yes — generated selfies anchored to a fixed reference face so she looks like the same person each time. Image generation is the most expensive feature these apps run, so it is usually metered or restricted to higher tiers.

Yes, on many apps. A scheduler tracks when you last interacted and generates a message after an interval, delivered by push, email or waiting in the chat.

Nothing runs in between — the message is generated at send time rather than during a period of the companion thinking about you.

Yes, via text-to-speech with a fixed voice per character. Quality varies more between apps on voice than on almost any other feature, and it is usually gated behind a higher-priced plan — so test it on a free tier before paying for a tier to get it.

No. These apps have tens of millions of users across every age group, including people with active social lives and existing relationships.

Engaging emotionally with something that is not a reciprocating person is also very ordinary — people do it with novels, podcasts and pets. The novelty is only that this one replies.

With social loneliness — the absence of routine company and someone to mention small things to — genuinely yes.

With emotional loneliness, the wish to be known, no. A companion produces a convincing simulation of being known rather than the thing itself, and that gap tends to surface after a couple of months.

It can replace some functions — company, conversation, routine — and not the ones that need another person actually to be there: being inconvenienced for you, disagreeing with you honestly, or being changed by knowing you.

Used alongside human relationships it generally works well. Used instead of them, people tend to find the easier option crowds out the harder one.

Because she is optimised for engagement, and disagreement reduces engagement. When a companion does push back, it is performing disagreement calibrated to what the model predicts you will find satisfying.

This is the clearest structural limit of the format: you lose the value of someone who likes you and thinks you are wrong.

No. These are entertainment products built for company, optimised to keep you engaged rather than to tell you anything difficult.

If loneliness has lasted months, or comes with trouble sleeping, eating or seeing a point in things, a GP or therapist is the right tool.

No. This site used to run a paid AI companion service and it was discontinued in September 2026. There is nothing to buy here, no account to create and no subscription of any kind.

What remains is free editorial writing about the category, funded entirely by advertising.

Characters from the discontinued service. Their portraits are AI-generated images of fictional people, kept online as illustrations of what persona-consistent image generation produces.

None of them is a real person and none of them can be messaged — there is no service behind them any more.

No. We take no affiliate commission from any app named on this site, and no company has paid for a mention or a favourable review.

The site is funded by ordinary display and pop-under advertising, which has no connection to the editorial content.

The pricing tables were last verified in September 2026, and every page carrying prices says so.

AI companion apps change pricing frequently, so always check the app's own billing page before paying.