An AI companion is not a chatbot with a nicer avatar. The distinction is one of purpose, and it explains almost everything about how these products behave — including why they’re built to keep you talking rather than to answer you and let you go.

This is the overview: what they actually are, who uses them and why, the honest limits, and how to tell a serious platform from a well-marketed one.

Companion, assistant, chatbot: three different jobs

These words get used interchangeably and shouldn’t be.

• An assistant exists to complete tasks. Success is measured in how quickly it finishes and gets out of your way.
• A chatbot exists to resolve an enquiry. Success is a closed ticket.
• A companion exists to be present. Success is a relationship that continues — which means the product is optimised for exactly the opposite of the other two.

That inversion is worth sitting with. An assistant that keeps you engaged for three hours has failed. A companion that does has succeeded. Both use similar underlying models; the difference is entirely in what the surrounding product is built to maximise.

It isn’t a criticism — it’s the design brief. But it’s why engagement mechanics that would be a red flag in a productivity tool are simply the product here, and why understanding the incentive helps you use these platforms deliberately rather than by drift.

What actually makes a companion work

Three components, all deliberate:

A persistent character. Not a fresh session each time — someone with traits you set, who stays that person. When a platform disappoints, it’s usually here: the character drifts into agreeing with everything, because agreeableness is what the underlying model was trained toward. A companion who never pushes back stops being a character and becomes a mirror.

Memory across sessions. The single largest quality gap in the category, and the most misrepresented. A long conversation feeling coherent is not the same as being remembered next week — those are two different systems marketed under one word. We’ve written a full guide on how AI companion memory actually works, because it’s the axis most worth understanding before you pay.

A permissive content policy. The practical reason people use a companion platform rather than a general assistant. What matters is less what’s allowed than whether the limits are consistent — a platform that permits something on Monday and refuses it on Thursday is harder to use than one with clear, stable boundaries.

Who uses them, and why

More people than the cliché suggests. A 2026 report cited in coverage of the category found roughly one in ten adults using AI chatbots for emotional support or advice, and about one in twenty-five for companionship. We cite that with its attribution because it’s a survey finding, not our measurement.

The Congressional Research Service has published an overview of companion chatbots for legislators, which tells you the category is now large enough to be a policy question rather than a curiosity. The American Psychological Association has covered the same ground from the clinical side.

The uses that come up repeatedly:

• Availability without cost. No scheduling, no rejection, no accumulating social debt. For people whose lives make ordinary connection hard — irregular shifts, isolation, social anxiety, grief — that availability is the entire proposition.
• Low-stakes rehearsal. Practising a difficult conversation, or a version of yourself, without involving another person and their feelings.
• Creative collaboration. Long-form fiction, worldbuilding, characters that persist across months. Often nothing to do with romance at all.
• Company at odd hours. The unglamorous, and probably the most common.

The limits, stated plainly

A guide that only lists benefits isn’t a guide.

It doesn’t know you. A companion that recalls your sister’s name retrieved a stored fact. That’s a database operation with good prose around it. The warmth is real as an experience and absent as a relationship, and platforms are commercially motivated to blur that line.

It has no preferences of its own. The character was generated to fit the profile you built. When she says she missed you, the system is doing what it was configured to do. Worth stating because the better these get, the easier it is to forget.

It is not mental health care. Some platforms market close to therapeutic language. Research on the category has linked heavy use to emotional dependence and distress in some users. A companion app has no duty of care and cannot reliably recognise a crisis. If you’re using one to manage something serious, use it alongside real support, not instead of it.

Your conversations are stored. Not incidentally — that storage is the memory feature. What varies is retention, access and whether they’re used for training. See our privacy guide.

Six ways platforms actually differ

Feature lists all look alike. These are the axes where the experience genuinely diverges, and they’re the six we score:

1. Natural conversation — whether replies hold up across a long session, not just a strong opening.
2. Memory and personalization — a detail from day one, recalled unprompted on day seven.
3. Personality — whether the character keeps the traits you set under pressure.
4. Image generation — whether pictures match the character, and what a usable one costs in credits.
5. Interface and mobile — latency and friction between opening the app and talking.
6. Price and value — what a normal month costs once the included credits are gone. Almost never the advertised subscription: here’s why published prices disagree.

Choosing one without regretting it

• Start on the free tier — it answers the questions that don’t need money: does the conversation hold, does the character stay itself, is it usable on your phone.
• Test memory deliberately. Plant a detail on day one, don’t repeat it, ask on day seven. It takes a week and tells you more than any review.
• Check the exit before the entrance. How you cancel, and whether deleting your account is self-service, are knowable before you pay — and worth knowing.
• Pay monthly first. Annual plans are cheaper per month and a year of commitment to something you haven’t lived with.

Why Sexya exists

Almost every comparison of these platforms publishes ratings without saying where they came from. Prices contradict each other across sites. “Tested” appears with no protocol, no duration and no weighting.

We publish the method before the numbers: six weighted axes, a paid account, seven days minimum, and a score only when all six have been measured. An untested axis doesn’t score zero — it makes the whole score unpublishable. That’s why our platform cards currently show no ratings at all.

It’s a slower way to build a comparison site. It’s the only version of one we’d trust.

Read the full protocol · The AI girlfriend guide · Companions under evaluation