AI Chatbot, Assistant, Companion: What the Words Actually Mean
“AI chatbot companion” is a phrase people search before they know the vocabulary of the category, and the vocabulary is genuinely unsettled. Chatbot, assistant, companion, character, agent and friend are used interchangeably in store listings while describing four or five different kinds of software, and the differences matter for what you should expect and what you should check. Decoding the words is the cheapest thing you can do before installing anything.
This post is about reading language rather than about any product. It exists because the words are doing work in store listings that most readers assume they are not.
The words, in order of how much they promise
Each of these has a defensible technical meaning and a much looser marketing use.
Chatbot is the oldest and the widest. Historically it meant any program you exchange text with, including entirely rule-based ones with no model at all. In current listings it usually does mean a language model, but it is the word most compatible with something very simple underneath, and it makes no promise of continuity or character.
Assistant implies tasks. The interaction is meant to end when the thing is done, and the product’s own measure of success is completion. Assistants can be prompted into companion-like behaviour, which is not the same as being built for it.
Companion is the one that carries an actual product commitment: a consistent character that resumes across sessions. When a listing chooses this word over “chatbot”, that is usually a signal about intent rather than noise.
Character signals that the personality is a defined, often user-authored object — something someone wrote, that other people can also talk to. On platforms built around this, who wrote the character is a real question.
Agent in a consumer listing generally means the software is expected to take actions elsewhere on your behalf. That is a much larger permissions question than conversation, and it deserves attention when it appears.
Friend, girlfriend, boyfriend, soulmate describe a relationship framing rather than a capability. They tell you what the product is positioned as and nothing whatever about what it does.
The two-word combinations to be careful with
Some pairings are informative and some are marketing sediment.
“AI chatbot companion” — the phrase in the search box — is three words claiming the same territory at once, and stacked terminology like this is more common in listings assembled for search visibility than in listings written by the people who built the thing. Worth noticing, not conclusive.
“AI-powered” attached to a feature usually means the feature involves a model somewhere. It is not a description of quality, scale, or whose model it is.
“Advanced AI” and “next-generation model” are unfalsifiable by construction. Neither is a lie and neither can be checked.
“Private” and “secure” are the pair that most needs unpacking. Both are used to mean anything from end-to-end encryption to “we have a privacy policy”. The only place either word means something specific is the operator’s own policy document.
“On-device” and “offline” are the rare technical claims in this vocabulary that you can actually test yourself, which is why they are worth taking literally. The test is in what on-device processing would look like.
Where the words come from
Understanding why listings read as they do makes them easier to discount.
Store listings are search surfaces. Titles, subtitles and description text feed app-store search, so listings accumulate every plausible synonym a searcher might type. That produces the stacked-terminology effect above and it is an ordinary, non-sinister incentive.
The category is young and has no agreed taxonomy. There is no standards body for what counts as a companion versus an assistant, so nobody is misusing a term that was ever pinned down.
Copying is rife. A listing that succeeded gets imitated, phrasing and all, which is one of several reasons a copy can read exactly like an original. The store fields that do distinguish them are in spotting a lookalike companion app.
Reading a listing usefully
What to look at once you have discounted the vocabulary.
Nouns over adjectives. “Voice replies”, “conversation export”, “character editor” are checkable features. “Deeply personal”, “truly understands”, “remarkably human” are not.
Anything expressed as a limit. Numbers of anything, caps, tiers, quotas. Not because the specific figures matter — they change constantly and no page should be trusted on them — but because an operator willing to write limits down is telling you more than one that will not.
Who published it. The developer field, the other apps under the same name, and whether the linked policy sits on a domain that matches. A minute of work and the highest-value minute available.
What the app asks for. Store listings enumerate declared permissions and data categories. Compare those against the features described; a mismatch is meaningful. What each request actually grants is covered in what a permission prompt actually grants.
What this does not tell you
Vocabulary is not evidence. An app that calls itself a chatbot may be considerably better than one calling itself an advanced AI soulmate, and the words tell you what a listing wants you to expect rather than what the software does. Nothing here separates a good product from a bad one.
It also does not settle what any of these apps are for. Deciding whether you want one is a different question from deciding what one is, and the second is prerequisite to the first only in the narrow sense that knowing the category’s boundaries — set out in what an AI companion app actually is — keeps you from evaluating a product against expectations it never claimed to meet. The commonest disappointment in this category is not deception; it is a reader who took “companion” literally and a listing that never meant it that way.