Three words get used loosely for what are, functionally, three different products. AI companion, chatbot, assistant. Each is a real category with real conventions, and the differences matter — mostly because they determine what you should expect the app to be good at, and what you should not.
One-sentence definitions
- A chatbot is a text interface layered on a task or knowledge base. Ask it a question, get an answer, close the tab.
- An assistant is a chatbot with tools and memory of your account context — it can do things, not just answer things.
- An AI companion is a chatbot with a fixed persona and long-term memory of you — the value is the relationship, not the answer.
The three axes that separate them
The words are fuzzy, but three product decisions cleanly separate the categories in practice.
1. Persona
A chatbot has no persona. It should sound helpful, professional, and generic — like your bank's help page rendered as text. An assistant has a light, consistent voice but no character (think: "as your assistant, I noticed…"). An AI companion has a name, a personality, a taste, opinions, and a voice that stays the same across sessions. The persona is the product.
2. Memory
A chatbot has session memory only — it forgets you the moment you close the tab. An assistant has account-context memory — your calendar, your files, your recent searches — but not usually a personal history. An AI companion has both: it knows you moved last month, it remembers your dog's name, it recalls a conversation you had in June.
3. Success metric
A chatbot succeeds when it answers your question correctly and you leave. An assistant succeeds when it completes a task you would otherwise have done yourself. An AI companion succeeds when you come back — the goal is the ongoing conversation, not any single answer. This has real implications for how each category is designed.
Concrete examples of each
Categorising real products against the three axes above:
- Chatbots: your bank's support chat, most e-commerce site "help" widgets, a legal Q&A tool built on a document corpus.
- Assistants: ChatGPT (in its default mode), Google's Gemini in a phone, Apple's Intelligence features in iOS, Copilot in Microsoft 365.
- AI companions: Character.AI's consumer product, Replika, Nomi, and every romantic/roleplay-first product on our reviews list.
When each category is the right choice
Use a chatbot when
You have a single question you need answered, ideally about a topic where the chatbot has been trained on the specific corpus (your bank's policies, a legal database, a product's docs). Chatbots are best when the answer needs to be exactly right and you do not want a personality getting in the way.
Use an assistant when
You want to save time on a series of small tasks — draft this email, summarise this PDF, book this meeting, extract data from this spreadsheet. Assistants shine where the value is in the doing, not the conversation.
Use an AI companion when
The conversation itself is the value. Roleplay, language practice, ongoing creative writing, low-stakes social conversation, or (in the adult segment) romantic personas. If you find yourself wishing an assistant "remembered" you and had more personality, you actually want a companion.
Why the categories are collapsing
The three categories are converging because the underlying tech is the same. In 2026 every product uses the same class of large language model; the differences are in the wrapper — memory, persona, tools, UI. It is now trivial for a chatbot to sprout a personality, or for a companion to gain the ability to draft emails. The lines that used to be firm are dissolving into a spectrum.
This is good news for buyers, because you no longer have to keep three different products. It is bad news for reviewers, because "which category does this belong to?" is now a marketing question rather than a technical one. Our approach: rank each product by the job it does best, not the label it wears.
A short buyer's heuristic
When a new AI product launches, three questions almost always tell you what it really is:
- Does the character on the other side have a name and a personality that stays the same? If yes → companion territory.
- Can it do things (schedule meetings, edit files, run code)? If yes → assistant territory.
- Is it optimised to answer specific questions from a single knowledge base? If yes → chatbot territory.
Most products score high on one and light on the others. If a product scores high on all three, it is either a very ambitious platform or (more often) a marketing page that will disappoint in at least two of the three dimensions.