AI terms every business owner hears, explained without the jargon

When a vendor hides behind jargon, something is being hidden: the words tell you what you are buying and what can break.

The vocabulary arrives in the meeting before the product does: model, chatbot, assistant, training, hallucination, grounding. Most owners nod along, sign, and learn the words later from a bill or a failure. This is not decoration; each term marks a real difference in what gets built, what it can do for you, and what it costs when it breaks. Learning them costs an evening. Not learning them costs a year.

A model is the engine: the part that produces an answer when asked. A chatbot is any program that converses, and the word promises nothing about intelligence, because a phone menu inside a chat window is a chatbot, and so is a system that reads your documents and replies in two languages. When a vendor says chatbot, you have learned the shape of the thing and nothing about its brain.

An assistant is a chatbot with access. It reads your actual business information and can act on your behalf: booking, updating records, drafting the reply that goes out under your name. That access is the value and the risk in the same wiring. The more it can do, the more it matters what it sees, what it is allowed to do, and who answers for what it does.

Training is the long, expensive process that shapes the engine itself, and it is not what happens when a system reads your documents. That reading is grounding, a discipline built on top of the model, and it is the difference between a system that answers from your material and one that improvises. The confident wrong answer has a name too: hallucination. It is a property of these machines, not a bug a patch removes, and grounding is how honest builders reduce it, never a promise that it cannot happen.

The vocabulary is commercial before it is academic. A vendor who hides behind jargon is hiding something, and what is hidden is usually the distance between a demo and a deployed system. The words tell you what you are actually buying: a script wearing a chat window, or an engine with your business behind it. They also tell you what can break, and breakage is where these projects quietly die.

Whatever the terms, run the ownership test before signing. Ask which parts of the system see your customer data, which parts act on your behalf, and who answers when it is wrong. A serious builder answers all three in plain sentences you could repeat to a client. A hesitant answer has told you something the proposal was careful not to.

We build assistants for a living, including our own, and we describe them the way this essay defines the words: transparent AI that says plainly that it is AI, grounded in the client's own material, with a human handoff for what it should not hold. That is guidance, not a guarantee, and it applies to us too: ask any builder these questions, weigh the plainness of the answer, and let the vocabulary do its job.