What it means
Artificial intelligence is the umbrella, not a technique. It covers everything from the rule-based expert systems of the 1980s to today's language models, united only by the goal: get a machine to do something that would count as intelligent if a person did it.
That breadth is why the word is nearly useless as a product description. In practice, essentially all commercially relevant AI today is machine learning, and most of what people mean when they say "AI" in conversation is a large language model with an interface attached.
A useful piece of history: the definition keeps moving. Optical character recognition, chess engines, and spam filters were all landmark AI achievements, and none of them feel like AI now. This is common enough to have a name — the AI effect — and it is worth remembering when judging what today's systems will feel like in a decade.
Why it matters
Because the term covers so much, "we use AI" tells a buyer almost nothing. The questions that carry information are narrower: what does the system learn from, what decision does it make, and what happens when it is wrong. Regulation runs into the same problem — a law written for "AI" has to define the term precisely enough to be enforceable without freezing it to one generation of technology.
In practice
Use the specific word when you have one. "A language model that drafts responses" is a claim someone can evaluate; "AI-powered" is not.