Most "AI in hospitality" content treats large language models as either magic or vaporware. Both framings are wrong and both stop you from making good decisions about what to actually deploy in your property. The honest explanation takes five minutes and gives you the conceptual model to evaluate every AI vendor pitch you will receive for the next three years.
What a large language model actually is
A large language model (LLM) is a statistical pattern-matcher trained on a very large corpus of text. Given a sequence of words, it predicts the next word — billions of times in a row, very fast. When you "ask it" something, you are giving it a sequence (your prompt) and asking it to predict what comes next (the response). It is not searching a database. It is not reasoning the way a human reasons. It is producing a statistically likely continuation of your input.
That sounds reductive but it is structurally accurate. The model — Claude, GPT-4, Gemini, Llama — has no memory between conversations, no awareness of facts after its training cutoff, no understanding in the human sense. What it has is an enormous pattern library that often produces correct, useful, and sometimes brilliant output when prompted well.
Why this matters for hotel decisions
An LLM is excellent at: drafting an email reply, summarizing a long review, translating a guest message, generating ten variations of marketing copy. It is bad at: arithmetic with specific numbers, looking up a guest record (unless integrated with your PMS), guaranteeing a fact is true (it will confidently make up facts — "hallucinate"), or doing anything it has not been shown patterns of.
When a vendor tells you their AI "understands your guests," ask what specifically the AI has been trained on, what data it has live access to, and what it does when it does not know. A vendor who cannot answer those three questions is selling marketing, not technology.
The three things to remember
You don't need to understand the math. You need to understand what the tool can and cannot do — so you can recognize hype when you hear it.