Natural language understanding (NLU)
Natural language understanding is the part of AI that works out what a piece of human language means: its intent, the things it names and its tone.
Natural language understanding, or NLU, is the part of AI that works out what a piece of text means. It goes past matching words. It tries to tell what the person wants, which things they're talking about, how those things relate, and whether they're pleased or annoyed.
It's why search engines stopped needing exact keywords, and it's the core of how AI platforms read both the questions people ask and the pages they use to answer.
What it does
Take the search "is Fernhill any good for decaf." An NLU system works out that "Fernhill" is probably Fernhill Coffee, a brand, that "decaf" refers to its decaffeinated coffee, and that the person wants an opinion before buying. None of this appears in the words alone.
The same skills apply to pages. A system reading a review can tell that "the beans arrived stale and the box was crushed" is a negative statement about freshness and delivery, even though neither "negative" nor "freshness" appears in it.
NLU in search
Google has been open about using language models to understand searches. In 2019 it explained how BERT helps it understand the small words that change a search's meaning, like "for" and "to." That shift is why pages written in plain, natural language started doing better than pages stuffed with keywords.
NLU and AI search
Large language models are, among other things, very capable NLU systems. When someone asks an AI platform a long, conversational question, the platform has to understand it well enough to decide what to search for, which pages are relevant and which brands fit. Then it reads those pages the same way.
For a marketer, that has a practical effect. The platform doesn't need you to repeat a keyword. It needs your pages to say clearly what you are, what you sell and who it's for, so a system reading them reaches the right conclusion.
In AnswerOrbit
We do a version of this on the answers. We work out which brands each answer names, the sentiment of each mention with the words it's based on, and which claims about you go against your facts. See How we count.