What Are Intents, Entities, and Context in Chatbot Design?

Before large language models could hold a fluent conversation, chatbot builders relied on three foundational concepts to make bots understand people: intents, entities, and context. These ideas powered platforms like Dialogflow and Rasa, and they did not vanish when generative AI arrived. They simply moved up a layer. If you design assistants, automate support, or build AI agents today, understanding this trio explains why bots succeed or fail at the moment a real person types something messy and unexpected.

What an intent is

An intent is the goal behind a message — what the user is actually trying to accomplish in a single conversational turn. "Where's my order?", "track my package", and "did my stuff ship yet?" are three different sentences that map to one intent: checking order status. In classic platforms, you train an intent by giving it example phrases, and the system learns to classify new, unseen messages into the closest match. This is intent classification, and it is the first decision any bot makes. Get it wrong, and everything downstream — the response, the action, the data lookup — goes wrong too. The practical lesson for builders is that users rarely phrase things the way you expect, so an intent is only as good as the variety of real language behind it.

What an entity is

If the intent is the verb, an entity is the noun. Entities are the structured pieces of information extracted from a message: a date, a city, an order number, a product name, a quantity. In "book a table for four on Friday," the intent is make a reservation, while four and Friday are entities the bot must capture to act. Platforms distinguish entity types (the category, like a date) from entity values (the specific data pulled from the text). Some entities are system-provided (numbers, dates, currencies); others are custom to your domain, like internal product SKUs. Reliable entity extraction is what turns a vague understanding into something a system can actually execute against a database or API.

What context is

Context is conversational memory — the awareness that a turn does not happen in isolation. When someone says "they're sold out," the bot needs to know what they refers to from earlier in the chat. Context lets the system carry information forward, disambiguate references, and decide which intents are even valid right now. Classic platforms modeled this explicitly with context objects, slots, or dialogue state that activate or expire as the conversation moves. Without context, a bot treats every message as the first one, which is why early bots felt frustratingly forgetful. With it, a bot can ask a clarifying question, remember the answer, and continue a multi-step task like a booking or a return.

How the three work together

In a working assistant, the three concepts form a pipeline. A message arrives, the bot classifies its intent, extracts the relevant entities, and interprets both through the current context. Suppose a user says "change it to Saturday" after booking a table. The intent is modify reservation, Saturday is an entity, and context supplies the missing detail — which reservation "it" means. Remove any one piece and the interaction breaks: no intent means no goal, no entities means no data, no context means no continuity. This separation of concerns is also what makes bots testable and debuggable. You can inspect exactly where understanding failed rather than guessing.

Why this still matters in the age of LLMs

Large language models blur these boundaries because they infer goals, extract data, and track context implicitly inside one model. That feels like progress, and it often is — but the concepts remain the right mental model for design and evaluation. When you build an AI agent, you still define which actions it can take (intents), which parameters those actions need (entities), and how it should carry state across turns (context). Tool-calling and function schemas are entities and intents wearing new clothes. Teams that ignore this find their LLM agents hallucinate parameters, lose track of the conversation, or trigger the wrong action. The vocabulary gives you a precise way to specify, test, and fix behavior that a vague prompt cannot.

Bottom line

Intents, entities, and context are not legacy baggage — they are the durable grammar of conversational AI. Whether you are configuring a no-code bot or orchestrating an LLM agent with tools, you are still answering the same three questions: what does the user want, what details do they need, and what has already happened? Naming those questions explicitly makes your assistant more predictable, easier to debug, and far more useful to the person on the other end.

Sources and further reading