Intent Recognition

Last updated: October 6, 2026

What intent recognition is

An intent is the goal behind a message. “How much is the Pro plan?”, “What does Pro cost per month?” and “pricing for pro??” use different words but share one intent: a pricing question. Intent recognition maps these varied phrasings to the goal behind them.

It is often paired with entity extraction, which picks out details such as a product name, a date or an order number.

How it works

There are two common approaches:

Classic intent classification: a team defines a fixed list of intents, writes example phrases for each and trains a classifier. Each intent links to a scripted answer or flow. This is predictable, but it needs ongoing maintenance, and messages that match no intent end in a fallback reply.
Language model reasoning: a large language model reads the message in context and infers the goal without a predefined list. It copes with new phrasings and several intents in one message, and its behavior is steered through instructions rather than a fixed map.

Example

A visitor on a software company's site writes: “We're 12 people, is there a team discount and can I try it first?” The message holds two intents, a pricing question and a trial question, plus one entity: a team size of 12. A classic bot may only catch one intent. A language model can answer both parts and use the team size to point to the right plan.

Why it matters for business chatbots

Reading intent correctly is what turns a chat into a useful conversation. It decides whether the bot answers from the knowledge base, offers a contact form, calls a tool or admits it cannot help. Intent is also a sales signal: questions about pricing or availability often mean a visitor is close to a decision.

Intent recognition in intoCHAT

intoCHAT does not ask you to build an intent list. The language model reads each message in context and decides how to respond, guided by your instructions. For lead capture, the AI trigger lets the agent decide when to show the lead form, based on your description of a good lead. Alternatively, rules show the form on keywords or after a set number of messages. The agent also decides when to call one of your custom API actions based on what the visitor asks.

Frequently asked questions

What is the difference between an intent and an entity?

The intent is what the user wants to do, such as checking opening hours. An entity is a specific detail in the message, such as a location, a date or a product name. Chatbots often need both to give a precise answer.

Do AI chatbots still need predefined intents?

Not necessarily. Chatbots built on large language models can infer intent from free text without a fixed list. Some platforms still use predefined intents for flows that must behave the same way every time.

What happens when a chatbot misreads intent?

It gives an off-topic answer or no useful answer at all. Clear instructions, good source content and a fallback, such as a way to leave contact details, limit the damage.