Skip to main content

AI glossary Conversational

What is an intent in conversational AI?

← All glossary terms
In short

The goal behind what a user says or writes, such as booking a flight or requesting a refund, as detected by a conversational system.

Explained

In conversational AI, an intent is the user's goal behind a message: for example, check an order, book an appointment or request a refund. Intent recognition, also called intent classification, maps each utterance to the intent that best fits it, and that decision drives what the chatbot or voicebot does next: which flow to start, which data to ask for, which system to call or when to hand over to a person. Intents usually come with entities or parameters, the specific details such as a date, a product or an account number. Large language models make recognition more flexible, but a clear catalogue of intents is still what keeps a service predictable.

How intent recognition works

In classic platforms, designers define a set of intents and give each one training phrases, examples of how people express that goal. A model learns from them and, for each new utterance, picks the best-matching intent with a confidence score. Below a threshold, the system asks for clarification or falls back to a default response.

Alongside the intent, the system extracts entities: the details needed to act, such as a city, a date or a policy number. Context from earlier turns helps interpret short replies like "yes" or "the second one".

Intents in the age of LLMs

Large language models can understand requests they were never explicitly trained on, so rigid lists of training phrases matter less. They do not remove the need for intents, though.

  • Intents still define what the service is allowed to do and which process or system each request goes to.
  • They give a stable set of categories for measuring volumes, containment and satisfaction.
  • An LLM can classify intent and extract details, while a defined flow handles the sensitive steps, such as payments or identity checks.

Designing a good intent catalogue

  • Start from real conversations and transcripts, not from the organisation chart.
  • Keep intents distinct: overlapping intents cause misrouting.
  • Decide what happens when the system is unsure, including a clear path to a human agent.
  • Review unmatched and misclassified utterances regularly and adjust.

Frequently asked questions

What is an example of an intent?

"I want to change my delivery address" and "Can you send it somewhere else?" are different phrasings of the same intent, say, change delivery address. The system recognises the intent and then asks for, or extracts, the new address as an entity.

What is the difference between an intent and an entity?

The intent is what the user wants to do; entities are the details needed to do it. In "book a table for four on Friday", the intent is a booking and the entities are the number of people and the day.

What is intent classification?

It is the step that assigns a user message to one of a set of predefined intents. Conversational platforms use machine learning models for this, and LLMs can now do it from a list of intent descriptions with few or no training examples.

Do LLM chatbots still need intents?

Usually yes, though in a lighter form. The model understands free language, but the business still needs to decide which requests it handles, how each one is resolved and when to escalate. Intents are a practical way to organise that and to measure it.

How is intent detection used in contact centres?

It routes calls and chats to the right flow or team, lets virtual agents resolve simple requests on their own and gives human agents context before they pick up. Aggregated intent data also shows why customers get in touch, which helps prioritise improvements.

Decision guides that use this term

How to put it to work

Where it shows up at Thinkia

Further reading

Sources

Updated: 2026-10-21

Want to see how this applies to your organisation?

Talk to an AI Expert