All termsGlossary

What is grounding?

Grounding means tying an AI agent’s answer to source material it has been given, such as help articles or account records, so that the answer reflects those sources and not the model’s general guesses.

Why it matters

A language model on its own writes fluent text whether or not it is correct. In customer service, a wrong refund period or an invented opening time causes real problems. Grounding limits answers to what approved sources say, which reduces made-up details and makes replies checkable. It also lets the agent say it does not know when the sources are silent.

How it works

The system places relevant source text in front of the model along with the customer’s question, and instructs it to answer only from that text. Sources can be documents, records fetched from other systems or details from earlier in the conversation. Many set-ups also keep a note of which source each statement came from, so a reviewer can check the answer against it.

In practice

A traveller asks an airline’s chat agent whether a cabin bag can weigh ten kilos. The agent is given the current baggage policy and answers from it, quoting the limit for the traveller’s fare type. When the traveller asks about carrying a drone battery, the policy supplied does not cover it. The agent says so and passes the chat to a member of staff.

Get Started

Let’s build AI agents, together.

Talk with us to see the top three use cases for AI agents at your business, and how quickly you could go live.

Live in 7 days · Works inside the tools your team already uses · Hands off with full context