AI glossary · Data and knowledge
Grounding
In one sentence
Grounding is the practice of making an AI model base its answers on specific, verifiable sources, such as your documents, a database or a search, rather than on what it remembers from training.
Also known as: AI grounding · grounded generation · source grounding · anchoring
What is grounding?
To ground something is to tie it to solid earth. Grounding an AI means making sure what it says rests on something you can check.
Put it to work
AI agents that look after your customers
From €49/mo + VAT
Voice, WhatsApp and email with the phone line included: agent, numbers and CRM in one fee.
See plans and pricingA language model without grounding answers from what it learned during training. That’s fine for explaining a general idea, but not for quoting your price or the status of an order. Without the facts, it can produce an answer that sounds convincing and is made up. Grounding gives it the right facts at the moment it answers and tells it to stick to them.
Related to this term
UTIO agents
- Available 24/7 in 26 languages
- Set it up yourself for €0: built from your website
- 14-day free trial, no card needed
From €49/mo + VAT
See plans and pricingHow it works
There are several ways to ground an answer, often used together:
- Documents. The relevant passages are found in your knowledge base and passed to the model. That’s RAG, with a vector database doing the searching.
- Live data. The agent checks a system: the calendar for free slots, the CRM for an order.
- Search. For general and current topics, public sources are consulted.
In every case, clear rules go into the system prompt: answer only from the information provided, say where it comes from and, if it isn’t there, say so.
Grounding doesn’t depend on the model alone. Above all, it depends on the quality of the sources. Messy or contradictory documentation produces answers grounded in bad data.
Example
A country guesthouse in the Lake District puts an assistant on its website. Without grounding, asked “do you allow dogs?”, the model answers with whatever is typical elsewhere. Sometimes it’s right, sometimes not.
With grounding, the assistant checks the guesthouse’s house rules: “Dogs up to 15 kg welcome in ground-floor rooms, with a supplement”. And for “have you got a room over the August bank holiday?” it assumes nothing: it checks real availability. If no source has the answer, it offers to take contact details so the guesthouse can call back.
What it means for your business
If an assistant talks to your customers on your behalf, its answers need to be grounded. Always ask where it gets its information and what it does when it can’t find it.
UTIO’s agents answer from your knowledge base, and in the support vertical it’s the only source: anything not in it doesn’t get answered, and a ticket is opened instead. For appointments, the agent checks the real free slots in your calendar before offering them. That’s the basis of a reliable AI agent.
Frequently asked questions
Are grounding and RAG the same thing?
Not quite. Grounding is the goal: answers that rest on sources. RAG is the most common technique for achieving it, by searching documents. You can also ground answers in a database, an API or a web search.
Does grounding eliminate hallucinations?
It cuts them sharply but doesn't eliminate them. The model can still misread an extract or mix two together. It helps if it cites its source, is told not to answer what it can't find and has a person review anything sensitive.
How can I tell whether an answer is well grounded?
You can point to the document or data it came from. If the system shows the source, check it. If an answer contains a detail that doesn't appear in any source, that detail isn't grounded.
Related terms
- RAG
RAG (retrieval-augmented generation) is a technique in which a language model first retrieves the relevant passages from your own documents and then writes its answer from them, instead of relying on what it memorised in training.
- AI hallucination
An AI hallucination is a response from an AI model that sounds convincing but is false or not backed by any source: a figure, quote, price or reference the model invents while generating the most probable text.
- Knowledge base
A knowledge base is a business's organised store of information (FAQs, terms, opening hours, product sheets, procedures) that an AI agent or a team draws on for answers.
- Vector database
A vector database stores embeddings, lists of numbers that represent the meaning of content, and finds the ones closest to a query very quickly. It's the search component in most RAG systems.
- Function calling
Function calling is a language model's ability to reply, instead of with text, with a structured request to use a specific function, giving its name and arguments. Your system runs that function and hands the result back to the model.
Updated on