What is grounding in AI?

Grounding and citations, explained

A grounded answer is one where every factual claim can be traced to a source the business approved: a help article, a policy page, a product listing or a data lookup. Grounding is usually achieved with retrieval-augmented generation, where relevant passages are fetched and the model is told to answer from them.

A citation is the evidence trail. It might be a link to the article, a label naming the source, or a quoted passage. Citations serve two audiences. Customers can open the source if they want the detail. Reviewers can check, quickly, whether the claim in the answer actually appears in the source.

The two are related but not the same. A system can be grounded without showing citations, and a citation can be attached to an answer that misstates what the source says. That second case matters: a link gives an answer an air of authority, so evaluate whether the cited source supports the specific claim, not just whether a link exists.

When no approved source covers a question, a grounded system should say so and offer another route, rather than answering anyway.

How to measure it

Check the claim against the source, not just the presence of a link.

Common mistake

How SuperBot relates

SuperBot trains on your own content and answers with citations: replies show a source label when matched knowledge supports the answer. Citations are not guaranteed on every reply, so review answers in your dashboard.

• AI chatbot evaluation checklist
• How SuperBot works

Related terms

• Retrieval-augmented generation (RAG)Retrieval-augmented generation (RAG) is a technique in which an AI system first searches a trusted knowledge source for passages relevant to a question, then gives those passages to a language model to write the answer. In customer support, RAG keeps replies tied to the business's current policies and help content instead of the model's general training data.
• Hallucination (AI)An AI hallucination is a response from a language model that sounds confident and fluent but is false, unsupported by its sources, or invented outright. In customer support, typical hallucinations are made-up policies, prices, delivery dates or product features, and they are dangerous because customers act on them and businesses may have to honour them.
• Knowledge baseA knowledge base is an organised library of help content, such as how-to articles, policies, troubleshooting steps and FAQs, that customers and support staff use to find answers. For AI support agents, the knowledge base is also the source material the agent retrieves from, so its accuracy and freshness directly limit the quality of automated answers.
• AI customer support agentAn AI customer support agent is software that uses a large language model to answer customer questions in natural language, grounded in a business's own content, and can take bounded actions such as looking up an order, capturing a lead or handing the conversation to a human when it cannot help.