A citation is a link inside an AI reply that points back to the source the AI used. If the agent says "our standard return window is thirty days", a citation links to the page in your knowledge base where that policy lives. It is the difference between "trust me" and "here, check".
Citations are how AI replies stop being a black box. The customer can verify the answer. Your team can audit it. If a policy changes, you can trace which conversations cited the old version.
Why citations matter
Generative AI is fluent. Fluency is persuasive. Persuasive is dangerous if the underlying answer is wrong. A citation reverses the burden: the AI does not ask you to believe it, it shows you where it got the answer.
In support specifically, citations turn AI from a confident stranger into a guide. The customer sees the AI is reading from your real help center, not making things up. Their trust in the answer goes up. Their patience for being on hold goes down. And if the answer is wrong, the citation makes the mistake easy to find.
What a good citation looks like
A good citation links directly to the passage used, not the homepage of your help center. It uses the source's real title, not "doc-7234". If multiple sources were combined, all of them are cited.
Equally important: the citation is honest. If the AI did not use any specific source, the answer should say so plainly rather than fabricating a link. A citation pointing to a page that does not contain the cited fact is worse than no citation at all.
When citations are not enough
Citations help with verifiability, not correctness. The AI can cite a source it misread. It can cite a source that is itself outdated. Citations need to be paired with grounding (which forces the AI to stay close to retrieved content), source quality (the documents being indexed are correct), and human review on high-stakes replies.
In Keloa
In Keloa, every AI agent reply is generated with RAG and ships with the sources it used. Citations link to the section in your knowledge base, so the customer can read further and your team can verify in seconds. See how the AI works for the full pipeline.