RAG & Retrieval · Ground answers in evidence
How do you generate reliable citations?
Short answer
Have the model cite stable passage IDs, then map those IDs to human-readable sources in code. Validate the mapping and show enough context for the user to verify the claim.
Asking a model to produce a URL or page number invites invention. A safer pattern is to give every packed passage a short ID, have the model cite the ID, and let the application render the real title, URL and page. The model never needs to know the final display format.
- Assign IDs at packing time and keep an ID-to-source map.
- Validate that each cited ID was actually in the prompt.
- Render citations next to the sentence they support, not only at the end.
- Link to the exact page or anchor when the source supports it.
Follow-ups you should expect
Interviewers use these to check whether the first answer was understanding or memorisation.
- 1How would you handle an answer that cites the same passage for every claim?
- 2What would you show when the source is behind a login?
- 3How do you evaluate citation quality at scale?
Say it in the interview
I do not let the model construct URLs. Each packed passage gets a short ID, the model cites the ID, and my code maps that to the real title, link and page. I validate that every cited ID was actually supplied. Then I render the citation next to the sentence it supports, so the user can check the claim directly. Existence of a citation is not enough; it also has to support what was said.
Use this as a model, not a script. Saying it in your own words is what makes it survive follow-up pressure.
Related questions
Reading is step one
Now say it out loud.
Understanding an answer and delivering it under pressure are different skills. Rehearse this topic in a live RAG & Retrieval mock interview and get scored feedback on depth, clarity and confidence.