Citations and Grounded Answers
Make RAG answers cite the right sources and check that every claim is actually supported by them.
A taste of a lesson
Our bot cites passage 3 for a refund deadline, but passage 3 never mentions a deadline. How do I catch this?
Add verification after generation. First, a cheap check: any number or date in the answer, such as 30 days, should appear in the passage it cites; if not, flag it. Second, a support check: for each cited sentence, ask a separate model call whether the passage supports the claim, answering yes, no or partial. Flagged answers can be regenerated with the feedback or sent for review. Then measure how often it happens. How many answers in your test set fail the number check?
Written by the teacher as an example. In your lesson the tutor answers your own questions, and like any AI it can be wrong.
What you will be able to do
- Format passages and instructions for claim level citations
- Explain the common ways citations look right but are wrong
- Verify cited ids, quotes, numbers and claim support automatically
- Design an interface that makes sources easy to check
- Measure citation precision and recall in your evaluations
Lesson plan
- 1 What grounded means Define grounded answers and why citations matter for trust. Start
- 2 Prompting for citations Format passages and instructions to get claim level citations. Start
- 3 How citations go wrong Recognise citation failures in real outputs. Start
- 4 Automatic verification Check citations and claim support in code and with a checking step. Start
- 5 Sources in the interface Present sources so users can verify answers quickly. Start
- 6 Measuring citation quality Add citation metrics to your evaluation set. Start
Try asking
About this tutor
For developers with a working retrieval system who need answers people can trust and verify. You learn what grounded means in practice, how to format passages and instructions so the model cites sources at the claim level, how provider citation features differ, and how to verify citations automatically: does the cited passage exist, does the quoted text appear, does the passage support the claim. You also design the interface for sources, handle conflicting documents and the not found case, and measure citation quality in your evaluations rather than assuming that a citation means a correct answer.
Reviews
4.5
2 ratingsSample
- Hiroshi K.Sample
Good on citation precision versus recall. The conflicting documents lesson helped us write a clear policy for old and new versions.
- Paolo F.Sample
The simple number check caught a surprising share of wrong citations. The support check handles the rest. Our legal team finally trusts the sources panel.
About the teacher
Takes retrieval systems from demo to dependable: parsing, citations, freshness, retrieval evaluation and debugging
9 tutors 267 lessons taught Sample
Most RAG demos work on the ten documents someone picked. I teach what happens after that: scanned PDFs, tables, documents that change every week, answers that cite the wrong page and users who ask things the documents never covered. My background is in document processing and internal knowledge tools, so I am practical about formats and sceptical of any setup...
See Fumiko's profile and tutorsMore like this
Other tutors on the same or nearby topics.