Ahmadreza SamadiThe confident answer from the old price list
In a fictional hotel, Leila asks an AI assistant for a meeting-room quote. A fluent answer appears. Before sending it, she spots the problem: its supporting sheet is an old version. This is a teaching scenario, not a report of an actual hotel or product test.
English film with synthetic narration and English captions. Leila, the hotel, the helper and the documents are fictional. Original animation, not footage of a real hotel, a product test or a real price quotation.
The situation
In a fictional hotel, Leila asks an AI assistant for a meeting-room quote. A fluent answer appears. Before sending it, she spots the problem: its supporting sheet is an old version. This is a teaching scenario, not a report of an actual hotel or product test.
How retrieval helps
Retrieval-augmented generation, or RAG, supplies external material to a model before it drafts an answer. Microsoft describes retrieval, added context and generation. Anthropic explains how isolated document chunks can lose context. Retrieval can help; it does not guarantee a correct answer. [MS] [ANT]
Record, date, answer
Our practical check is record → date → answer. Record: can you open the exact supporting passage? Date: is this the currently approved version, rather than merely a recently retrieved file? Answer: does that passage actually support this quote and its conditions?
When the evidence is missing
For the fictional hotel, a useful response names the approved sheet and relevant conditions. If that record is missing, pause the quote and say what is missing. Do not turn a search result into invented certainty.
Limits
This framework is an editorial decision aid, not a benchmark or a security guarantee. A citation is a route to evidence, not proof of every generated sentence. This machine-assisted article has source checks; it does not claim human fact-checking.
About the author
Ahmadreza Samadi — https://a-samadi.com