Why cited answers beat a black-box chatbot

The fastest way to lose a team’s trust is to hand them a confident number they cannot check. Most assistants that sit on top of your data are black boxes: you ask a question, you get a sentence, and you have no idea whether it came from your tables or from thin air. The moment one figure turns out to be invented, every other figure is suspect too.

Query, not vibes

When you ask iDBQuery a question, the AI does not free-associate an answer. It writes a query against your connected sources, runs it, and answers only from the result set. We surface the underlying query so anyone can read exactly what was asked, edit it, and re-run it. The answer is a consequence of real rows, not a guess dressed up as one.

That makes the whole system inspectable. If a figure looks off, you open the query, see the join and the filter, and find the bad row — instead of arguing with something that cannot explain itself.

Citations as a contract

Every number carries the source it came from: the table and row for structured data, the file and exact page for a document. Conversations persist with an audit log, so a finding from last quarter can still be opened and traced today. The rule is simple — nothing is asserted that cannot be traced back to its evidence.

This is why a cited answer is not a nicer version of a chatbot. It is a different kind of object: a result you can put in front of an auditor, a board or a regulator and defend line by line.