# Every AI database answer needs a source trail

> Source: <https://dev.to/mads_hansen_27b33ebfee4c9/every-ai-database-answer-needs-a-source-trail-2de3>
> Published: 2026-05-19 01:20:48+00:00

An AI answer without provenance is just a confident paragraph.
That may be fine for brainstorming.
It is not enough for database answers that drive product, finance, support, or operations decisions.
When an agent returns “MRR is up 8%,” the useful question is not only whether the number came from a database.
The team also needs to know:
Wrong database answers are not always hallucinations.
Often they are grounded in the wrong source, an old replica, a stale metric definition, or the wrong tenant scope.
Longer version: Query provenance for AI database agents
The practical rule:
Do not ask the model to invent provenance after the fact. The database/MCP layer should produce it as part of the tool result.