Ask an LLM a GIS question, and it answers like a math problem: one plus one equals two. GIS data isn't that simple. If a jurisdiction maintains 5,000 miles of road, that number comes from centerlines, so it excludes split roadways and says nothing about who's actually responsible for which side: a county might maintain one side, a city the other. A GIS professional works through that without being asked. An LLM doesn't. GeoProof surfaces the complexity up front instead of waiting for the follow-up question a GIS professional already knows to ask.

GeoProof and ProofSimple are a pair of open-source projects in the MapSimple suite, built to close that gap. GeoProof is an MCP server that sits between AI agents and spatial data, attaching five dimensions to every answer: accuracy, completeness, freshness, provenance, and authority. Authority is assigned to the individual contributor or data steward responsible for the data and surfaced through the response. Sensitive data is masked before a response is ever returned. It runs at production speed, is fully configurable, and carries no license fee. ProofSimple is the reference chat client that consumes GeoProof and makes its work visible: the layer determines what constitutes a trustworthy answer, and the client shows the user what that looks like. You need both to see either one work.

The talk walks through a live GeoProof and ProofSimple question, showing how each trust dimension maps to a response, how masking works, and how authority traces back to a named contributor rather than a generic or no source. Attendees see the same question answered in two ways: a fast, confident number and a complete answer with all the caveats.

Attendees leave with the same five-part framework for judging whether any GIS answer, AI-generated or not, deserves their trust. The framework applies whether or not they ever run GeoProof and ProofSimple: this is one implementation of an idea the field will need as more spatial questions are answered by machines.