Tools · 02 · built here

The GEO Scorer.

Paste a URL or a block of text. The scorer decomposes it into atomic statements and predicts, claim by claim, how likely each one is to be retrieved and reproduced by a frontier model.

Built herev0.1 · preview
§ 01
What it scores

Six signals, weighted by survival.

The score is built from the structural features that distinguished surviving claims from non-survivors across a controlled corpus of 3,200 documents.

01 · high

Entity specificity

Named entities — companies, products, people, places — survive retrieval at 3.4× the rate of generic categories.

02 · high

Quantification

A claim with a number — percentage, count, date, measurement — survives at 2.1× the rate of one without.

03 · highest

Source proximity

Statements published close to a verifiable primary source survive at 4.7×.

04 · medium

Self-containment

Claims that survive chunking intact — the assertion and its qualifier in the same retrievable span.

05 · medium

Epistemic framing

Hedged, sourced, defensible prose is trusted over confident marketing assertion.

06 · variable

Contradiction risk

Claims the model already holds a competing belief about score lower until reconciled.

Need the real thing?

The GEO Scorer is a preview heuristic. The LLM Visibility Audit runs the full controlled probe against fourteen frontier models with a 30-day persistence window.

See the full audit →