MediaReceipts — Accuracy, over time.

The Mission

Building the record on media accuracy

Existing fact-checkers focus primarily on politicians. Outlet-level bias raters score networks as institutions. No one tracks individual cable news personalities — the anchors and commentators who shape how millions understand the news — with a consistent, transparent methodology applied equally regardless of network or political orientation.

MediaReceipts fills that gap. We process broadcast transcripts, extract verifiable factual claims, verify them against authoritative primary sources, and publish what we find. Every verdict, every source citation, and the full scoring methodology are publicly auditable.

We are nonpartisan by structure, not just by intention. Inaugural tracked figures are selected simultaneously from CNN, Fox News, and MSNBC. The same rubric, the same standard, the same process — regardless of who said it.

Methodology

How it works

Every published verdict passes through a six-stage pipeline designed to prioritize accuracy over speed. AI assists at scale; humans make the final decisions.

1

Broadcast

A tracked media personality makes a statement on air.

2

Transcript Ingestion

The broadcast transcript is acquired and timestamped.

3

Claim Extraction

AI identifies individual verifiable factual claims — statistics, legislative details, economic figures. Opinions and editorial commentary are excluded.

4

Gate 1 — Human Review

A human editor reviews each claim and decides whether it moves forward for a full verdict. The editor can override the AI at any point.

5

Gate 2 — Verdict Assignment

A second review produces the final verdict with full reasoning, primary source citations, and confidence assessment. No verdict publishes without editorial sign-off.

6

Publication

Verdicts enter the public database. Individual scorecards update to reflect the longitudinal record.

Verdict Categories

Four categories, clearly defined

Every claim receives exactly one verdict. The categories are mutually exclusive, operationally defined, and applied identically regardless of who made the claim.

True

The claim is fully supported by authoritative sources with no material omissions.

Mostly True

The core assertion is accurate but contains imprecision, rounding, or minor contextual gaps that do not change the fundamental conclusion.

Misleading

The claim contains a factual element but omits, frames, or presents information in a way that would lead a reasonable viewer to a substantially incorrect understanding.

False

The claim's core factual assertion is contradicted by authoritative sources.

Source Standards

What counts as evidence

Every verdict is grounded in a defined source hierarchy. Higher-tier sources carry more weight. No verdict is based solely on secondary reporting.

Tier 1

Primary institutional records. Government data, congressional records, court filings, census data, peer-reviewed indexed research, official transcripts, legislation text.

Tier 2

Wire services and official statements. AP, Reuters, AFP reporting that cites primary sources. Official organizational statements and press releases.

Tier 3

Context and corroboration only. Commentary, editorial analysis, and secondary reporting. Never the sole basis for a verdict.

Design Principles

What guides the work

Nonpartisan by structure

Balanced editorial selection across networks and the political spectrum. Quarterly monitoring for output imbalance. Independent advisory board oversight.

Longitudinal, not reactive

One claim doesn't tell you much. Patterns over quarters tell a meaningful story. We'd rather be right than first.

Human-verified

AI assists at scale. Humans make every final decision. Two mandatory editorial review gates before any verdict publishes.

Radically transparent

Full methodology published. Every verdict shows its sources. Correction log public. If we make a mistake, we correct it openly.