After my previous post on passive voice and hedging by news topic, I looked at the same idea at the outlet level. So I ran the same style of metadata-only linguistic analysis across 8 news outlets from 2016–2026.

The two metrics shown are:

Hedging Rate : Share of sentences containing uncertainty/speculative language, such as “may,” “might,” “could,” “reportedly,” or “allegedly.”

Passive Voice Ratio : Share of sentences detected as passive voice, used here as a rough signal for less direct agency/attribution structure.

The analysis is filtered to hard-news topics and excludes sports, entertainment, lifestyle, weather, and similar categories.

Important caveats:

This is not a truthfulness ranking.

It measures linguistic style, not factual accuracy.

Topic mix and article type can affect the results.

Some source trend lines/panels cover fewer years because this sample applies minimum-per-year data thresholds and only plots years with sufficient observations.

The y-axis is zoomed to the observed range;…