Teaching AI-written news when to stop hedging

Ask an AI to summarize a news story and you'll usually get something cautious: "according to reports," "it appears that," "sources suggest." That hedging is often the right call — a single claim shouldn't be stated as settled fact. But when several independent outlets are all reporting the same thing, hedging every sentence stops being honest caution and starts being noise. Rhyter Times's story synthesis is built around that distinction.

The idea

Every write-up on Rhyter Times is original — written in our own words, never copied from any outlet — and always credits its sources below the piece. But the voice changes with how corroborated a fact is:

• When independent outlets are still lightly aligned, the tone stays attributive: "according to X and Y."
• When enough independent outlets agree, the piece states the fact directly. The corroboration itself is the confidence signal, so the writing doesn't need to hedge every sentence to be honest.

What never changes, regardless of confidence: nothing is stated that isn't actually supported by the source material. Confidence in tone and license to invent are two completely separate things, and only the first one moves.

We say exactly what this is

Every synthesized piece carries a plain disclosure explaining that it's AI-assisted synthesis of existing reporting, not original investigative journalism — cross-referencing convergent reporting from multiple named outlets is a real, if limited, form of verification, but we never let a confident tone imply it's something it isn't.

A related but different feature

Rhyter Times also has a Perspectives view that's easy to confuse with this but does something different: instead of measuring how much outlets agree, it shows how coverage of the same story differs by outlet across the political spectrum, built without any AI writing involved at all.

Both are live at rhytertimes.com, and the same synthesis is now available through an API and MCP server for anyone building on top of it. Curious what others think about corroboration as a proxy for confidence in AI writing — happy to discuss in the comments.