About Satcove News

What this is

A single AI answer can sound sure and still be wrong or incomplete. Satcove News puts the same reports in front of several models and publishes the comparison. When they agree, that is a useful signal. When they differ, that is where to look closer. It is not a replacement for the original reporting, which we always link.

How a story is made

1. 1

   We pick a story

   We follow stories reported by established news outlets. We link every original report so you can read it yourself.

2. 2

   Several AIs read it separately

   The same reports go to several AI models, each working alone. They do not see each other's answers.

3. 3

   We compare the answers

   We list the points each model makes and mark where they agree, where they differ and where one adds something the sources do not say.

4. 4

   We write it up

   The analysis shows the comparison, not a verdict. Models that failed to run are listed as skipped, never hidden.

5. 5

   You can check

   Prompts, outputs and scores are published. The open dataset lets anyone re-run the comparison.

How to read an analysis

The agreement score

A number from 0 to 100 for how similar the models' texts are. It measures similarity of wording and content, not truth: models can agree and still be wrong together.

Splits

A split is one point on which the models differ. We label each one:

Contradiction

Two models state opposite things about the same point.

Critical omission

A model leaves out a fact that changes how the story reads.

Neutral omission

A model leaves out a detail that does not change the meaning.

Emphasis

Models agree on the facts but stress different ones.

Unverified addition

A model adds a claim that the source reports do not contain.

Original research

Some pieces are our own studies, such as putting the same questions to six AIs. They are marked "Satcove original research" and have no outside report to compare against, so they carry no agreement score.

What we do not do

• We do not say who is right. A split is context, not proof that a source or a model is wrong.
• We do not give investment, medical or legal advice. Market quotes are delayed and for information only.
• We do not publish an analysis without linking the reports it is based on.
• We do not edit the models' answers to make them agree.

Sources, photos and data

Photos come with their credit and licence, often from Wikimedia Commons; where there is no licensed photo we show a generated chart or poster. Market quotes come from third-party data providers, are delayed, and indices are shown through ETFs (SPY, QQQ, DIA). The comparison data is published as an open dataset and the stories are available as an RSS feed.

Errors and corrections

Spotted a mistake or a missing source? Tell us through the contact page. We fix errors in the article and say so on the page.