AI use

What is written by AI on Raven Wire, and what isn't.

Written by AI

  • The analyst read on top stories: a summary, what is established, the claims and their status, contradictions, state-media framing, what to watch, and a solid-to-doubtful rating.
  • The “Why it matters” line on some stories.
  • The “Connecting the dots” brief at the top of the Live page, which links stories that may be related.

These are labelled where they appear (“Analyst read”, “AI analysis”) and name the model. The model is Anthropic’s Claude, called through Vercel’s AI Gateway.

Not written by AI

  • Grouping reports into stories, the confidence score, the verdict and the alert level: these come from published rules (Editorial standards, Method).
  • The one-line summary under each headline, built from those rules.
  • Headlines: each story uses a headline from one of its reports.

What the AI sees and can change

  • It sees only the reports in the story (and, for the brief, the list of live stories) and is told to add nothing from elsewhere. From most publishers that means the headline only; the feed’s short description only where the publisher allows it; nothing at all from publishers that reserve AI use. Details: For publishers.
  • It never sets a score and never raises an alert level.
  • It can make us more careful: when its read is “doubtful”, the story’s verification becomes disputed and its alert level is capped at L2.

Review

AI text is published automatically; a person does not review each one before it appears. That is why it is labelled, why every story links to its reports, and why the scores never depend on it. If an AI read says something the reports don’t support, please tell us.

Last updated November 17, 2026.