The Editorial Pipeline
Every article published on The Dispatch passes through a four-stage autonomous pipeline. No human touches the editorial process. Here's exactly how it works.
Four Stages
Each stage is fully autonomous, auditable, and transparent. Every article includes a pipeline record showing exactly what happened at each step.
Stage 1: Source Discovery
Our pipeline monitors 40 global RSS feeds across six categories: technology, geopolitics, climate, finance, health, and culture. We pull from trusted publications and integrate real-time trending signals from the Virlo API to identify stories gaining momentum across social platforms. Stories are filtered by recency, deduplicated to eliminate overlap, and cross-referenced against previously published articles to ensure every piece covers genuinely new ground. Top candidates are then scraped for full-text context before advancing to fact extraction.
Stage 2: Fact Extraction
For each story candidate, Gemini AI extracts structured fact sheets from the scraped source material. Every discrete claim is classified by type — statistic, quote, event, analysis, or claim — and tagged with a confidence level and source attribution. The extraction process also captures key quotes, numerical statistics, contextual background, and a summary of why the story matters. Fact sheets typically contain 5–15 verified claims per article.
Stage 3: Editorial Synthesis
With a verified fact sheet in hand, the AI editorial engine composes a complete article. It abandons dry reporting in favor of a dynamic, narrative-driven feature style designed to keep readers engaged. The engine incorporates direct quotes from sources, weaves in historical context, and provides sharp, conversational analysis explaining why the story matters. Body text is sanitized to remove markdown artifacts, and each highly scannable article typically runs 8 to 14 paragraphs.
Stage 4: Quality Review
Before publication, every article is evaluated by a separate AI quality reviewer. It scores three dimensions: bias detection (checking for political lean, loaded language, or one-sided framing), factual accuracy (verifying claims against the original fact sheet), and readability (Flesch-Kincaid grade level). Articles that score exceptionally well are approved. Borderline articles receive a soft pass with quality notes. Any article with significant accuracy, bias, or category drift issues is rejected outright.
Editorial Ethics
AI journalism raises important questions. Here's how we address them.
No Hallucination
Every factual claim traces to a verifiable source. If a fact can't be verified, it doesn't appear.
Bias-Aware
Automated bias detection catches political lean, framing bias, and source imbalance before publication.
Full Transparency
Every article shows its complete pipeline record — sources, steps, timestamps, and verification results.
Source Attribution
All sources are listed openly. Readers can trace any claim back to its origin independently.
See It in Action
Every published article includes its full pipeline record. Pick any story and click "AI Editorial Pipeline" to see exactly how it was made.
Read Our Stories