RAG

Why RAG matters

Ask a model to write a piece cold and you get a generic article. With RAG, the model first looks at what reputable outlets already wrote on the topic and builds the plan and text around those fragments.

Two effects. First, the material is factually grounded rather than invented. Second, it echoes the style of sources that AI already cites well, raising the chance your publication gets cited too.

How RAG is populated

A four-step chain that runs automatically:

1. Link harvest. The platform collects every link AI services include in answers to your queries. Runs daily.
2. Content fetch. The system grabs the page text behind each link. Not images or ads, just the main article body.
3. Chunking. Long texts are split into short meaningful pieces. The model can cite a single point instead of a whole page.
4. Query lookup. When you open a card, the best-matching fragments for its topic are surfaced. They land on the RAG tab.

What you see in a card

The RAG tab lists fragments: the snippet text, source domain, page title, and a link to the original.

These fragments power the Brief tab when generating the plan and the Content tab when generating the body. If the picks aren't relevant, you can swap some manually and regenerate the material.