The steps
- Collection. The pipeline runs at the top of every hour and reads the sources listed below. New items are cleaned up and stored with their original URL.
- Grouping. Items about the same story are grouped by URL and by text similarity. A story covered by only one source is not summarized.
- Summary. Stories with two or more sources are summarized by a Gemini Flash model. The prompt allows only facts found in the sources and requires every paragraph to cite the sources it draws on. A summary that cites a source outside the group is discarded.
- Review. JT reviews a sample of published stories each day and adjusts the prompt and the source list based on what the review finds.
- Publication. Summaries appear on /discover and on the topic pages, with the full source list under each story. The public page puts commerce, marketing, social, creator, AI and platform stories first and leaves out sport, politics, grief and celebrity stories.
Sources
- News services: GDELT, GNews, Mediastack, NewsData, and Reuters
- Community threads: Reddit and Hacker News
- Social posts: Bluesky, Pinterest, and YouTube
Report an error
For a misattribution, a missing source, or a summary that misreads its source, email hello@mora-marketer.com with the subject line “Discover correction”. The editorial policy explains how corrections are made.