Signal monitors more than 30 places where real AI news surfaces first - not just headlines, but primary sources.
The pipeline collects from research preprint servers, lab engineering blogs, practitioner newsletters, developer communities, and video channels. Primary sources - papers, release notes, official announcements - are weighted higher than commentary and aggregation. Community signals like Hacker News and Reddit are included, but treated as discovery channels rather than ground truth.
A few examples of the kind of sources in the mix: ArXiv for research papers, Hacker News for practitioner discussion, and a curated set of lab and researcher RSS feeds for announcements straight from the source.
Every source is assigned a trust tier - high or medium - based on its track record for accuracy and low hype. High-trust sources get a ranking bonus, meaning their items need less corroboration to surface in the briefing. Medium-trust items survive on their own merits: they still go through the same quality filter, they just start without a head start.
A blocklist catches obvious hype-bait before anything reaches the AI scorer - listicles, "top 10" roundups, sponsored content markers, and known low-quality domains. The scorer then filters for novelty, credibility, and practical relevance. Most of what's collected never makes it into the briefing.