How we build the News board
Where the stories come from, how they're chosen, what AI writes, and how we credit each source.
What it is
The News board collects the AI news that matters to people who use AI in their work: model and product launches, regulation, companies, research, chips and energy, and effects on jobs and society. It’s updated several times a day, and every story links to its original source.
Where the stories come from
We read the public feeds (RSS, sitemaps, and news pages) of these sources:
- Labs and companies: OpenAI, Anthropic, Google DeepMind, Google, Microsoft, Meta, Mistral AI, Nvidia, and Hugging Face, plus the Google and Microsoft research blogs.
- English-language media: TechCrunch, The Verge, Ars Technica, The Decoder, MIT Technology Review, Wired, IEEE Spectrum, The Register, Tom’s Hardware, and DatacenterDynamics.
- Spanish-language media: Xataka, Genbeta, Hipertextual, El País, and elDiario.es.
- Regulation: the European Commission and the Spanish Agency for the Supervision of Artificial Intelligence (AESIA).
- Research: the day’s most upvoted arXiv papers on Hugging Face Daily Papers.
How stories are chosen and ranked
A program checks the sources every few hours, respecting the crawl frequency each one asks for. It merges into a single story what several outlets are reporting at the same time, and it discards anything that isn’t news: deals, buying guides, podcasts, events, or pieces where AI only comes up in passing.
Each story gets an importance score from 1 to 5, with people who use AI every day in mind. The ranking combines that score, how many outlets are covering the story, and whether it comes from the official source. The News board’s front page shows stories rated 2 or higher; each day’s archive shows all of them.
What part AI plays
The headline, the summary, “The 5 things that matter today,” and the “What changes for people who use AI” notes are written by an AI model (Claude, from Anthropic) following our rules: write in its own words, use only the facts the source provides, and don’t make up figures or details. To write each story, the program reads the original piece we link to (when the site allows it), and the model works only from that text.
Then a second automated pass checks the headline, summary, and notes against the original piece, sentence by sentence: it corrects anything that isn’t supported and drops the story if it can’t be verified. Even so, no human reviews each story before it’s published. That’s why every story links to its original: if something doesn’t add up, the source wins.
Anthropic, the company that makes Claude, is also one of the sources. Its stories go through the same rules as everyone else’s.
English edition
The News board is also published in English. What gets translated is our already-checked text (headline, summary, notes, and the day’s key points), using the same model with instructions not to add or remove anything; the source and link are the same. A story appears in English once its translation is ready, usually in the same pass in which it’s published in Spanish, and if the Spanish text is corrected, the translation is redone.
How we credit sources
We don’t copy text or images from our sources. Each story has its own headline and summary, the name of the outlet or organization, and a link to the original piece. When several outlets report the same thing, we link to them too. Information from the European Commission is reused under its license (CC BY 4.0), and AESIA’s under Law 37/2007 on the reuse of public sector information (Spain’s public sector information reuse law).
Errors and removals
If you spot an error, or you’re with an outlet and would rather not appear on the News board, write to us and we’ll correct it or take it down.