How we choose and verify sources

In short: AI Daily and AI Daily Tech publish a story only when it rests on a primary source: the company, project or institution that did the thing, saying so itself. Independent press adds context and confirmation; aggregators are only where a story is noticed and are never cited. The working list has 140 sources in three tiers, each URL fetched and checked on November 11, 2026.

The three tiers

Tier 1 — primary (107 sources)
Official blogs, changelogs, release notes, documentation, repositories, papers and public filings: the party responsible for the news, in its own words. A story needs at least one. A claim about what a product does, costs or changes is only stated as fact when a Tier 1 source says it.
Tier 2 — trusted technical press (24 sources)
Independent reporting with editorial standards. It is used for context, for what a company did not say, and as the second, independent source. It is never the only basis of a claim, and what it reports is attributed to it by name.
Tier 3 — discovery only (9 sources)
Aggregators, trending lists and individual commentary. They show what people are talking about and point to the original. They are never cited and never the basis of a story: if the original cannot be found and read, the story is not published.

What a story needs before it is published

These rules are checked by the publishing system itself, not left to judgement. A story that does not meet them is rejected, and a day with no story that meets them has no story.

  • At least one primary source, marked as such on the article.
  • A second source, or a written explanation of why there is only one. That explanation is shown to the reader on the article ("Single-source story").
  • The event happened today, in the editorial timezone (Buenos Aires). The date of the event is stored separately from the date of publication so that a late publication cannot pass yesterday's news off as today's.
  • Every source is listed under the article with its publisher, a working https link, what kind of source it is, and the day it was read.
  • The English and Spanish versions tell the same story: same facts, same number of key takeaways. Neither is a machine translation of the other.

Who we read

The list is a working tool, not a wall: a story can cite a primary source that is not on it (a company covered for the first time), under the same rules. Verified on November 11, 2026.

Tier 1 — primary
Amazon Web Services, AMD, Angular, Anthropic, Apple, Arm, arXiv, Astro, Bun, ClickHouse, Cloud Native Computing Foundation, Cloudflare, Cybersecurity and Infrastructure Security Agency, DeepSeek, Deno, Docker, DuckDB Foundation, Elastic, Expo, Fly.io, GitHub, Go, Google, Google Cloud, Google DeepMind, Groq, Hugging Face, Internet Engineering Task Force, Meta, Microsoft, Mistral AI, Model Context Protocol, MongoDB, Mozilla, National Institute of Standards and Technology, Neon, Netlify, NVIDIA, Ollama, OpenAI, OpenJS Foundation, OpenRouter, PlanetScale, PostgreSQL Global Development Group, Prisma, Python Software Foundation, Raspberry Pi, React, React Native, Redis, Stripe, Supabase, Svelte, Tailwind Labs, The Kubernetes Authors, The Linux Foundation, The Linux Kernel Organization, The Rust Foundation, Together AI, V8, Vercel, Vite, Vue.js, WHATWG, World Wide Web Consortium, xAI
Tier 2 — press
404 Media, Ars Technica, BleepingComputer, El País, Genbeta, Hipertextual, IEEE, Infobae, InfoQ, Krebs on Security, LWN.net, MIT Technology Review, Phoronix, Rest of World, SemiAnalysis, Stack Overflow, TechCrunch, The New Stack, The Register, The Verge, Tom's Hardware, WIRED, Xataka
Tier 3 — discovery
GitHub, hnrss.org, Hugging Face, Lobsters, Product Hunt, Reddit, Simon Willison, Techmeme, Y Combinator

AI Daily and AI Daily Tech

AI Daily covers what happened in AI: models, products, research, policy, funding. AI Daily Tech covers one technology story a day — cloud and hosting platforms, web frameworks and runtimes, developer tools, hardware — and only when it has something to do with AI or with building software. A changelog entry is not a story because it exists; it is one when it changes what someone who builds with these tools can do, must do, or pays.

Both editions follow the same rules on this page. Both are researched with AI assistance and signed "Promptea Editorial": there is no invented human byline.

Corrections and late stories

  • A published article is never edited silently. A correction keeps the same URL and the original publication date, marks the article as corrected and shows a note saying what changed and when.
  • An article that should not have been published is withdrawn: it leaves the index, the sitemap and the feeds, and its address stops working.
  • A story about an earlier event can only be published by a person, on purpose, up to 14 days late, and the article shows the date of the event, the date of publication and the reason. The automated routine cannot do it.

Found an error? Promptea is built by eterlab: write through the Feedback link at the bottom of any page, or at eterlab.co.

What this does not guarantee

  • A primary source is primary for what its author says, not proof that it is true. A company's announcement is reported as an announcement; a preprint is reported as a preprint.
  • Research and writing are assisted by AI and checked against the sources by an automated process. There is no human editor reading every story before it goes out.
  • Being on the list is not an endorsement, and a source can be removed when it stops being reliable or reachable.

Frequently asked questions

Do you cite Hacker News, Reddit or social media posts?

No. Aggregators and social feeds are Tier 3: they are where a story is noticed. The article cites the original document they point to, after reading it.

Can a story have only one source?

Only when that source is primary and the article says why there is no second one yet, for example an announcement nobody else has reported on. The explanation is shown on the article.

How is the list kept up to date?

Every URL on it is fetched by a script, and an address that does not answer is not added. The last full check was on November 11, 2026; the date on this page changes with each new check.

Last updated:

Read AI DailyRead AI Daily TechHow prompts are scored