INDEX 46 ▲1 todaySPLIT OF THE DAY Broadcom to lend Anthropic up to $42 billion for chip leases50 STORIES · 387 REACTIONSANTI-AI 75% · MIDDLE GROUND 20% · PRO-AI 5%LATEST Wes Moore forms bipartisan governors group on AI

Editorial policy

Last updated September 26, 2026

How a story gets onto Splitfeed, how a quote gets into the Split, how a camp gets assigned, and where a machine did the work and where a person did.

1. Sourcing

Splitfeed does not report news. An automated pipeline reads public RSS feeds from a fixed list of outlets and lab blogs: technology press, general news, research institutions and the labs' own announcement pages. The list is editorial and changes only by a deliberate decision, never automatically.

  • Coverage of one event from several outlets is grouped into a single story. Each grouped article is listed in the story's Sources section with the outlet, the headline and a link.
  • We use only the title and the summary each publisher chooses to syndicate in its feed. We do not scrape article bodies, bypass paywalls, or reproduce text beyond what the feed carries. Readers go to the publisher for the full article.
  • A story is published when enough independent coverage exists to establish the event. Stories that rest on a single thin source are held or removed.

2. Attribution

  • Quoted reactions are verbatim. We trim for length, marked where it changes the flow, and never paraphrase.
  • Every quote links to the original post on the platform it came from.
  • Authors are named as they present themselves on the platform: display name and handle. We do not add identifying information the author did not publish.
  • Reactions are collected from public posts only: Hacker News comments, Bluesky posts, selected Reddit threads, a watchlist of accounts we always read, and the Splitfeed YouTube channel.
  • An author who wants a quote removed can ask, and we remove it.

3. Camp assignment

Each published reaction is filed under one of three camps. The label describes the take on that particular story, not the person, and the same author can sit in different camps on different days.

  • Pro-AI: this is progress; build it, ship it, lean in.
  • Anti-AI: this is dangerous, moving too fast, or costs people more than it gives.
  • Middle Ground: cautiously optimistic, nuanced, conditional, or wait-and-see.

Software proposes a camp and a confidence for each reaction. A person reviews proposals before publication and can change the camp, mark a reaction as skeptic (dismissing the story as overblown, whichever camp it sits in), or reject it. Reactions that are off-topic, abusive, or not about the story are not published. Low-confidence proposals always go to review; nothing with an unclear camp is published automatically.

4. AI-assisted drafting, human review

Splitfeed uses an AI language model to draft parts of every story. Specifically, the headline, the summary, the key points and the analysis section are generated by a model that is given only the linked sources and the published reactions, and is instructed to write only what those sources establish: no speculation, no framing words, no invented detail, and no claims about facts the sources do not contain. Drafts that fail those rules are discarded.

A person reviews stories before they are published and re-reads them when new sources or corrections arrive. The byline “Splitfeed Editorial” on a story means exactly this: drafted with AI assistance from the sources shown, reviewed by a human, and the responsibility of the editorial team. Reactions and reader takes are never generated or altered by a model.

5. The Doom/Boom Index

The Index runs from 0 (full Doom) to 100 (full Boom). Every story carries a score of its own judged on what happened, and every published reaction pulls that score toward its camp: Pro-AI toward 100, Anti-AI toward 0, Middle Ground toward 50. Reader votes count the same way. The Index is the weighted average across the day's stories, weighted by coverage and reaction volume, so a story many outlets cover and many people react to counts for more. The live number updates through the day; the daily value is frozen at midnight UTC. The method is described in more detail on How it works.

6. Reader takes

Signed-in readers can vote for a camp on any story and post a take of up to 500 characters, filed under the camp they voted. One vote per reader per story, changeable at any time. Takes are plain text, shown with the reader's Google display name, and can be removed by the reader. We remove takes that are abusive, spam, or off-topic, and accounts that repeat that behaviour lose posting.

7. Corrections

Corrections are handled ahead of all other mail. Email hello@splitfeed.ai with “Correction” in the subject, the story link and what is wrong. Clear-cut attribution errors are usually fixed the same day. Corrected stories show a “Last updated” stamp, and material corrections carry a note describing the change. Our About page has the full correction policy.

8. Independence and advertising

Splitfeed has no ownership or financial ties to any AI company, lab or investor. The site is free to read and supported by advertising served through Google AdSense. Advertising is kept apart from editorial: advertisers do not choose stories, do not see stories before publication, and have no influence over summaries, camps or the Index. There are no sponsored stories and no paid placement in the Split. Affiliate links, where they appear, are labelled.

9. Contact

Questions about this policy, feed suggestions and watchlist nominations all go to hello@splitfeed.ai. See the contact page for response times.