What This Site Picked

What happens in a day

2026-09-23 · recorded for 47 days

1. 1,619 feed items
2. 135 last 24h
3. 86 main candidates
4. 10 published

How the 135 last-24h items split

• main-pool candidates86
• separate Korea pool44
• self-promo / discussion4
• already published0
• outside AI scope1

Checks out: 135 = 86 (main) + 44 (Korea pool) + 4 (self-promo) + 0 (duplicate) + 1 (off-scope).

The Korea pool of 44 breaks down into 36 candidates, 1 dropped as promotional, and 7 dropped as outside AI scope. Korea stories are chosen from a separate pool and never mix with the main candidates.

Where the 86 candidates went

• published 10
• outranked 66
• same story 8
• 3-per-outlet cap 2
• original unreadable 0

The bar above counts what happened at selection time. In practice 3 of these 10 started as different articles and were swapped for backups after we read the originals.

• OpenAI News original unreadable
• AI Times duplicate story
• MIT Technology Review AI no usable facts

Across the 36 days on record we swapped 70 articles · duplicate story 50 · promotional 9 · original unreadable 6 · no usable facts 3 · not news 2.

Recent selection log

Only days with a recorded funnel. "Published" is the actual count; "swaps" is how many were replaced by backups. Days before swap-tracking show —.

Stories with a next step

On 2026-09-23 it was 3 of them, and 49 across the 28 days on record. We only write one when there is a screen to open or a setting to change today. When there is nothing to do we leave it out. Ten out of ten every day would mean writing sentences that look like actions rather than actions.

What came in, what ran: 47-day total

Across the 47 days with candidate records. Bar length is what came in; the solid part is what ran.

The "% of candidates published" is not a quality score for an outlet. It is affected by story volume, duplicate coverage, daily competition, and the per-outlet cap.

• AI Times 1554→87 5.6% AI Times, 87 of 1554 candidates published, 5.6 percent of candidates
• TechCrunch AI 371→109 29.4% TechCrunch AI, 109 of 371 candidates published, 29.4 percent of candidates
• The Decoder 220→60 27.3% The Decoder, 60 of 220 candidates published, 27.3 percent of candidates
• The Verge AI 179→47 26.3% The Verge AI, 47 of 179 candidates published, 26.3 percent of candidates
• Hacker News (AI) 151→0 0.0% Hacker News (AI), 0 of 151 candidates published, 0.0 percent of candidates
• Wired AI 122→54 44.3% Wired AI, 54 of 122 candidates published, 44.3 percent of candidates
• SiliconANGLE AI 93→6 6.5% SiliconANGLE AI, 6 of 93 candidates published, 6.5 percent of candidates
• Ars Technica AI 90→41 45.6% Ars Technica AI, 41 of 90 candidates published, 45.6 percent of candidates
• VentureBeat AI 70→8 11.4% VentureBeat AI, 8 of 70 candidates published, 11.4 percent of candidates
• OpenAI News 49→1 2.0% OpenAI News, 1 of 49 candidates published, 2.0 percent of candidates
• MIT Technology Review AI 39→22 56.4% MIT Technology Review AI, 22 of 39 candidates published, 56.4 percent of candidates
• Google AI Blog 22→14 63.6% Google AI Blog, 14 of 22 candidates published, 63.6 percent of candidates
• Google DeepMind Blog 12→10 83.3% Google DeepMind Blog, 10 of 12 candidates published, 83.3 percent of candidates
• MarkTechPost 4→2 50.0% MarkTechPost, 2 of 4 candidates published, 50.0 percent of candidates

Which outlets we ran

14 sources in total. No outlet can take more than 3 of the 10 slots in a day. Percentages are of all 617 published stories.

• TechCrunch AI 15825.6% TechCrunch AI, 158 of 617 published stories, 25.6 percent
• AI Times 9214.9% AI Times, 92 of 617 published stories, 14.9 percent
• Wired AI 6911.2% Wired AI, 69 of 617 published stories, 11.2 percent
• Ars Technica AI 6510.5% Ars Technica AI, 65 of 617 published stories, 10.5 percent
• The Decoder 609.7% The Decoder, 60 of 617 published stories, 9.7 percent
• The Verge AI 609.7% The Verge AI, 60 of 617 published stories, 9.7 percent
• MIT Technology Review AI 243.9% MIT Technology Review AI, 24 of 617 published stories, 3.9 percent
• VentureBeat AI 213.4% VentureBeat AI, 21 of 617 published stories, 3.4 percent

What the stories were about

Each percentage is the share of all 617 published stories that carry that topic. A story can have more than one topic, so the counts can exceed the story total and the percentages can add up to more than 100%.

• Business & Funding 16326.4% Business & Funding, 163 of 617 published stories, 26.4 percent
• Products & Services 14323.2% Products & Services, 143 of 617 published stories, 23.2 percent
• Safety & Alignment 14122.9% Safety & Alignment, 141 of 617 published stories, 22.9 percent
• Agents & Dev Tools 11218.2% Agents & Dev Tools, 112 of 617 published stories, 18.2 percent
• Policy & Regulation 10316.7% Policy & Regulation, 103 of 617 published stories, 16.7 percent
• Models & Research 9715.7% Models & Research, 97 of 617 published stories, 15.7 percent
• Society & Work 8013.0% Society & Work, 80 of 617 published stories, 13.0 percent
• Infrastructure & Chips 6811.0% Infrastructure & Chips, 68 of 617 published stories, 11.0 percent
• Open Source 284.5% Open Source, 28 of 617 published stories, 4.5 percent
• Generative Media 254.1% Generative Media, 25 of 617 published stories, 4.1 percent
• Robotics & Physical AI 223.6% Robotics & Physical AI, 22 of 617 published stories, 3.6 percent
• Data & Training 193.1% Data & Training, 19 of 617 published stories, 3.1 percent

Terms we explained

Each term is explained on the day it actually appeared in a briefing. We don't write a dictionary separately. Repeats across days are counted once. Browse the glossary →

How these numbers are counted

You can download this page's aggregates as machine-readable JSON: selection data (JSON)