The machinery behind what you see.
Internet culture and platform mechanics: micro-drama apps, viral storylines, recommendation systems, AI slop and the people caught inside all of it.
Feeds
A Sora mammoth clip shows the weak point in AI labeling: platforms disclose what they know on their own pages, then let downloads and screen recordings leave that context behind.
Cass Ito · August 11, 2026 · 7 min
Feeds
Autocomplete turns repeated curiosity into a durable label. A joke, denial or recycled clip can keep a rumor beside someone’s name long after its premise has fallen apart.
Reuben Adeyemi · 8 min read
Feeds
TikTok’s search box compresses popularity, personalization and controversy into the same few suggested words. The interface supplies no evidence, but it can still make an accusation feel pre-checked.
Cass Ito · 7 min read
Feeds
A suggested search can state the accusation that every video merely circles. TikTok then turns the gap between rumor and proof into another reason to keep watching.
Priya Nandakumar · 7 min read
Feeds
A local name can disguise a remote content operation built from police logs, weather alerts and recycled outrage. The useful clues sit in Page transparency, the website footer and the ad stack.
Reuben Adeyemi · 8 min read
Feeds
X’s crowd-sourced fact checks can produce credible corrections. The fake Pentagon explosion showed the catch: consensus arrives on human time, while falsehood travels at feed speed.
Cass Ito · 8 min read
Feeds
TikTok Shop search folds paid placements, commission-backed videos and seller ratings into one stream. Here is a repeatable way to separate product evidence from sales machinery.
Priya Nandakumar · 8 min read
Feeds
A suggested-search row can make an unverified celebrity rumor look organized, popular and ready to investigate. The menu is not evidence. Its packaging still changes the story.
Cass Ito · 7 min read
Feeds
A suggested query can attach an allegation to a person before anyone searches for it. TikTok offers reporting tools, but the named subject gets little visibility into why the prompt exists.
Reuben Adeyemi · 8 min read
Feeds
A platform warning is easy to crop away. The useful evidence lives in the file’s history, earlier copies, visual structure and the account that put it into circulation.
Priya Nandakumar · 8 min read
Feeds
Fake storm maps and disaster images copy the graphics people already trust. Platforms reward the recognition hit long before anyone checks who issued the forecast.
Cass Ito · 8 min read
Feeds
The viral Rotten Tomatoes crop looked definitive. Its repost trail shows how fandom screenshots lose dates, denominators and corrections while gaining authority.
Priya Nandakumar · 8 min read
Feeds
A rusted iron belt becomes proof of medieval cruelty after faceless videos copy the same weak claim. Search then mistakes repetition for corroboration.
Priya Nandakumar · 8 min read
Feeds
A complete watch of The Double Life of My Billionaire Husband shows how micro-drama episodes turn interrupted kisses and delayed reversals into a chain of coin purchases.
Reuben Adeyemi · 7 min read
Feeds
Creators are skipping the setup and handing viewers a mystery. The feed cannot measure comprehension, but it can measure how long confusion keeps your thumb still.
Priya Nandakumar · 8 min read
Feeds
TikTok can learn that you dislike one slime clip without learning that you dislike slime. A controlled feed test shows what to suppress, what to log and when to use stronger controls.
Priya Nandakumar · 8 min read
One update a day
One story each morning — no hype, no filler, no algorithm deciding for you.