Field notes · 2026-08-22
Why keyword filters aren’t enough
The useful part of a feed filter is understanding what a post means, not simply spotting a word.
A keyword is a blunt instrument. The word “crypto” might appear in a scam promotion, a thoughtful explanation of a protocol, or a friend’s joke. A filter that treats those three posts as identical forces you to choose between noise and missed context.
That is the uncomfortable trade-off behind most feed controls: they are easy to explain, but they make you do the hard work. You keep adding words, exceptions, and people to a list until the list becomes another thing to maintain. Meanwhile, the post you actually wanted to avoid changes its wording and slips through anyway.
Sift starts with cheap, local checks and only reaches for semantic matching when the obvious signals are not enough. That keeps everyday filtering fast while still letting a rule such as hide hype and get-rich-quick pitches describe an idea rather than a vocabulary list.
The useful distinction is intent. A post can mention a topic without being the thing you are trying to avoid. A good filter should understand that difference, show its reasoning, and let you correct it when it misses. Those corrections become part of your preferences instead of disappearing into a black box.
The result is a feed that follows your intent more closely. You can hide a category without maintaining a constantly changing blacklist, and every decision remains visible in a receipt so you can tune the rule when it gets something wrong.