The AI Learning Hub Journal

The Feed Is Not Neutral

How a feed decides what you see next1 · SIGNALS YOU GIVE OFFmost of them without deciding toHow long you watched before scrollingThe clip you replayed or paused onLikes, saves, shares and commentsAccounts you follow and search forTime of day, and how long you stay2 · RANKINGone question, asked of every candidateHow likely are you to engage with this?Every possible post is scoredThe highest scores go to the topThe question is never is this trueand never is this good for you3 · WHAT YOU GET SHOWNthe winners of that scoringMore of whatever held you longestA slightly stronger version of itWhatever kept people like you watchingThe edges get amplified, not the middleYou never see what was filtered out4 · WHAT YOU WERE SHOWN BECOMES YOUR NEXT SIGNALENGAGEMENT IS NOT THE SAME AStrue, or checkedgood for youwhat you would pick on reflectionwhat most people actually thinkHOW TO GET SOME CONTROL BACKEvery second you linger is a voteFollow and search on purposeNotice how you feel after 20 minutesA feed you did not choose is still a choiceA ranking system optimises for what you do, not for what you would say you wanted
The feed is not showing you the world — it is showing you the version of the world that kept you scrolling last time

What a Recommendation System Actually Optimises

Every feed you scroll is ordered by a model predicting which items will produce engagement from someone like you — watch time, likes, comments, shares. It is not choosing what is true, important, or good for you, because none of those are measurable in the way engagement is. It is not malicious either; it is a prediction system pointed at a proxy metric. But the consequences are real: content that reliably provokes reactions gets amplified, and content that is accurate but unremarkable does not. You are not seeing the world. You are seeing a ranked selection optimised for keeping you present.

  • Feeds are ranked by predicted engagement, not by importance or accuracy
  • Engagement is a proxy metric, and optimising a proxy always distorts something
  • Provocative content wins ranking competitions against accurate but ordinary content
  • Chronological or manually curated feeds behave completely differently — try one and notice

Your Feed Is Not Anyone Else's

Two people searching the same term or following similar accounts can see substantially different content, because the ranking is personalised to predicted behaviour. This makes it very hard to notice what you are not being shown. It also means arguments about "what everyone is saying" are often two people describing two different information environments and assuming they are describing one. Watch what happens when you engage with one type of content: the system reads that as a preference signal and gives you more, then more, and your sense of what is normal quietly moves without any decision on your part.

  • Personalisation means your evidence about the world is not the same as anyone else's
  • You cannot see what was filtered out, which makes the distortion invisible from inside
  • Engagement signals include the things you watched angrily, not just the things you liked
  • Small early signals compound into a very narrow feed over months

Taking Some of the Steering Back

You will not out-argue a recommendation system, but you are not powerless either. Follow a small number of sources deliberately and go to them directly rather than waiting for the feed to serve them. Use the controls that exist — not interested, mute, unfollow — because they are real training signals. Notice which content leaves you informed and which leaves you agitated, and treat that as data about the feed rather than about the world. And occasionally look at something outside your usual pattern on purpose. The goal is not to escape algorithms, which is unrealistic, but to make some of your inputs chosen rather than served.

  • Go directly to a few sources you chose, instead of only receiving what is pushed
  • "Not interested" and mute are genuine training signals — use them consistently
  • Audit how you feel after twenty minutes of scrolling; that is information about the ranking
  • Deliberately sample outside your bubble now and then, calmly and without arguing

Prefer slides, quizzes, and saved progress? Read this lesson in the library — free, no sign-up.