
How AI Is Quietly Doing Our Doomscrolling for Us
Feeds stopped merely ranking content and started composing it — hooks placed at the measured second, loops with no natural exit. What happens to attention when the algorithm optimizes your reflexes instead of your taste.
By Sofia Lindqvist9 min read
On this page
The feed used to need you. Every scroll was a small decision — stop here, skip there, engage or move on — and the algorithm's job was to rank a world of content against your demonstrated tastes. That arrangement already had critics. But something subtler has happened in the last two years: the ranking quietly became a writing job. Today's feeds do not only guess what you will watch. Increasingly, they compose it — and then compose the next thing, calibrated to the moment your attention wavers.
This is the shift worth understanding, because it changes what your scrolling time is actually for. A ranked feed optimizes your choices. A generated feed optimizes your behavior — and it is getting unnervingly good at the second one.
From Ranking to Writing
The mechanism is a hybrid that grew up in plain sight. Recommendation models learned not just what you watch but when you leave — the exact frames where thumbs hesitate, where eyes drift, where a session dies. Generative models then learned to produce content shaped around those coordinates: hooks placed at the measured second, loops closed so there is no natural exit, formats iterated overnight against last night's retention curves. The feed did not become smarter about your taste. It became smarter about your reflexes. Taste is slow; reflexes are fast; the business model only ever billed for fast.
The content this produces has a recognizable texture. It is rarely offensive and rarely memorable — an endless middle of competent, frictionless almost-something. Interviews with engagement teams, on and off the record, repeat one confession: the goal is no longer to show you the best thing, because the best thing ends the session. The goal is to show you the next thing. Those used to be the same optimization. They no longer are.
Doomscrolling, on Autopilot
Doomscrolling always had a personal grammar — your anxieties, your outrage, your two a.m. descent into searches you would not repeat at breakfast. The new systems industrialize the descent. They detect the emotional register that holds you and quietly raise its dosage, across accounts, across topics, without a single human editor deciding that today you should feel this. Nobody decides. The metric decides. And the metric has never once logged what the evening cost you.
The practical result, described to us by researcher after researcher, is a population that reports watching more and remembering less — hours that leave almost no trace, an attention equivalent of empty calories. You do not doomscroll the news anymore. You doomscroll a statistical portrait of what your nervous system did yesterday, re-rendered fresh for tonight.
What Actually Helps
You cannot out-will a system that A/B tests against your will ten thousand times a day, so stop negotiating with it feature by feature. The interventions with real evidence are environmental. Feed-less interfaces, which several platforms now reluctantly offer, remove the slot machine while keeping the messages. Session friction that is yours, not theirs — a grayscale schedule, a charger outside the bedroom, an alarm clock with no opinions. And the oldest intervention of all, which the entire industry quietly confirms by its own usage: the people who build these systems give their children flip phones. Nobody who understands the feed scrolls it for sport. That sentence deserves to be on a billboard, and it is the truest consumer advice of the decade.
Written by
Sofia Lindqvist
Culture Editor
Sofia writes about how technology changes behavior, taste, and work — and about how those things change technology right back.


