- Four situations
- You choose, then we debrief
- Zero “delete the app” lectures
“But I never liked anything like that”
A friend’s feed has filled up with videos about a breakup drama between two creators. “I never liked a single one of those,” they say. “I don’t even follow them. Why does it keep showing me this?” True on both counts — they never tapped like once. They did, however, watch about thirty of the videos all the way to the end.
Which of their actions told the algorithm the most?
What’s actually going on
Likes are the signal you know you’re sending. Watch time is the one you don’t. TikTok’s own published explanation of the For You feed says a strong indicator of interest — like finishing a longer video from beginning to end — gets more weight than weaker signals, alongside rewatches, shares, comments, and follows. YouTube’s engineering team says the same about its recommendations: clicks, watch time, shares, likes, and survey answers all feed a system that processes tens of billions of signals. The feed isn’t listening through your microphone; it doesn’t need to. Every second you linger is already a vote.
That’s the core skill of this whole module: the algorithm can’t tell the difference between “I love this” and “I can’t look away.” Both look identical from the data side. Hate-watching, stress-watching, and “one more to see how dumb this gets” all count as demand.
From one gym video to somewhere darker
Someone watches a few workout videos before tryouts. A week later their feed has drifted: less “how to train,” more “what’s wrong with your body.” Extreme diets. Before-and-afters that feel bad to look at — but hard to skip. They didn’t search for any of this.
Why did the feed go there?
What’s actually going on
This is the “rabbit hole” effect, and it’s not a conspiracy theory — TikTok itself acknowledged it publicly in December 2021, naming “extreme dieting or fitness, sadness, or breakups” as exactly the kind of content where too much of the same thing can be harmful, and saying it was testing ways to break up repetitive streams. The mechanism is the one from scenario 1 running in a loop: intense content is hard to skip, hard-to-skip reads as interest, interest gets you more of it, and each round is a little more concentrated than the last. Nobody at the company chose your rabbit hole. The optimization did.
The tell that you’re in one isn’t the topic — it’s the narrowing. A healthy feed is varied; a rabbit hole is the same emotional note over and over. Once you can name that, you’ve caught it early. What to actually do about it is scenario 4.
“The algorithm just knows me”
Two friends compare feeds at lunch. Completely different worlds — different sounds, different drama, different everything. One shrugs: “It just knows me. My feed is basically my personality.”
What’s wrong with “my feed is my personality”?
What’s actually going on
The system doesn’t know you. It knows a version of you assembled from watch time — and as scenario 1 showed, watch time can’t tell fascination from car-crash-can’t-look-away. If rage-bait holds you for eight seconds longer than the stuff you actually care about, the machine’s model of “you” is eight seconds more rage-bait. MediaSmarts — Canada’s digital-literacy organization, whose #ForYou card game for ages 13–18 is built on exactly this — frames it as a business question, not a mind-reading one: algorithms exist because your attention and your personal information are worth money to somebody.
One honest limit, since this page doesn’t do guesses: platforms publish the kinds of signals they use, but the full recipe — the exact weights, the exact model — is a trade secret at every major app. Anyone online claiming to know precisely “how the algorithm works” is selling something. What’s on this page is what the companies have said themselves.
Resetting a feed that’s gone somewhere bad
The friend from scenario 2 wants their old feed back. “It’s ruined,” they say. “Once it decides what you are, that’s it. The only fix is deleting the app.”
Are they right?
What’s actually going on
The feed is trained by behaviour, which means behaviour can retrain it — and the platforms have shipped real controls, because regulators and parents kept asking. The documented toolkit: “Not interested” (long-press a video) tells TikTok to back off that kind of content; TikTok added a full “Refresh your For You feed” button in 2023 that starts your recommendations over like a new account; Instagram later added a “reset suggested content” option of its own; and on YouTube you can pause or clear your watch history, which cuts off the main fuel line. None of these are hidden features — they’re in each app’s own settings and help pages.
The everyday version needs no buttons at all: spend attention like it’s money, because to the algorithm it is. Skip fast on what you don’t want more of — even skippable-but-fascinating garbage. Search, follow, and finish the things you do want more of. It feels slow, but it’s the same voting system that built the rabbit hole, pointed the other way. You’re not stuck with the version of you the machine guessed.
First-try answers. The score was never the point — the point is that the feed stopped being magic. It’s a machine with inputs, and you’re holding most of them.
The four things worth keeping
- Watch time is the loudest signal you send. Every second of lingering is a vote, whether you liked the video or not.
- The algorithm can’t tell “I love this” from “I can’t look away.” That confusion is where rabbit holes come from.
- Your feed reflects what holds you, not what you value. It’s a model of you, not a mirror.
- Feeds are retrainable. “Not interested,” feed refresh, watch-history controls — and spending attention only on what you want more of.
If the feed keeps serving you heavy stuff
- A feed that keeps circling body image, sadness, or self-harm content isn’t a personal failing — it’s the scenario-2 loop doing what it does. Use the reset tools above, and don’t carry the feelings part alone.
- Kids Help Phone — call 1-800-668-6868 or text CONNECT to 686868. Free, 24/7, confidential, anywhere in Canada. A real person.