The algorithm isn't neutral. How harmful content finds teens.
Recommendation systems don't decide content is harmful or beneficial. They optimise for watch time. When a distressed teenager engages with content about distress, the system does what it was built to do — and shows them more.
US/Canada: 988. UK: Samaritans 116 123. Australia: Lifeline 13 11 14. If your child has expressed thoughts of suicide or you've found evidence of self-harm, act now. Ask directly — asking about suicide does not increase risk, and clinicians are unanimous on this point.
What the mechanism actually is
This isn't a conspiracy and it doesn't require anyone to intend harm. The sequence is mundane:
- A teenager engages with content about sadness, isolation, or feeling different. Ordinary adolescent experience, and there's a lot of it.
- The system registers engagement — watch time, rewatches, saves.
- It shows more from that cluster, because that's the objective function.
- Distressed engagement is intense engagement. The signal is strong.
- Within a small number of sessions the feed has reorganised around it.
Independent research and regulatory investigations have repeatedly found that new accounts registered as minors, engaging minimally with mental health content, receive self-harm-adjacent recommendations quickly. The exact figures vary by study and platform. The direction of the finding does not.
Why this is worse than what came before
- No search required. Previously a child had to look for it. Now it arrives.
- It targets the vulnerable specifically. The system identifies distressed engagement and serves accordingly — meaning the teenagers most at risk get the most of it.
- Volume. Not one article. Hundreds of short videos in a sitting.
- Community reinforcement. The content connects to communities that normalise the behaviour, presented as understanding.
- It's invisible to parents. Personalised feeds mean your child's version of the app looks nothing like yours. Checking their phone shows you their feed only if you scroll it as them.
Reduce the exposure window
Defenras blocks self-harm content domains and lets you schedule when feed-based apps are available at all.
Get Defenras free →What you can do at the platform level
Turn off watch history
The most effective single change. Without history the recommendation engine loses its targeting.
- TikTok: Settings → Privacy → Personalisation and data → turn off personalised feed where available; clear watch history.
- YouTube: profile → Your data in YouTube → turn off watch and search history.
- Instagram: Settings → Content preferences → Sensitive Content Control → Less.
Actively retrain the feed
"Not interested" on relevant content works, and it works faster than parents expect. Doing it together with your child is more effective than doing it for them — it teaches them the feed is something they can control.
Use the teen accounts where they exist
Several platforms have introduced restricted account types for minors with tighter content defaults. They're worth enabling and they're not sufficient on their own.
What you can do at the device level
- Block self-harm content domains as a category, covering the standalone sites platform moderation doesn't reach.
- Schedule feed-based apps. Late-night solo scrolling is when this content does the most damage — low mood, no interruption, no one to talk to. Making these apps unavailable after a set time is one of the higher-impact things available to you.
- Keep phones out of bedrooms overnight. Blunt, unpopular, and consistently associated with better sleep and lower exposure.
The conversation that matters more
Filtering reduces exposure. It doesn't address why the content resonated.
If your child is engaging heavily with this material, the content is a symptom. The underlying question — what's making them feel this way — is the one that needs an answer, and it usually needs a professional.
Things that help when you talk:
- Ask directly. "Have you been feeling like hurting yourself?" Directness is protective; ambiguity leaves them guessing whether you can handle the answer.
- Don't react with panic. Visible alarm teaches them the topic is too much for you and closes the door.
- Don't remove the phone as the first response. It cuts them off from support and confirms that telling you has a cost.
- Get professional help early, not once it's severe enough to feel undeniable.
It's worth being clear that this is a design outcome, not a parenting failure. These systems are built by large teams to maximise engagement, tested continuously, and they are extremely good at it. No individual parent is going to out-engineer that with settings.
What you can do is reduce the hours of exposure, keep the worst of it blocked, and be the person your child talks to. That combination is genuinely effective — and it's a fairer description of what's available to you than any claim that the right app configuration solves this.
Frequently asked questions
How quickly can a feed turn harmful?
Research on new minor accounts has repeatedly found self-harm-adjacent recommendations appearing within a small number of sessions. Figures vary by study; the pattern is consistent.
Does turning off watch history actually help?
Yes, substantially. It's the single most effective platform-level change available, because it removes the targeting signal the recommendation engine depends on.
Can I see what my child's feed shows them?
Only by scrolling it as them, on their device, logged in as them. Feeds are personalised, so your version tells you nothing about theirs.
Is blocking social media the answer?
It reduces exposure and doesn't address the distress. If your child is engaging heavily with self-harm content, that's a mental health situation needing professional support, with filtering as one supporting measure.
Defenras blocks all of this — without collecting your child's data.
One app, every device, every browser. Adult content, gambling, vaping, scams, malware, social media — all blocked. PIN-locked. Free version available, no credit card.