Study: Young people try to train social media algorithms, but algorithms train back

A new international study has found that young adults actively attempt to shape their social media feeds to protect their wellbeing, but algorithmic systems frequently override their preferences. Conducted across Estonia, Finland, Austria, and the UK, the research reveals that user control over content feeds is often illusory. Researchers are calling for platforms to be held more accountable rather than placing the burden on individuals.

Young people aged 18-30 are not passive consumers of social media, they actively strategise, click "not interested," search specific keywords, and deliberately scroll past content they don't want to see. Yet a new international study has found that these efforts are routinely outpaced by the very algorithms they're trying to influence.

The research behind the findings

The findings come from the TRAVIS project (Trust and Visuality in Everyday Digital Practices), an international research initiative that interviewed 125 young adults across Estonia, Finland, Austria, and the United Kingdom. The study, whose results are being published in the academic journal Information, Communication & Society, was co-authored by Jaana Davidjants and Katrin Tiidenberg.

Participants described deliberate attempts to "train" algorithms on platforms including Instagram, TikTok, YouTube, and Facebook, seeking more content that supported their wellbeing and less that caused anxiety, body dissatisfaction, or guilt. These judgements were deeply personal: what feels harmful or helpful varies significantly from person to person.

Algorithms remember, users forget

One recurring theme in the interviews was the long "memory" of algorithms. Participants noted that changing a profile built up in the early days of account creation felt almost impossible, even when their interests evolved over time, platforms continued serving content reflecting earlier behaviour.

Users also described a pattern of algorithmic over-reaction. Watching just a few fitness videos in search of inspiration, for instance, could trigger a relentless stream of fitness content. The researchers note that the issue is not just what content appears, but how often, several motivating workout videos a day can quickly tip into anxiety-inducing pressure.

The opposite problem also emerged: algorithmic under-reaction, or sluggishness when users tried to reset their feeds. Pressing "not interested" repeatedly with no visible result left many participants feeling that real control lay with the platform, not with them.

The gap between intention and behaviour

Perhaps the most striking finding concerns the gap between conscious intent and unconscious signals. A young person may wish to avoid content related to eating disorders, for example, but pausing even one second longer on such a video sends a signal that overrides any deliberate counter-action. The algorithm interprets hesitation as interest, and responds accordingly.

Some participants also described attempting to improve the experience not just for themselves but for others, reporting accounts or posts to make the platform environment safer for the wider community. This, the authors suggest, reflects a broader cultural shift in which personal wellbeing is increasingly treated as an individual responsibility.

Bans are not the answer

The study enters an active policy debate. Australia introduced a social media ban for under-16s at the end of 2024. The UK government has announced similar plans, with restrictions expected to come into force in spring 2027. France, Denmark, Spain, and Germany are all discussing comparable measures. Estonia has not yet adopted an outright ban, though debate around young people's digital wellbeing is ongoing.

The researchers caution against placing too much faith in age-based restrictions. In Australia, an estimated 60-70% of children and young people continue to use social media despite the ban. Those unable to circumvent the restrictions may miss out on important digital skills, potentially widening the digital divide.

What should change

The study's authors argue that the policy conversation should shift away from digital literacy programmes and age bans and toward platform accountability. Platforms, they say, should be required to offer less intrusive recommendation systems that do not progressively narrow and intensify a user's feed based on every click and view.

The EU's Digital Services Act already prohibits deceptive design practices, so-called dark patterns. The researchers say enforcement of existing rules and stronger obligations on platforms should be the priority. As participants in the study put it themselves: «we don't want smart algorithms, please make them dumber!»