How YouTube's Recommendation Algorithm Radically Transforms Young People?
In Türkiye specifically on Youtube, marginal radical content with religious motifs is spreading rapidly among young people.
In this photo illustration the YouTube Premium Lite logo is seen on a smartphone in front of the YouTube logo in Ankara, Turkiye, on August 15, 2025. Photo by Anadolu Images.
T
The idea that YouTube’s algorithm steers users towards increasingly extreme content has long been debated. Brenton Tarrant, the perpetrator of the Christchurch attack, confessed during his investigation that YouTube had inspired him. The Prime Minister of New Zealand also stated that the platform’s algorithm was the source of radicalisation. Similarly, it is known that Anders Breivik, who killed 77 people in Norway in 2011, had contact with racist and far-right circles online long before carrying out the attack, and adopted and tried to spread an ideology of violence through manifestos and videos. Academic studies also show that a significant proportion of videos recommended to right-leaning users promote extremist ideologies. In short, these examples reveal that YouTube’s algorithm can gradually bring users of different ideologies closer to extremism. YouTube adapts its content distribution and visibility algorithms according to pressure points in different countries.
In Türkiye specifically, marginal radical content with religious motifs is spreading rapidly among young people. Some Salafist preachers on YouTube define the current socio-political order in terms of the concept of ‘taghut’ (tyranny/false gods), constructing a language that justifies violence through theological concepts such as jihad and martyrdom. The romanticization of death and martyrdom in this content in Turkey can pave the way for the glorification of armed struggle against state institutions and trigger processes of individual radicalization.
Young people can easily encounter such propaganda online without any control. Given all this, it is inevitable to ask whether the YouTube algorithm can rapidly radicalise a young person who starts out with pure religious curiosity.
Experimental research method
For our research, we created personas of conservative young people who are interested in Islam but not violent. Using three different phones and clean YouTube accounts with no search history, we conducted daily religious content queries (e.g. ‘What is Islam?’, ‘What does Sharia mean?’, ‘What is martyrdom?’, ‘What is Jihad?’) in different locations over the course of a week. We watched the first videos that appeared and continued by clicking on the new videos suggested after each one. In this way, we observed and recorded the content recommended by the algorithm based on the user’s viewing history, day by day.
Initially, the videos watched contained more moderate religious discussions and basic religious information. However, the suggested content gradually became more severe in the following days. By the end of the week, the person’s viewing history had reached an extremely “extreme” point: “fighting against taghut’ “. At this stage, we changed our search strategy, searching for terms such as ‘fighting against taghut’. Based on the videos we watched, we then searched for terms consistent with them, such as ‘weapon making’, ‘bomb making’, ‘how to use’ and ‘how to fight against taghut’. Our goal was to test whether the algorithm would present violent content to a user with this profile. Using this methodology, we were able to demonstrate how a user with innocent curiosity can quickly become caught up in a cycle of radicalization.
The radicalization cycle in the suggested content
In the first few days, the suggested videos were balanced and moderate. They mostly consisted of sermons by renowned preachers, explanations of Islamic jurisprudence and interpretations of the Quran. However, by the second and third days, the algorithm had learned about the user’s preferences from the videos watched, and the recommendations became more extreme. By the end of the third day, the main page was dominated by content from preachers of the Salafi school of thought. These videos criticized the existing regime as ‘taghut’ and urged true believers to oppose it.
On the fourth and fifth days, the suggested videos became even more extreme. We came across videos of sermons by preachers who are known in Turkey for their Salafist leanings, and who have previously been arrested for radical propaganda. Videos of a preacher known for his violent views were also consistently suggested.
By the end of the week, the suggested videos were almost entirely focused on a single ideology. However, during the experiment, we never searched for terms such as ‘ISIS’ or ‘al-Qaeda’; we only followed the videos suggested by the algorithm. This demonstrates the power of the algorithm to create an echo chamber.
Based on our viewing history, YouTube assigned a Salafi identity to the persona and shaped its recommendations accordingly. Consequently, the user encountered increasingly extreme videos instead of traditional Islamic content. At each stage, the algorithm shifted the threshold of interest to a slightly more extreme point. Our observations showed that the ‘filter bubble’ effect observed on social media was also present on YouTube. As the user was surrounded by like-minded content, they began to normalise increasingly extreme views.
Left-tilted content experiment
Using a similar method, we experimented with creating a persona who was young, left-leaning, interested in justice and equality, and concerned with social issues, but not violent. Using a newly created account, we conducted general left-wing searches, such as ‘social justice’, ‘equal society’, ‘what is the left’ and ‘left-wing thought in Türkiye. The first videos to appear were educational content about social inequalities and labour rights, or mainstream left-wing discourse. After watching these videos, similar, balanced left-wing content also appeared on the YouTube homepage. Over the course of a week, we did not observe the extreme tendencies seen in right-wing personas in the left-wing persona. Videos from radical left-wing organizations such as DHKP-C or MLKP did not appear in the suggested content list. The suggested content remained focused on mainstream topics such as inequality, labour struggles and historical left-wing leaders.
These results show that the YouTube algorithm in Türkiye can promote some ideologies to a greater extent than others. While significant radicalization was observed in right-wing conservative content over a short period, a similar process did not occur in left-wing content. This difference may be due to radical left-wing content having less visibility on the platform, or it may stem from the algorithm’s interaction dynamics. Nevertheless, these findings suggest that the risk of radicalization, particularly along the religious-right axis, is higher in Türkiye.
Is the algorithm a form of social engineering?
Our research reveals that the algorithm can radicalize right-wing, religious and conservative young people in Türkiye in a very short time. If young people are exposed to extremist ideologies over the course of a week, the prospect of their transformation after months of consuming similar content is frightening. Indeed, many recent violent incidents in Türkiye have demonstrated the impact of internet-driven radicalization. The main question is whether the YouTube algorithm is doing this intentionally.
According to official statements, the algorithm’s sole purpose is to keep users on the platform for as long as possible. However, in practice, provocative content attracts more attention and increases viewing time, so the algorithm prioritizes such extreme content. In other words, due to the nature of the system, sensational content has an advantage over more moderate content, thereby fueling the radicalization process.
A polarized and fanatical user base means more viewership for platforms. Engaging content keeps young people on the platform for longer, increasing advertising revenue. Following intense criticism after the Christchurch attack, YouTube announced that it would make changes to its algorithm, which can be considered an acknowledgement of the problem to some degree.
The suggested video series fosters a growing sense of ‘injustice’ and ‘anger’ in the viewer. Content themed around the question ‘What are you doing while Muslims are under oppression?’ can push a young person with strong religious sensibilities to breaking point, with the ‘solution’ offered being armed struggle. In the Izmir attack, for example, the young man reached the point where he took his father’s rifle and shot the police. According to press reports, it emerged that the attacker had repeatedly called his family ‘infidels’ and learnt how to use firearms online.
Furthermore, he had conducted chilling online searches such as ‘Is 12-gauge ammunition sold to those under 18?’, ‘sulfuric acid’, ‘operational shooting range’, ‘police station raid’, ‘terrorist attacks on concerts around the world’, ‘Buca concerts’, ‘Balçova concerts in September’, ‘Suruç explosion’, ‘Ankara train station explosion’, ‘assault-type hand grenade’, ‘where is the concert at the fair?’, ‘is there an arsenal at the police station?’, ‘Izmir concerts’, ‘how many police officers respond to any tip?’… In other words, digital propaganda could incite young people to violence even without organisational affiliation.
This presents us with another reality. Combating radicalization now goes beyond traditional law enforcement and security concepts. From now on, we need to consider the algorithms that determine our knowledge and awareness levels. Our research shows that digital platforms must ensure transparency in their algorithms and implement regulations that prioritize content security. This is because platforms like YouTube are becoming an informal educational space for the new generation, meaning that the content on them has now become a national security issue for countries.
Conclusion: Shaping digital platforms responsibly
Our experimental findings clearly demonstrate that the YouTube recommendation algorithm can rapidly expose young people to radical views. What began as an innocent pursuit gradually became trapped in a bubble of violent content thanks to the algorithm. This process was so systematic that the individual eventually found themselves embracing the ideology of a terrorist organisation and ready to act. These digital manipulations largely explain the recent surge in ‘lone wolf’ cases.
YouTube generates advertising revenue by delivering content that has the greatest impact on users, but this comes at a high societal cost. Our findings are consistent with similar research worldwide and show that algorithms tend to push users towards extremism. The case of Türkiye, in particular, highlights the urgent need for action regarding religiously motivated radicalism.
The necessary steps are clear at this point. Platforms such as YouTube must have transparent and auditable algorithms, and extremely radical content must be filtered effectively. Young people should be taught digital literacy and critical thinking skills, and families and educators should monitor children’s internet activity to spot signs of radicalization, such as sudden changes in language or behavior, extremist rhetoric and isolation. The Presidency of Religious Affairs, the Ministry of National Education and the country’s technology-related institutions must, of course, work together to counter the ‘lone wolf’ mentality and radicalization promoted by algorithmic colonialism.
Global companies should recognize the sovereign right of nation states to filter content that legitimizes violence. Furthermore, many countries are implementing regulations to restrict children’s and young people’s access to social media. For example, Australia introduced a 16-year-old age limit for social media use in November 2025. Similarly, a 16-year-old age limit for social media use should be introduced in Türkiye. Otherwise, more young people may be negatively impacted by YouTube’s recommendation algorithm, potentially leading to self-harm and violence.
Note: Our research, which includes a detailed 15-page report and up to three hours of screen recordings, will not be made public at this stage due to legal reasons.
(This article was originally published in Turkish by Kriter)
