You have a short video application you open to have a few minutes of rest. And before you know it an hour has passed.
The videos are eerily realistic. One video leads to the next and before you know it you are in some hole of the internet you never realized was there. It is intimate–intimate.
The same thing applies to streaming sites. You complete a show and another one is on the list. The home page appears to be aware of your mood even before you are. We are inclined to refer to this as convenience. The app is simply something that gets us.
But what would happen when the app is not only reading your preferences but also creating them?
This blog believes that AI recommendation algorithms are accomplishing a two-fold goal concurrently. They truly simplify and make our lives on the internet more enjoyable. Yet they also silently influence what we desire, what we observe and what begins to seem normal to us. And you can now see governments all over the world, scrambling to keep pace with the implications.
Taking TikTok as the prime example, this article unravels the working principle behind the recommendation systems, why it is even stronger than it might seem, and what the policymakers are (and aren’t) doing with it.
Fig. 1. Someone scrolling through a TikTok feed. Source: https://unsplash.com/photos/someone-is-scrolling-through-a-tiktok-feed-XthsrkTNZ6I
“It just knows me” — why recommendation systems feel so helpful
Recommendation systems address this issue. They operate as described by Just and Latzer (2017) by automatically assigning relevance to content i.e., in effect pre-sorting a massive catalogue in such a way that you do not have to look at everything, only the stuff that is most likely to be of interest to you. That is why your TikTok For You page feels personalized, and why Netflix suggests movies on its home page that seem to fit your preferences, your mood, even your recent viewing activity.
This is a real convenience. Recommendation systems save time in search, assist you in finding something you would have never encountered otherwise, and make online media feel natural and comfortable. According to Just and Latzer (2017), the recommender systems are now among the most significant types of algorithmic selection in the consumption of media – especially in the entertainment of music and videos.
Working, another thing is, is trust. Similar to the masses in terms of search engines, the majority of people regard the algorithmic systems as neutralized systems that simply offer the best, or the most relevant, solutions. This sense of neutrality is significant as it enables the personalisation to appear more like a service than a commercial intervention. In the case of a platform that seems to be familiar with you, its recommendations are not so much marketing as it is a helpful hand.
Yes, yes, the recommendation systems are helpful to the users and it would be unfair to write that off. What has a greater value is what they are doing in the meantime and whose interests are optimised in the process in the first place (Just and Latzer, 2017; Noble, 2018).
However, this is what is actually occurring, the algorithm is not only reading your taste but also forming it.
It is this where it becomes awkward.
Recommendation systems have been said to be almost like reading our preferences and presenting it back to us. However, when a platform chooses what is and is not visible, what is or is not ranked first, buried, or not, it ceases to be a neutral mirror. It begins to behave more as a curator, one with its own commercial agenda.
In order to survive in a world filled with content, Andrejevic (2019) writes, social work of organising culture has been transferred to automated systems without much noise. And such systems are commercially owned. Not only are platforms making it easier to find content, but they are also becoming a bigger and bigger part of the cultural world you are interacting with to begin with.
This is important since what is prescribed is not done with the best interest of you as an individual. It is selected according to what retains your stay on the platform. The metrics that recommendation systems are optimised to are engagement, watch time, retention and return visits (Crawford, 2021). A programme that is geared towards long-term wellbeing may at times recommend something new or difficult. An engagement-optimised system will continue to show you more of what you already stopped to view, rewatched, or browsed.
Another issue that Pasquale (2015) describes is the fact that the reasoning behind recommendation rankings is mostly unknown to the users. You watch the result, the resulting video, the show that it recommends, the list of shows it thinks you might like, but you never get to see the mechanism. Social networks amass vast quantities of behavioural data on you, and the judgments that make decisions on which you are shown are opaque. It is a great imbalance of power.
According to Crawford (2021) there is a simple answer to this question: it is not whether AI systems work, but what they are optimising, who they benefit, and who has the final decision to make. The answers matter, because they reveal that recommendation is not a neutral technical process — it is also a political and economic one (Andrejevic, 2019; Crawford, 2021; Pasquale, 2015).
TikTok: a case study in how curiosity becomes a habit you didn’t choose
One of the most vivid illustrations of the practicality of recommendation systems and the extent to which they impact individuals is TikTok, which, in fact, is more influential than many individuals think.
You do not search anything when you open Tik Tok. You are instantly put into an Into You feed, a never-ending video stream that you already have been algorithmically chosen by the platform before you have made a single conscious decision (TikTok, 2020). According to TikTok, For You feed ranks the videos based on signals such as likes or shares, accounts followed, comments, and video information (such as captions, sounds, and hashtags), among others, but the device settings have a lower influence (TikTok, 2020).
The response time is abnormally quick. You do not have to actively select a preference of the platform in order to begin building one. Stopping a video on skincare, re-watching the clip about a sad break up, or even watching a cooking tutorial to the end all these actions are all counted as cues. Incidental, non-reflective behaviour is data. And the information rapidly creates a portrait that informs your next hour of watching. What starts as a mere passing interest, may turn into a flow of similarities in a surprisingly short period of time.
TikTok does not just observe what you like. It helps turn scattered moments of attention into increasingly fixed habits — and it does this before you have even decided what you want.
To its credit, Tik Tok has recognized this dynamic. The company itself, in its explanation of the For You feed, acknowledges that its system can generate a more homogeneous viewing experience, which it specifically refers to as a possible potential filter bubble (TikTok, 2020). In 2023, it launched an option to refresh their feed and essentially start fresh (Grover and Wang, 2023). It is an informative confession: the system of recommendations can be so deep-rooted that its users will have to take steps to evade it.
This is precisely what Crawford (2021) cautions against: it is not that algorithms occasionally misjudge our tastes. The reason is that the usefulness of a recommendation system cannot be entirely independent of the business interests of the platform on which it operates. The situation with TikTok cannot be ignored (Andrejevic, 2019; Crawford, 2021; Pasquale, 2015).
Fig. 2. TikTok’s rise and the role of its algorithm. Source: https://www.theguardian.com/technology/2022/oct/23/tiktok-rise-algorithm-popularity
So who’s watching the algorithm? The policy gap
When this level of behaviour is being influenced by the recommendation systems, the question that arises is what the governments are doing about it. The reason remains: not enough, and not fast enough.
The European Union has come up with the most ambitious regulatory response to date. On 17 February 2024, the Digital Services Act became generally applicable and the European Commission may impose up to 6% of the global annual turnover of major platforms in case of serious violations. The DSA also necessitates more transparency and risk evaluation regarding recommender systems and online harms, of very large platforms (European Commission, 2026).
Australia has a less extensive regulation. The Online Safety Act 2021 enables the eSafety Commissioner to regulate some types of harmful online content, whereas the Safety by Design initiative promotes safety in designing products. However, this framework still fails to explicitly govern recommendation systems as systems that scale content, and much of the design guidance is still voluntary (eSafety Commissioner, 2025).
In America, there is definite increase in public concern but policy response is disjointed. The problem of recommender systems design logic has received less attention than child safety and platform responsibility more generally, but federal attention has been drawn to the issue through Senate hearings on social media and teen mental health (United States Senate Committee on the Judiciary, 2023).
It is startling how the magnitude of the problem and the aspirations of the regulatory response are out of proportion. According to Crawford (2021), meaningful accountability entails inquiring about what the AI systems are optimising, who they optimise on and who decides. At this point, that power is still to a significant degree in the control of the platforms themselves.
Why this is important outside your screen.
It is easy to envision that recommendation algorithms are an intimate issue – your feed, your screen time, your habits.
But there is still more than that. They are not just dictating personal preferences by platforms determining what is shown and what is sunk. They are shaping concepts being propagated, seemingly mainstream worldviews, and are heard. Social work is increasingly being socialised, organised by recommendation systems, says Andrejevic (2019) – and in a way that is driven by commercial, instead of social, interests.
This, as Pasquale (2015) puts it, is a one-way mirror: platforms build detailed profiles of their users and obscure their ranking and recommendation logic. Users need to trust a system, which they cannot substantially verify or challenge. The less transparent this process is the more one is able to disorient platform curation and personal taste.
Confusion between what the algorithm displays and what we tend to like is a puzzle that is not a by-product of recommendation systems. To a large degree, it is their greatest strength (Andrejevic, 2019; Crawford, 2021; Pasquale, 2015).
The feed is not neutral – and it is time we should treat it that.
AI recommendation algorithms are beneficial indeed. They speed up, streamline and even improve content discovery. And that is a truth, and it will be observed.
However, there is more than just convenience. These systems are not merely responsive to what we desire, with time, they assist in creating it. Ranking visibility, turning behaviour into data, and constantly optimising to engage rather than to benefit us, recommendation algorithms are engaged in the silent process of influencing our habits, tastes, and attention in a manner that a majority of us do not even notice.
Even the implicit recognition by TikTok that it has filter bubbles and the option to refresh the feed already demonstrates that recommendation systems are capable of influencing the user experience in a manner that is hard to control.
The point is to remember that the feed is not neutral. All the videos that show up on the top of your screen represent commercial, technical and political decisions – and largely unknown to you.
What can you do? Begin by realising that what you are presented with is not what is there. Actively attempt to find non-feed content. And enhance more rigorous regulation of algorithmic accountability – the Digital Services Act by the EU is a good beginning, but other countries such as Australia should do more. Recommendation systems working on this scale and with such influence cannot afford voluntary codes of practice or even narrow online safety laws.
The algorithm is aware of much about you. We need to learn more about it.
References
Andrejevic, M. (2019). Automated Media. Routledge. https://doi.org/10.4324/9780429242595
Crawford, K. (2021). Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. Yale University Press.
eSafety Commissioner. (2025). Industry regulation | eSafety Commissioner. ESafety Commissioner. https://www.esafety.gov.au/about-us/industry-regulation
European Commission. (2026). The Enforcement Framework under the Digital Services Act | Shaping Europe’s Digital Future. Digital-Strategy.ec.europa.eu. https://digital-strategy.ec.europa.eu/en/policies/dsa-enforcement
Grover, S., & Wang, M. (2023). Introducing a way to refresh your For You feed on TikTok – Newsroom | TikTok. Newsroom | TikTok. https://newsroom.tiktok.com/introducing-a-way-to-refresh-your-for-you-feed-on-tiktok-au?utm_source=chatgpt.com&lang=en-AU
Hern, A. (2022, October 24). How TikTok’s algorithm made it a success: “It pushes the boundaries.” The Guardian. https://www.theguardian.com/technology/2022/oct/23/tiktok-rise-algorithm-popularity
Just, N., & Latzer, M. (2016). Governance by algorithms: reality construction by algorithmic selection on the Internet. Media, Culture & Society, 39(2), 238–258.
Noble, S. U. (2018). Algorithms of Oppression: How Search Engines Reinforce Racism. New York University Press.
Pasquale, F. (2015). The Black Box Society The Secret Algorithms That Control Money and Information. Harvard University Press.
Swello. (2025, April 30). Someone is scrolling through a tiktok feed. Unsplash.com; Unsplash. https://unsplash.com/photos/someone-is-scrolling-through-a-tiktok-feed-XthsrkTNZ6I
TikTok. (2020). How TikTok recommends videos #ForYou – Newsroom | TikTok. Newsroom | TikTok. https://newsroom.tiktok.com/how-tiktok-recommends-videos-for-you?lang=en
Be the first to comment