The True Face of TikTok’s Algorithm: Anatomy of an Attention-Capture System

tiktok algorithm: inside the system of control. The “For You” feed is not yours, it is not discovery, it is not digital serendipity—it is behavioral engineering applied at industrial scale, built through three precise mechanisms (behavior analysis, automatic categorization, cascading distribution) that construct your personalized prison while you call it freedom of choice.

One billion people consume information through this system, and understanding how it works is no longer optional but the only way to see the bars of the cage before they become invisible, before the capture mechanism becomes completely naturalized in your everyday perception of the digital.

How the Algorithm Works: The Three Pillars of the System

Three categories of signals—that’s all it takes to build a complete map of your attention, to track every movement of your interest, to statistically predict with high precision what will keep you glued to the screen in the next sixty seconds.

1. User Interactions: The Decisive Factor

Watch time, the single most important signal, outweighing all the others combined, and determining in an almost deterministic way the success or failure of every single piece of content published on the platform.

TikTok doesn’t just measure whether you watched a video until the end, it monitors whether you rewatched it in a loop, how many times, how quickly you scrolled away, where you hit pause, at which fraction of a second you hesitated before moving on to the next content—a video that maintains 80–90% completion gets an immediate algorithmic boost that can turn zero views into millions within hours.

Likes, comments, shares, saves
Accounts followed after watching a video
Videos marked as “not interested”
Speed at which you scroll away from a piece of content

Video completion is so central that creators design content specifically around it: short videos (15–30 seconds) that visually end exactly where they begin, creating automatic loops that the algorithm interprets as high engagement.

2. Video Information: TikTok SEO

The algorithm doesn’t “read” in the traditional sense—it sees through computer vision, it hears through natural language processing, it analyzes every single frame of the video with a precision that goes far beyond simply reading captions and hashtags, a technique that seemed advanced in 2018 but today looks primitive compared to the platform’s current capabilities.

Now the algorithm identifies objects, scenarios, people with a precision that instantly recognizes a kitchen from a park from a gym, analyzes facial expressions to categorize dominant emotions (anger, joy, shock, surprise, disgust), transcribes audio in real time to extract keywords even when they never appear in written text, and classifies the video into micro-thematic clusters you didn’t even know existed until you suddenly find yourself immersed in a hyper-specific niche the algorithm has built just for you.

This is TikTok SEO in its most advanced form, and in 2025, 20% of young people aged 18–24 use TikTok as their primary search engine for information—not Google, not Bing, but TikTok, where the keywords in captions, on-screen text, and even in transcribed audio determine whether your video appears in search results or disappears into the digital void without leaving a trace.

3. Testing Pool and Cascading Virality

You publish a video with zero followers and zero apparent relevance, and yet TikTok still shows it to 15–50 carefully selected people—a small initial group of users who have shown interest in related topics, users who are part of phase one of the process, what the industry calls “the test” but which is in fact a continuous experiment in behavioral optimization.

If the video surpasses a critical performance threshold (typically 80–90% watch time, a brutal metric that eliminates the vast majority of published content), then it moves into phase two, where it’s shown to hundreds of users, then phase three with thousands, and potentially millions in an exponential cascade that can turn an unknown account into a viral phenomenon within hours.

This mechanism creates the illusion of democratization, the feeling that “anyone can go viral,” which is technically true but hides a crucial structural detail: the algorithm doesn’t reward followers, it rewards watch time, and this difference is not semantic—it represents a complete paradigm shift compared to how traditional social media used to work.

It doesn’t matter who you are, it doesn’t matter what story you have to tell, it doesn’t matter how important your message is—what matters is only how well you can prevent people from scrolling away, how effectively you can capture and hold attention in an environment where every single video competes against millions of others for the exact same milliseconds of attention.

What the tiktok algorithm Really Rewards

In 2023, TikTok CEO Shou Zi Chew revealed a technical detail nobody was asking about but everyone needed to hear: when three users like the same video, the algorithm automatically categorizes them into a unique group and from that moment on begins showing them similar content, including videos other members of that group have already watched—building, in real time, small echo chambers where three people who reacted to the same stimulus are locked into a self-reinforcing bubble of related content.

The question then becomes: what exactly is this system optimizing for, what parameters is it trying to maximize through this process of clustering and segregating users into increasingly homogeneous groups—content quality, informational accuracy, value for the user?

The answer is simple and brutal: none of the above. The algorithm optimizes exclusively for time spent on the platform, for the total time you spend scrolling regardless of what you’re watching or how it makes you feel.

A video that makes you furiously angry has exactly the same algorithmic value as one that makes you laugh hysterically, as long as both keep you on the platform for the same number of seconds—the algorithm is completely agnostic to the content of your emotional experience, it only cares that you don’t leave, that you don’t close the app, that you keep scrolling to the next video and the next and the next, in an endless chain of stimuli engineered to be irresistible.

The practical outcome of this logic is mathematical and predictable: emotionally loaded content that short-circuits critical thinking and triggers instantaneous instinctive reactions systematically outperforms content that requires reflection, analysis, time to be understood—and this is not a bug in the system, not an unwanted side effect, but the expected and intentional functioning of an algorithm designed from the ground up to maximize engagement at any cost.

The Architecture of Filter Bubbles

You watch a video on a topic and the algorithm sharpens your profile; you watch more similar videos and the AI tightens the circle further; you interact, and the bubble narrows again—every single video you watch, every like you leave, every comment you post progressively shrinks the range of future content you’ll be shown, building a spiral of specialization that may look like smart personalization but is in reality algorithmic segregation.

This is not an echo chamber in the traditional sense, where you actively choose who to follow and what content to consume, but an algorithmically constructed echo chamber where every single interaction is interpreted by the system as explicit consent to narrow even further the range of what you’ll be shown, where diversity of perspectives is progressively eliminated not by conscious user choice but by automatic system optimization toward maximum possible engagement.

The academic research of 2025 documents this phenomenon through a systematic analysis of 30 peer-reviewed studies conducted between 2015 and 2025, covering multiple platforms and methodologies. From this meta-analysis, three patterns emerge that repeat with statistically significant consistency across all the studies examined, regardless of specific platform or method.

First pattern: algorithmic systems structurally amplify ideological homogeneity, reinforce selective exposure to content that confirms pre-existing views, and actively limit diversity of viewpoints—not because of malicious or intentional design, but simply because the system’s design is optimized for engagement, and engagement is maximized when users see content that confirms rather than challenges their existing beliefs.

Second pattern: young people display partial awareness of the phenomenon and develop some adaptive strategies to try to navigate algorithmic feeds more consciously, but their agency is structurally limited by opaque recommendation systems that do not provide sufficient tools to understand or modify the criteria by which content is selected—you can be fully aware that you’re in a cage and still not have the keys to get out.

Third pattern: echo chambers don’t just fuel ideological polarization as is commonly assumed; they also function as protected spaces for cultural identity reinforcement, where people find communities of similar individuals who validate their experiences and perspectives—bubbles are therefore not only a problem to be solved but also comfort zones where identity is continuously validated and reinforced through interaction with content and people that confirm who we think we are.

A specific 2025 study on TikTok users in Indonesia found that while some participants genuinely appreciated how the platform provided personalized and relevant information for their specific interests, many others felt that the filter bubble actively and problematically limited their view on important social and cultural issues—the subjective perception of the system’s usefulness therefore coexists with critical awareness of its structural limitations.

But there is a crucial detail emerging from the latest research that significantly complicates this apparently clear picture: a study published in PNAS in February 2025, using naturalistic experiments on YouTube, showed that the short-term polarization effects of filter bubbles may be more limited than previously assumed by prevailing theories, with heavy manipulations of recommendation feeds producing surprisingly small causal effects on participants’ political opinions over the study period.

This result does not mean that filter bubbles don’t exist or aren’t a real problem, but that the effects of these systems are likely more complex, nuanced and mediated by other factors than early theories suggested—the difference between a controlled six-week experiment and continuous everyday use over years may turn out to be enormous in terms of cumulative impact on opinion formation and reality perception.

Disinformation and Viral Speed

The very structure of the algorithm—designed to maximize speed of distribution and immediate emotional impact rather than accuracy and source verification—creates the perfect conditions for the rapid, uncontrolled spread of disinformation, not because ByteDance intentionally wants to promote fake news (there is no evidence of this), but because the system structurally rewards speed and emotion regardless of their relationship to factual reality.

The ultra-fast consumption of content on TikTok, where users scroll through dozens of videos per minute without stopping on any single one for more than a few seconds, makes the platform particularly vulnerable: false information can go viral and reach millions of people before professional fact-checkers can even identify it—let alone verify it and produce corrections that, in any case, would arrive far too late to contain the damage already done.

During the COVID-19 pandemic, systematic analysis of content published on the platform empirically documented that vaccine disinformation spread significantly faster than corrections from authoritative sources—much faster, with time gaps of days or even weeks between the publication of viral disinformation and the availability of accurate fact-checking.

The key distinction emerging from academic research is between misinformation (unintentional falsehood) which typically stems from ignorance, misunderstanding or honest error, and disinformation (intentional distortion) which is specifically and deliberately designed to influence public opinion by exploiting the platform’s engagement mechanisms—the latter is far more dangerous because it is optimized to maximize viral spread.

TikTok has introduced fact-checking systems and reporting tools that allow users to flag problematic content, but the issues persist structurally due to the platform’s viral nature, where the speed of diffusion always exceeds the speed of moderation—the video format also makes false information significantly more convincing and shareable compared to text-based platforms like Twitter, because psychologically it is easier to believe what you see in motion than what you read in static text.

The Impact on Traditional Media

In 2025, for the first time in the history of mass communication, social media overtook television as the main source of news in the United States—a shift that is neither temporary nor marginal but represents an irreversible structural change in how people access everyday information.

The data are unequivocal: 54% of Americans access news through social media and video networks, significantly surpassing both traditional TV stuck at 50% and news websites hovering at 48%—and for young people aged 18–24, this share rises above 50%, indicating that for an entire generation social media are not “also” a news source, but “the” primary channel through which they understand the world around them.

Newspapers are not simply losing ground in a competition they might still win with the right strategies—they are fighting on a completely redesigned playing field where the rules of the game are written by a Chinese algorithm optimized for capturing attention rather than distributing accurate information, and in this new environment the traditional skills of journalism (source verification, in-depth analysis, historical context) paradoxically become competitive disadvantages compared to the ability to produce emotionally charged content that generates instant watch time.

How Newspapers Are Adapting

Legacy outlets have had to completely reinvent their approach:

The New York Times has built a TikTok presence based on service content and human-interest stories rather than breaking news. They use journalists on camera who narrate stories with authenticity, adopting TikTok’s native language while maintaining brand quality. Their videos are short but dense with visual information, optimized to maximize completion rate.

Vice Media has adapted its on-the-ground reporting DNA to the TikTok format. Matthew Champion, Editor-in-Chief EMEA, explains their approach: “We have a correspondent on the ground walking through Kyiv with the sound of sirens in the background. They are living the experience but also telling you what’s happening.” Their war reporting from Ukraine hit 21.6 million views. The strategy: always start with a question to create a “curiosity gap” that instantly grabs attention.

Le Monde has experimented with live broadcasts on topics such as “the rise of manga in France,” and repackages content using metaphors, drawings, mock video games and acting.

BBC and CNN use their anchors as creators, with Victoria Derbyshire approaching half a million followers through quick news takes, and Jake Tapper promoting TV exclusives with behind-the-scenes content from his work.

The common strategy: broadcasters and mass-market tabloids repurpose existing content (footage, scoops), while newspapers and social-native brands create more bespoke and risky content.

The Structural Problem

The algorithm does not reward brand authority. Period.

An independent creator with 500 followers who packages information well can easily outperform The New York Times in views. Not because it’s better journalism, but because it’s better optimized for the algorithm.

Traffic from social to news websites—the traditional model of monetization via banners and subscriptions—is almost impossible to obtain. TikTok wants to keep users on the platform. Every external link is a failure for the algorithm. Every user who leaves is lost revenue.

The result: traditional media are not competing with other newspapers on TikTok. They are competing with an infinite variety of creators producing “infotainment”—content that looks informative but is optimized first and foremost for engagement, not accuracy.

Cyberpunk glitch illustration of TikTok algorithm, in neon digital style — Algorithm as invisible infrastructure: neon aesthetics, industrial logic of attention capture.

The Regulatory Front: Digital Services Act

The European Union has tried to impose transparency through the Digital Services Act, a legislative instrument that on paper should force digital platforms to be accountable for their operations and guarantee minimum protections for European users—TikTok and Meta have become the first primary targets of this new regulatory architecture.

On 24 October 2025, the European Commission issued what in legal jargon are called “preliminary findings,” a technical formulation which, translated into practical terms, means: “We have gathered enough evidence to state with reasonable certainty that you are breaking the law, and if we confirm these initial findings the consequences will be severe”—it is not yet a formal conviction, which would require a longer process, but it is significantly more than a simple warning and represents a concrete step toward potentially enormous sanctions.

The Specific Violations

First violation: Data Access for Researchers

Article 40 of the DSA is clear. Very Large Online Platforms must provide independent researchers with meaningful access to data in order to study systemic risks. Illegal content. Harmful content. Disinformation.

TikTok has made this process “excessively burdensome.” Bureaucratic translation: “We have put in so many obstacles that it is practically impossible to access the data.” The Commission’s preliminary findings explicitly use the term “burdensome.” Researchers end up with partial or unreliable data. They cannot properly study whether minors are being exposed to harmful content. They cannot analyze the mechanisms of disinformation spread.

ByteDance claims to have granted data access to nearly 1,000 research teams. The European Commission argues that access has been made so complicated as to be substantially useless. Both statements can be true at the same time.

Second violation: Ad Repository

TikTok does not maintain a transparent, freely accessible ad repository. The content of the ad is missing. The targeting data is missing. The identity of the payer is missing.

Why does this matter? Because without an ad archive it is impossible for citizens and researchers to identify disinformation campaigns, hybrid threats, coordinated influence operations—especially in electoral contexts. Advertising transparency is not a regulatory nice-to-have. It is the only way to detect manipulation at industrial scale.

Potential Consequences of the tiktok algorithm

If the preliminary findings are confirmed through the full investigation process that will unfold over the coming months, TikTok faces fines of up to 6% of its global annual turnover—and for a company of ByteDance’s size, this does not mean a few million dollars but potentially billions, numbers that even for a global tech giant constitute a serious economic deterrent, not just a marginal cost of doing business.

The DSA also requires structural changes to how the platform operates: users must be able to choose non-algorithmic feeds (chronological or based solely on followed accounts), recommendation systems must be explainable in sufficiently clear terms to allow users to understand why they are seeing certain content, and platforms must implement concrete, effective measures to mitigate the systemic risks created by their algorithms.

On 29 October 2025, the delegated act on data access officially came into force, an extension of the DSA that allows qualified and vetted researchers to access not only public data but also certain non-public datasets maintained by platforms for internal analytics—on paper this represents a significant leap forward in terms of transparency. In practice, we will see how accessible these data actually become, and what technical or bureaucratic obstacles platforms will erect to limit their real-world usefulness.

Tiktok algorithm and tLimits of Regulation

Giving users the technical possibility to switch to a chronological feed is like offering raw vegetables to someone who has developed an addiction to refined sugar—technically it’s an option on the menu of possible choices, but in practice how many people will actually choose it when the alternative has been optimized through thousands of iterations to be irresistible?

The system has been deliberately built so that the default choice (the hyper-personalized algorithmic feed) is so superior in terms of immediate gratification that choosing the alternative requires a level of discipline and awareness that the overwhelming majority of users simply do not possess or are not willing to exercise continuously—thousands of A/B tests, millions of live experiments on real users, all fine-tuned to maximize what the industry calls “zero friction,” which in practical terms means eliminating every possible obstacle between the impulse to open the app and the gratification of seeing the perfect piece of content for you at that precise moment.

Even if ByteDance were to publish the algorithm’s full source code on GitHub tomorrow, making it accessible to anyone who wanted to study it, most users would lack both the technical skills (we’re talking advanced machine learning, neural networks, multimodal optimization) and the time required to analyze it (we would likely be dealing with millions of lines of code spread across hundreds of repositories)—the system’s real opacity is therefore not technical but structural, it does not stem from a will to hide information but from the intrinsically complex and ever-evolving nature of modern machine learning.

The system is based on machine-learning algorithms that continuously self-adjust by learning from data, which means that not even its original designers can fully predict how it will behave in every possible situation or input combination—it is not a black box because ByteDance deliberately wants to keep trade secrets, but a black box by the very nature of modern machine learning, which produces systems too complex to be fully understood even by those who built them.

The Pressure on Creators

Creators are simultaneously the greatest beneficiaries and the most obvious victims of this system—the algorithm theoretically offers everyone the chance to reach the For You Page regardless of follower count: zero followers are not an absolute barrier, the right video at the right moment can take you from complete obscurity to millions of views in a matter of hours.

But this democratic potential comes at a cost that becomes increasingly clear as creators try to turn occasional visibility into a sustainable career: constant, obsessive posting, often multiple times a day, every day, without breaks, because the algorithm rewards frequency in ways that feel almost punitive—if you disappear from the platform for a week because you need a break or simply because life presents other priorities, your reach does not decline gradually, it collapses dramatically, as if the algorithm were actively punishing you for daring to interrupt the content flow.

Creators speak openly about high levels of burnout, and academic research confirms it through qualitative and quantitative data—it is not just physical or mental exhaustion but a systematic erosion of the capacity to remain creative under constant performance pressure, where every video is a test that can make the difference between continued growth and sliding into irrelevance.

Creative standardization is another inevitable side effect: when you see a video perform exceptionally well, reach millions of people and generate thousands of interactions, the rational temptation is to make another similar one; when that also works, you make a third; and after a hundred iterations of this process you find yourself producing infinitesimal variations of the same piece of content because the algorithm has empirically demonstrated that this specific format generates watch time—creativity becomes optimization, originality becomes an economic risk that few can afford.

The obsession with the opening hook is perhaps the most visible aspect of this pressure: the first three seconds of each video literally decide everything, determining whether the algorithm will continue to show that content or mark it as a failure—creators therefore focus maniacally on the initial hook, the editing rhythm of the first frames, the immediate emotional impact, often at the expense of depth, context and complexity that would require time to be properly developed, not because creators are incapable of producing deep content but because the algorithm structurally does not reward depth that takes time to be grasped.

The winning strategy in 2025 according to researchers and successful creators:

Hook in the first 3 seconds. Provocative questions. Fast cuts. Familiar audio. The user must stop. You don’t get a second chance.
TikTok SEO. Keywords in captions, on-screen text, subtitles. The AI reads everything to categorize. If you’re not optimized for search, you don’t exist.
3–5 targeted hashtags. A mix of broad (#TikTokTrend, #Viral) and niche (#BookTok, #FitnessTips). No more. No less.
Trending sounds. Even at zero volume. It signals to the algorithm that the content is fresh, current, part of the ongoing conversation.
Engagement in the first 60 minutes. The algorithm gives maximum weight to early interactions. If the video doesn’t take off in the first hour, it probably never will.

Specialization in micro-niches. This has become essential. The algorithm excels at connecting users to extremely specific interests. Not “sports” but “restoring old rally cars.” Not “food” but “under-10-minute Japanese vegan recipes.”

Creators who focus on specific niches build more loyal audiences. Higher engagement. Less competition. More value for the algorithm.

Cultural Fragmentation and the Atomization of Experience

The algorithm creates ever tighter and more impermeable individual bubbles, generating as a direct consequence a level of cultural fragmentation unprecedented in the history of mass media—there are no longer truly global trends in the classic sense where everyone was exposed to the same content at the same time and could discuss it as a shared experience; instead, there are micro-trends that last at most a few days before being replaced, trends visible only within specific user clusters carefully segregated by the algorithm.

What is viral and omnipresent to you right now may be completely invisible to your sister sitting next to you on the couch, living inside a totally different algorithmic bubble despite sharing the same physical space—this is not a minor technical detail but represents the structural collapse of shared experience that for decades provided the common ground necessary for public discourse.

The 2025 data document this migration with a brutality that leaves little room for optimistic interpretation: 43% of Gen Z prefer YouTube and TikTok over traditional TV and streaming services, not as a marginal or temporary choice but as a structural preference that defines their media consumption patterns—we’re not talking about a slim majority but almost half of an entire generation that has completely abandoned traditional formats.

Average daily time spent on streaming video will rise to 4 hours and 8 minutes, while time spent on traditional TV will simultaneously collapse to 1 hour and 17 minutes—these are not gradual adjustments or seasonal fluctuations but a structural breakdown in which the old paradigm is entirely replaced by a new one within a historically short time frame.

Streaming revenues will grow 18–19% annually, while traditional TV revenues will fall 4–6% year-on-year—the industry is not simply pivoting toward new formats while essentially preserving the same economic structures, it is migrating entirely to a radically different economic model where past skills and assets rapidly become irrelevant.

This fragmentation perfectly serves the platform’s economic interests, because it eliminates the very notion of collective saturation—there can never be a moment when “everyone has already seen everything,” because each user lives in a separate universe with content that is always new for them, even if already seen by others; there is always another niche for the individual user to explore, another algorithmic rabbit hole to fall into, and the algorithm can keep optimizing forever because every path is unique and potentially infinite.

But the social cost of this fragmentation is devastating in ways we are only beginning to understand: when there is no longer any shared experience among people living in the same society, when each individual lives inside a fully personalized and impermeable information bubble, the very possibility of public discourse—which requires at least a minimum of common ground on which to build debate—collapses structurally before we even realize we have lost it.

Surveillance Capitalism in Action

ByteDance does not sell your data in the traditional sense. It doesn’t need to. The real product is something far more valuable: the ability to shape behavior in real time.

Every video is a test. Every interaction is a data point. Every session is an experiment.

TikTok’s AI creates a hyper-granular “interest profile” that has nothing to do with traditional demographic categories. You don’t just like “pop music.” You like “Korean pop songs with a certain tempo and specific vocal features while you’re on the move between 6 p.m. and 8 p.m. and have scrolled away from at least three videos in the previous 60 seconds.”

The more accurate the system becomes, the more intense the risk of bubbles. The algorithm does not seek diversity. It seeks optimization of time on platform. The difference is vast.

Internal TikTok data: the 60-minute time-limit prompt reduces usage by about 1.5 minutes. Not 58 minutes. Not 30 minutes. Just 1.5 minutes. The system is engineered to be resistant to individual self-control interventions. This is not a design accident. It is the design itself.

The Concrete Numbers of 2025

Numbers that tell a story. The story of an industry that is not changing. It is migrating.

30% of the global population uses social media as their main news source (Reuters Institute Digital News Report 2025). This is not a niche. It is one-third of connected humanity.

Over 50% of 18–24-year-olds use social media for news. For this generation, social platforms are not “also” a news source. They are “the” news source.

20% of 18–24-year-olds use TikTok specifically for news. A steady 2% annual growth. The trajectory is clear. And irreversible.

955 million monthly active users in 2025. Projection: 1.9 billion by 2029. Nearly double in four years.

Average time on platform: 55.8 minutes per day. More than double the 27.4 minutes of 2019. In six years, time spent has doubled. Growth is not slowing.

6.4% of university users may meet criteria for TikTok addiction (systematic review 2025). This isn’t alarmism. It’s epidemiology.

ROI underestimated by 10.7x. When measured with advanced models (Bayesian MMM) instead of last-click attribution, TikTok’s ROI turns out to be 10.7 times higher. Companies that rely on obsolete measurement models are underestimating impact by an order of magnitude.

What This Means for Information

Independent creators and “internet personalities” are the primary news sources on TikTok—far ahead of traditional journalists and legacy publications. This is not a temporary anomaly. It is the new normal.

This reflects a structural distrust of mainstream media. But it also creates a situation where news is filtered through the personal bias of individual influencers. The process is less transparent than in legacy outlets—even taking into account all the documented problems of the latter.

When traditional media fail, at least the mechanisms of that failure are visible. Newsrooms. Editorial boards. Declared editorial lines. Transparent policies. When an influencer distorts information for engagement, the mechanism is invisible—and personalized for each individual user.

There is no oversight. No accountability. No professional standards. There is only the algorithm rewarding what works, regardless of accuracy or integrity.

No Simple Solution

There is no quick fix that will solve this problem with a software update or policy tweak, no magic button that will suddenly transform TikTok into a space of healthy cultural exchange and accurate information—the problem doesn’t lie in a technical bug that can be identified and patched, but in the fundamental design of the system itself, in the economic logic that sustains it, in the algorithmic architecture that defines it.

You can uninstall the app from your phone as an individual act of resistance, you can consciously choose digital abstinence and live outside this attention economy—but that personal choice does nothing to change the structural fact that hundreds of millions of people (and especially younger generations currently forming their understanding of the world) continue to consume information through an algorithmic filter deliberately engineered to maximize engagement rather than understanding, time on platform rather than real learning.

These people are, in real time, developing cognitive attention models calibrated to 60-second videos that change every few seconds, with a dramatically reduced ability to sustain the prolonged concentration required for complex analysis, long texts, articulated arguments—they are literally learning to communicate not with the goal of being deeply understood by others but with the goal of “hacking the feed,” generating measurable watch time, triggering the recommendation mechanisms that distribute visibility.

European regulation via the Digital Services Act is certainly a necessary step toward greater platform accountability, but it is also structurally insufficient relative to the scale and complexity of the problem—even assuming fines are actually enforced, controls are genuinely implemented, and platforms are forced into greater transparency, the fundamental economic logic remains intact: attention capitalism, in which users’ conscious time is the scarce resource to be systematically extracted and monetized through advertising and data collection.

TikTok has built what is arguably the most efficient machine ever created in human history for converting behavioral neuroscience and addiction psychology research into corporate profit at global scale—it has turned decades of academic addiction science into executable code optimized to maximize retention, it has fully industrialized the capture of human attention, transforming it from an approximate art into an exact science.

TikTok’s algorithm is neither unique nor an anomaly in today’s digital landscape—it is simply the most sophisticated and technically advanced version of a much broader logic that is transforming the entire digital ecosystem, where any platform that chooses to optimize primarily for user engagement rather than their psychological and social wellbeing is essentially walking the same path. TikTok has just walked it faster and is therefore further along what looks increasingly like an inevitable trajectory.

The fundamental question is not whether TikTok is intrinsically “bad” in some absolute moral sense—that question distracts from the real structural problem—but whether we, as a society, are truly willing to accept a future in which access to information, opinion formation, and the very construction of shared reality that makes democratic discourse possible are all mediated by algorithmic systems designed not to enlighten but exclusively to retain, not to educate but to capture, not to liberate but to chain attention into infinite loops of deferred gratification.

The algorithm continues to function with perfect efficiency precisely because, for the average user, it does not look like a control system at all—it looks exactly like its opposite: personalized freedom of choice, algorithmic discovery of perfectly calibrated content tailored to your interests, a system that gives you exactly what you want moment by moment without you even having to articulate what you’re looking for.

And in the meantime, as this illusion of freedom solidifies into normalized everyday experience, video after video and second after second, the algorithm is silently but systematically rewriting the fundamental parameters through which you perceive external reality, redefining what you consider normal in terms of information-consumption speed, progressively reducing the maximum span of your attention before your brain automatically demands the next dose of algorithmic stimulation.

There is no final epiphany that will magically resolve the problem, no collective awakening we can expect to arrive spontaneously and save the situation—there is only this question that remains suspended without a definitive answer: who really decides which thoughts deserve your time and your limited cognitive attention? Is it you, through conscious and deliberate choices, or is it the algorithm, specifically designed and optimized through millions of iterations to prevent you from stopping scrolling even when you would rather do something else?

neural network: the Probabilistic Regime of LLm and ai

engagement social: media power and the rule of attention

algorithms decoded: the dream of rationality

THE POWER OF DATA: AI, SOCIAL AND DIGITAL GOVERNANCE