ISTANBUL – She had already bought the dress. Three days later, it was still chasing her across every app she opened – in her Instagram Explore tab, in a TikTok ad, tucked between YouTube recommendations. The purchase was done; the algorithm had not gotten the memo. For this user, a 29-year-old participant in a new Turkish consumer study, that loop was the moment she stopped believing the platform understood her at all.
It is a complaint so common it has become a digital-age idiom. Yet a study released this month by Turkish research firm Twentify puts numbers and emotional texture around what has long been dismissed as anecdote. Titled “Digital Mirrors: Do They Really Know Us?” the qualitative study drew on in-depth interviews with 32 active users aged 18 to 54, spanning social platforms, e-commerce sites, and music and video streaming services. What it found is that the gap between what algorithms claim to know and what users actually feel is wider, stranger, and more consequential than the platforms’ marketing departments would care to admit.
The study arrives at a moment when AI-driven personalization has never been more technically sophisticated. Streaming giants are racing to replicate TikTok’s behavioral modeling, feeding watch time, pause patterns, and emotional resonance signals into ranking systems that operate well beyond what any user can see or control. Algorithms now analyze a broader range of data, including browsing history, location, and even mood-based preferences, to deliver more tailored content. Users spend, on average, more than two hours a day on social platforms – every scroll a data point, every pause a signal.
But sophistication is not the same as understanding. That distinction is the heart of what Twentify found.
On Instagram, the platform most users identified as delivering the strongest sense of personalization, the feeling of being known was real – but fragile. The Reels tab and the Explore section drew consistent praise for surfacing content in fashion, food, personal care, and humor that felt genuinely tailored. Then two things broke the spell: when the same content appeared repeatedly, or when a viral clip that clearly had nothing to do with the user’s interests invaded the feed, the sense of individual recognition collapsed. “In a social setting, when I see my friends getting the same videos I got, I start to feel like everyone gets the same thing,” one 41-year-old male participant said. The platform had stopped reading him; it was reading a category he had been sorted into.
TikTok produced a sharper divide along generational lines. Younger participants – those between 18 and 24 – credited the platform with strong personalization and accepted its algorithmic logic as a natural feature of the digital environment. Older participants were more likely to flag TikTok as the weakest performer, questioning whether the content it served had any meaningful connection to their actual preferences. The gap is not just aesthetic. Research published in Humanities and Social Sciences Communications found that younger individuals appear more amenable to personalization, while older demographics tend to associate algorithmic targeting with intrusion and surveillance.
X – the platform formerly known as Twitter – generated the lowest personalization expectations of any platform in the study, and that turned out to be its relative advantage. Users who opened it primarily for news and sports reported less frustration precisely because they were not expecting it to know them. What irritated them instead was repetition – the same topic cycling through their feed long after the news peg had passed.
The e-commerce finding is the one that will sting most for platform executives. Online shopping sites scored lower on the “feeling known” metric than every social platform in the study. Product recommendations were seen as functional – useful for narrowing choices and saving time – but they failed to produce the emotional register that Instagram’s best moments generated. The reason, participants said, was granularity. A recommendation engine that suggests similar shoes to the pair you browsed last week has learned your size and approximate price range. It has not learned your style, your context, or the difference between a weekday preference and a weekend one. “Sometimes I end up changing my mind based on a suggestion,” the 29-year-old participant said. “That bothers me.”
That discomfort – the sense that a recommendation has altered a decision rather than assisted one – is the line between personalization and manipulation. Twentify found a small but meaningful cluster of users who were not just neutral on algorithmic suggestions but actively suspicious of them, particularly when the same product followed them across multiple apps. Courts have begun to take that suspicion seriously, with recent litigation over addictive design and behavioral targeting reshaping how regulators in the United States and Europe approach what platforms are permitted to infer about a user.
Spotify was the outlier. It was the only platform in the study to generate what researchers described as a “soul-read” moment – the experience of a recommendation so precisely matched to a user’s mood that it felt less like an algorithm and more like a friend who knew them well. Daily mix playlists, year-end Wrapped summaries, and real-time mood-responsive suggestions drew consistent praise. On the core question of whether its personalization generates emotional resonance, the study’s participants gave Spotify a category of its own.
Netflix sat at the opposite end of the streaming spectrum. Users in the study described it as defaulting to what is popular rather than what is personal – surfacing whatever the platform’s marketing operation had decided needed an audience that week rather than what an individual’s viewing history might actually suggest. The distinction participants drew between Spotify and Netflix is instructive: both are subscription services with rich behavioral data, but one is perceived as using that data to serve the user, and the other as using the user to serve the catalog.
The study’s most consequential finding was not about any platform’s algorithmic quality. It was about what happens when trust collapses. One participant described becoming a victim of a fraud scheme originating on Instagram. After that experience, every personalization feature the platform had previously offered became evidence of the system’s menace rather than its usefulness. Research published in April 2026 in Information, Communication and Society, based on interviews with 32 users across two countries, found that initial emotional responses to perceived privacy violations include surprise, fear, creepiness, annoyance, and helplessness. Once that emotional threshold is crossed, no algorithmic improvement recovers what was lost.
The gender divide in the Twentify data is sharper than most platform design conversations acknowledge. Women in the study were more likely to experience personalization as useful and affirming, particularly in lifestyle, fashion, and food verticals where the recommendations tracked closely with what they were already seeking. Men reported a more detached relationship with the same systems – willing to call them functional but reluctant to grant them the authority of genuine knowledge. That difference suggests personalization is not a universal product but a highly context-dependent one, and that the design assumptions built into most recommendation systems may reflect the demographic that built them more than the full range of people using them.
Twentify conducted its research using semi-structured one-on-one interviews with 32 active users recruited across gender and age cohorts, all of whom had used the platforms studied within the prior 30 days. The methodology was qualitative; the findings are not projectable to a broader population in the way a large-scale survey would be. What qualitative research does well – and what the “Digital Mirrors” study does particularly well – is surface the emotional logic underneath behavior that surveys cannot reach.
What that logic reveals is an irony platforms have not yet solved. The more precisely an algorithm profiles a user, the more the user feels both served and surveilled – and the more any single failure of that precision shatters the relationship. One hundred accurate recommendations cannot undo the feeling produced by showing someone an item they already bought. Personalization that reaches a certain threshold of intimacy stops reading as convenience and starts reading as intrusion. The platforms that have come closest to solving this are those – like Spotify – where the emotional register of the product category makes a well-timed recommendation feel like a gift rather than a trace.
Whether the rest of the industry is capable of threading that needle remains the open question. Regulation in the European Union is now pushing platforms toward greater disclosure of how AI feeds use personal data – but enforcement, as multiple analysts noted this year, remains uneven. The gap between what the algorithm knows and what the user feels may not close until users have more visible control over the data generating those recommendations. Until then, the dress will keep appearing long after the purchase is done.
