Your Playlist Predicts Your IQ: 58,247 Songs Analyzed

Pull up your streaming history right now. Count the last twenty songs. More Radiohead or more Pharrell? More Billie Eilish or Dua Lipa? A team of German researchers claims the answer correlates with your intelligence — and not in the direction most people assume.

Scientists at Ludwig Maximilian University of Munich installed a tracking app on 185 volunteers’ smartphones. For five months, the app quietly logged every song played. Then a machine learning algorithm tried to predict each person’s cognitive ability using nothing but their listening data. 58,247 unique tracks. 215 features per song. The results surprised even the authors.

Five Months of Eavesdropping

Fluid reasoning — the ability to solve novel problems without relying on learned knowledge. It is one of the core components of general intelligence and among the hardest to train.

No questionnaires about favorite genres. No self-reports. The app ran silently in the background, capturing real behavior over five months — a crucial upgrade over typical one-shot surveys. Each participant also completed a brief cognitive test on their phone, measuring fluid reasoning, vocabulary, and mathematical knowledge.

From each song, the algorithm extracted 215 features: tempo, key, energy, danceability — standard audio descriptors. But it also parsed lyrical content, emotional valence of words, and thematic clusters. A nonlinear machine learning model then searched for patterns linking this mountain of data to test scores.

Lyrics Outweigh the Beat

The central finding: song lyrics predicted cognitive ability far better than musical characteristics. Tempo, key, loudness — these added almost nothing to the model’s accuracy. Words did.

People who scored higher on cognitive tests gravitated toward songs with less positive emotional tone. Not gloomy. Not nihilistic. Reflective. Melancholic. Introspective. Lead author Larissa Sust suggests these listeners use music as a tool for self-reflection rather than pure entertainment.

Several other patterns emerged. Lyrics focused on the present moment, perceived as honest, and related to themes of home were associated with higher cognitive scores. Heavy use of social words and tentative language pointed in the opposite direction.

The ‘Smart Listener’ Profile

Only one audio feature consistently predicted higher intelligence: low «liveness» — studio recordings over live performances. Think about it. Live recordings carry crowd noise, imperfect sound, spontaneous variations. A studio track is deliberate, layered, precisely mixed.

Total listening time mattered too. The more someone listened overall, the higher their predicted intelligence. Another marker: a preference for songs in foreign languages.

Assemble the pieces and a profile emerges: heavy listener, prefers studio recordings, drawn to thoughtful lyrics in non-native tongues. Before you rush to rebuild your queue — an important caveat.

Correlation Is Not Causation

Sust is upfront about it: the predictive power of music listening alone is «quite small.» The effects are statistically significant but too modest for any practical application. Swapping Dua Lipa for Radiohead will not raise your IQ. The relationship runs the other way: a certain cognitive profile manifests partly through musical taste.

The sample of 185 participants is modest. Age, education, socioeconomic background — all potential confounders — were not fully controlled. Reddit commenters (11,000 upvotes, 900 comments) were blunt: «Confusing cause and effect.» «Poorly conducted study with a clickbait headline.»

The criticism is partly fair. But the study has one clear advantage over older work: it measured actual behavior, not answers to a survey asking «what music do you like?» Five months of continuous data carries more weight than a one-time questionnaire.

Music as a Digital Fingerprint

This study fits into a broader trend: predicting psychological traits from digital traces. Facebook likes, YouTube watch history, smartphone usage patterns — all of these already function as indirect personality tests. Music is simply another data layer.

Digital phenotyping — using smartphone and wearable data to assess health and psychological states. A growing field in psychiatry and cognitive science that turns passive device usage into clinical signals.

For now, these models are too weak for individual-level predictions. At the population level, though, they work — and they will sharpen as samples grow. The question is no longer whether a playlist can reveal something about intelligence. The question is who gets access to that signal and what they do with it.

In the meantime, listen to whatever you want. Your IQ will not budge. But your playlist may know you a little better than you think.