Credit: Alex Shuper For Unsplash+
In A Nutshell
- Researchers analyzed nearly 70 years of Eurovision entries and found that singing in English, choosing pop, maximizing danceability, and packing in more lyrics became a shared competitive “code” learned collectively by nations over decades.
- Four countries, France, Italy, Portugal, and Spain, appear to have knowingly resisted the English-language advantage, likely prioritizing cultural identity and language promotion over winning.
- Once the winning formula became universal, it stopped guaranteeing success and became a baseline just to stay competitive, a pattern researchers call the “Red Queen” effect.
- Eurovision’s organizers have also been learning, repeatedly adjusting voting rules and expanding participation to keep outcomes unpredictable and audiences engaged.
Sing in English. Make it a pop song. Pack in more words than the competition. And get people moving. That is the rough “code” to winning the Eurovision Song Contest, according to a new study that analyzed nearly 70 years and more than 1,700 songs from the world’s most-watched music competition.
But this is not a magic recipe. A key point of the study is that these traits eventually became table stakes, not a guarantee of victory. No single genius songwriter cracked the code. Instead, entire nations learned these lessons over decades by watching what won and what did not, then adjusting their entries accordingly, almost like students copying the strategies of classmates who keep acing the test.
Published in Royal Society Open Science, the study also uncovered something more provocative: four countries, France, Italy, Portugal, and Spain, appear to have resisted the biggest lesson of all. English lyrics were consistently linked with stronger Eurovision performance, yet these four nations kept returning to their own languages. Researchers argue this is consistent with valuing national culture and language promotion over pure competitiveness, a posture that may reflect broader attitudes about Anglo-Saxon cultural influence well beyond a song contest.
How Researchers Cracked the Eurovision Song Contest Code
Led by Luis A. Nunes Amaral of Northwestern University along with Arthur Capozzi and Dirk Helbing of ETH Zürich, the team combed through every Eurovision contest from 1956 to 2024, covering 1,763 songs from 51 countries. They used Spotify’s audio data to measure traits like danceability and acousticness, an AI model to classify musical genres, and a large language model to identify the emotional themes running through each song’s lyrics.
English Dominates Eurovision, but Four Countries Pushed Back
In Eurovision’s early decades, songs arrived in a range of European languages, with French dominating among winners. By the 1970s, roughly 40 to 60 percent of winning songs were already in English, even when rules required native-language entries. By the early 1990s, that share had reached about 80 percent among winners.
When a permanent rule change in 1999 allowed any language, roughly 80 percent of all competing songs switched to English or a mix of English and a native language almost overnight. Countries had been watching the winners for years and drew the same conclusion: English lyrics meant a better shot at the top.
France, Italy, Portugal, and Spain bucked that trend. Even after restrictions lifted, these nations mostly submitted songs in their own languages. Researchers note it is unlikely they missed the pattern. Rather, the behavior appears consistent with prioritizing cultural identity over winning. French, Spanish, and Portuguese are spoken by enormous populations worldwide, making language promotion a rational trade-off.
Italy’s case is further shaped by a deep domestic music tradition: the paper notes that competitions such as the Sanremo Music Festival can draw far more attention at home than Eurovision, and Italy’s entry has typically come through that tradition. Germany, despite its size and economic weight, did not show the same resistance.
How Pop Music and Longer Lyrics Took Over Eurovision
Beyond language, the researchers tracked how Eurovision songs changed sonically. Early contests featured acoustic performances with orchestras. Over time, that gave way to electronic production and higher danceability. Statistical models confirmed that countries adjusted their entries based on the average danceability of top-five songs from the previous five years, with the pattern statistically significant for eight countries.
Pop has steadily taken over, though not as completely as English. About half of recent top-performing songs are classified as pop, but rock and other styles still appear among winners, meaning the contest retains some musical variety.
Lyrics evolved too. Average word counts held steady around 230 through most of the late 20th century, then jumped roughly 20 percent around 1999. Top-three songs were already trending longer by the late 1980s, suggesting wordier songs had a competitive edge countries gradually noticed and copied.
Emotional content shifted as well. Nostalgia, once dominant, has been in steep decline, possibly because of increasing distance from World War II and improving economic conditions, the researchers suggest. Pain, Rebellion, and Desperation have been rising. Since about 2010, top-performing songs have featured Pain more than the average competing entry, a trend the authors note may not be coincidental given the aftermath of the Great Recession.
Why Following the Eurovision Winning Formula No Longer Guarantees a Win
As English, pop, and high danceability became standard practice across the field, those features stopped being advantages and became baseline requirements. Researchers call this a “Red Queen” effect: countries must keep running just to stay in place. Top-performing songs do not simply meet the baseline. They deviate from the norm in ways that set new standards for everyone else to chase.
Countries that chose not to play by the rules at all, like France and Portugal, essentially opted out of that arms race in favor of something they appeared to value more: cultural identity on a continental stage. In a competition where nearly everyone else has converged on the same language, the same genre, and the same danceable beat, holding firm to your own tongue turns out to be its own kind of statement.
Disclaimer: This article is based on an observational study and reflects the findings and interpretations of the researchers. It does not constitute definitive proof of intent or strategy on the part of any country’s broadcasters or songwriters. Findings about language, genre, and lyric trends are based on statistical patterns across nearly seven decades of contest data and should be understood as associations, not certainties.
Paper Notes
Limitations
The study relies on several external tools whose inner workings are not fully transparent. Spotify does not disclose exactly how its system assigns audio features to songs, though its tools are widely used in research and considered sufficiently reliable. The AI genre classification model had a weighted accuracy of 0.815, meaning roughly one in five songs could be misclassified. Song theme annotations were generated by GPT-4o, which introduces potential variability in how themes were assigned. The statistical models for country-specific learning did not converge for all countries, meaning the evidence of learning is stronger for some nations than others. Finally, the Expansion period is relatively recent, and the authors caution it is still too soon to fully judge the effects of increased public voting or the potential for manipulation through automated voting schemes.
Funding and Disclosures
Luis A. Nunes Amaral received partial financial support from the project “CoCi: Co-Evolving City Life,” funded by the European Research Council under the European Union’s Horizon 2020 program, grant agreement no. 833168. Arthur Capozzi received partial financial support from the European Union Horizon 2020 Program under grant agreement no. 871042, “SoBigData++: European Integrated Infrastructure for Social Mining and Big Data Analytics.” Dirk Helbing coordinated and contributed to research in both projects. No competing interests were declared. Large language models were used to classify themes of song lyrics. No ethical approval was required.
Publication Details
Title: Breaking the code: Multi-level learning in the Eurovision Song Contest | Authors: Luis A. Nunes Amaral (Department of Engineering Sciences and Applied Math, Northwestern University, Evanston, IL, USA), Arthur Capozzi (Computational Social Science, ETH Zürich, Switzerland), and Dirk Helbing (Computational Social Science, ETH Zürich, Switzerland; Complexity Science Hub, Vienna, Austria) | Journal: Royal Society Open Science, Volume 13, 251727 | DOI: https://doi.org/10.1098/rsos.251727 | Received: 9 September 2025 | Accepted: 6 February 2026 | Data and code availability: https://github.com/amarallab/RS-Open_Science-Breaking_the_Code/tree/main







