AI Music Isn't English — The Hidden Languages of 650,000 Generations

In our earlier analysis of who actually authors AI music, one figure sat almost unremarked: across roughly 650,000 generations, about 93% of prompts are written in Latin-script languages. Read quickly, that number says "AI music is an English, Western thing." Read correctly, it says close to the opposite. The Latin alphabet is not English. It is the writing system of Spanish, Portuguese, Indonesian, Vietnamese, Turkish, French, Polish, Swahili, and thousands of other languages — and inside this dataset, English-language pop is a minority of the activity. The people making music with these tools are mostly not making it in English.

This piece looks at what the 93% actually contains. The sessions sort into roughly ten distinct musical worlds — Spanish ballads and devotionals, Brazilian sertanejo and funk, Indonesian dangdut, Turkish arabesk, Balkan songs, and more — and each of those worlds maps onto a real, documented surge in the streaming economy. The geography of AI music creation is not a surprise. It is the non-English music boom, arriving one step earlier in the pipeline.

The misread: Latin script is not English

The instinct to read Roman characters as English is a literacy illusion. The Latin alphabet is used to write more than 3,000 languages, spoken by about 70% of the world's population — across western and central Europe, sub-Saharan Africa, the Americas, and Oceania. A script is just a set of shapes; the language and the meaning are layered on top. As the linguistic reference puts it, a single writing system can represent many languages, and a single language can be written in several scripts. "Written in Latin characters" tells you almost nothing about which language you are looking at, and even less about whether it is English.

The dataset makes the point concrete. The most frequent word across all 650,000 prompts that isn't a structural tag is que — the Spanish and Portuguese function word — which appears 573,000 times, more often than chorus (452,000), the single most common English song-structure marker. amor ("love" in Spanish and Portuguese) appears about 79,500 times. When the corpus is clustered by meaning, the English-language pop cluster is only about 18% of a 50,000-session sample. Even counting English as it appears across the instrumental and electronic clusters, English-language activity stays under a third. The Latin script that a Western observer pattern-matches to themselves is, in this data, mostly carrying Spanish, Portuguese, Indonesian, Turkish, and French.

That is not a quirk of this dataset. It tracks a measured, accelerating shift in what the world listens to. English's share of Spotify's 10,000 most-streamed songs fell from 67.0% in 2021 to 62.1% in 2022 to 54.9% in 2023 — roughly a point of share lost per quarter, with Spanish, German, Portuguese, French, Korean, Hindi, and Indonesian picking it up. By 2024, music in eight different languages each generated more than $100 million in royalties on Spotify — English, Spanish, German, Portuguese, French, Japanese, Korean, and Italian — up from just two languages (English and Spanish) at that level in 2017. Among artists earning over $1 million, the number of languages represented also more than doubled. The economics of recorded music are de-Anglicizing, and the creation data is one step ahead of them.

Ten worlds under one alphabet

W1 named six of the clusters in passing. Here is what they are, and the documented music economies they sit on top of. The estimated composition of the sample — by language and region — looks like this:

Spanish: the planet's most-streamed language isn't English

The largest non-English world is Spanish, and it is not a niche. Bad Bunny was Spotify's most-streamed artist on the planet for three years running through 2022, with more than 18.5 billion streams in 2022 alone — recording entirely in Spanish. The number of Latin-music listeners on Spotify grew 986% from 2014 to 2023; in 2013 there were zero Latin songs in the global Top 100, and by 2023 more than one in five of the global Top 100 was Latin. In the United States — the supposed center of Anglophone pop — Latin music revenue hit a record $1.4 billion in 2024, growing faster than the overall market for the ninth straight year, with 98% of it from streaming.

Inside that world are sub-scenes the data shows clearly. Regional Mexican — corridos tumbados and música mexicana — grew more than 440% globally on Spotify between 2018 and 2023 and spread across Latin America (Colombia up 445%), with Peso Pluma reaching the global top five. And a quieter but large component is devotional: contemporary Christian music grew about 60% globally on Spotify over five years, surging specifically in Brazil, Indonesia, the Philippines, India, and South Africa. The Spanish cluster in the data is full of prayers, worship songs, and quinceañera tributes — a non-Western consumption base largely invisible to mainstream Western coverage.

Portuguese: Brazil builds its own charts

Brazil is now the ninth-largest recorded-music market in the world (2024 data), the first time two Latin American markets sat in the global top ten simultaneously (Brazil ninth, Mexico tenth) — and the fastest-growing of them all, up 21.7% in 2024 against a global rate of 4.8%. What dominates that 200-million-person market is Portuguese-language music. Sertanejo — Brazil's equivalent of US country music — is its single most popular genre, and the sertanejo duo Henrique e Juliano were Spotify Brazil's most-streamed act of 2024. Its grassroots counterpart, funk carioca, can top the literal global chart: Anitta's funk single "Envolver" made her the first solo Latin artist to reach No. 1 on Spotify's Global chart in March 2022.

The Brazilian scene is also a preview of what AI music tools accelerate: a democratized, independent creator base. Brazilian artists earned over R$1.2 billion from Spotify in 2023 — more than quadruple the 2018 figure — and over 70% of that came from independent artists and labels. The number of Brazilian artists earning more than R$100,000 a year rose over 500% since 2017. This is a Portuguese-speaking, streaming-saturated, independent-heavy market — exactly the profile of someone who reaches for a tool that removes the studio.

Southeast Asia: the volume the revenue charts miss

Southeast Asia is the cluster that most exposes how Western measurement misses the activity. Indonesia — the world's fourth-most-populous country — runs on dangdut and dangdut koplo, its national popular music, sung in Indonesian and Javanese, crossing every age and class. Yet in revenue terms Indonesia is invisible: as of a Billboard analysis, it contributed about 0.9% of Asia's recorded-music revenue while generating roughly 66 million minutes of Indonesian-artist music streamed worldwide per day, and around 10 billion total streams since Spotify launched there. Activity and money have decoupled — and the Western dashboard only reads money.

The pattern repeats across the region. On the inaugural IFPI Official Vietnam Chart in 2025, eight of the top ten tracks were by Vietnamese artists. In Thailand, local-language acts own the chart — Jeff Satur led both top local artist and song in 2024, with streaming making up roughly 92% of the market. Indonesia itself now has 212 million internet users and 356 million mobile connections — about 125% of its population. These are mobile-first, locally-driven music markets that the "AI slop" narrative, written from a Western desk, never enters.

Turkey and the Balkans: arabesk, turbo-folk, and the local-language majority

The electronic and Balkan clusters resolve into specific regional genres with no English-pop lineage. Arabesk — Turkish popular music built on Ottoman makam scales, mostly in minor keys — was one of Turkey's dominant forms for decades and spread across the Balkans, the Caucasus, and the Middle East. Its rise was the rise of an overlooked population: academic study ties arabesk's popularity to rural-to-urban migrants, disdained by the Westernized secular elite for its Eastern associations. Next door, turbo-folk dominates the former Yugoslavia so completely that when Billboard launched its Croatia Songs chart in 2022, the only Western acts on it were Glass Animals and the Red Hot Chili Peppers.

This isn't an Eastern-European footnote — it is the European norm. In IFPI's first "Music in the EU" report, domestic, local-language artists made up an average of 60% of the Top 10 across EU markets — all ten in Greece and Hungary, nine in Croatia and Italy, eight in Poland. And the fastest-growing music markets in Europe are the eastern ones: Bulgaria grew 44% in 2023, the Baltics 21.2%, and Poland 18.3%, against 4–5% in France and the Czech Republic. The local-language majority is the whole continent.

This was predictable: the clusters shadow the streaming surge

Lay the ten clusters over the documented streaming data and they line up almost one-to-one. The Spanish cluster sits on a market where Latin listening grew 986% and the world's most-streamed artist records in Spanish. The Brazilian cluster sits on the fastest-growing top-ten market on earth. The Turkish and Balkan clusters sit on the world's fastest-growing region and Europe's fastest-growing markets. AI music adoption is not landing in random places — it is landing exactly where music demand was already accelerating, just earlier in the pipeline, before the song reaches a chart.

The macro numbers say the same thing. Global audio streaming grew 14% to 4.8 trillion streams in 2024, but the growth was almost entirely outside the US — international streams up 17.3% versus 6.4% in the States. The US, UK, and Canada's combined share of premium streams slipped from 57.52% to 56.22% in a single year, with Mexico, Brazil, and India posting the biggest gains. By revenue, the three fastest-growing regions of 2024 were the Middle East and North Africa (+22.8%), Sub-Saharan Africa (+22.6%), and Latin America (+22.5%) — every one of them a non-English, largely non-Western market, all growing roughly five times faster than the global average. The center of gravity of music is moving, and it is moving toward the languages this dataset is already full of.

The blind spot runs two ways

Here the data turns genuinely strange. The regions doing the most AI music creation are the regions the AI itself was built to serve worst. A study of AI-music training data, "Missing Melodies," found roughly 86% of training hours come from the Global North versus 14.6% from the Global South — 6,128 hours of European music against just 27.5 hours of African music. The general-purpose models underneath are no better: GPT-3's training data was 93% English by word count, and Stanford researchers describe a "digital divide" in which nearly half of all websites are in English despite English being spoken by under a fifth of humanity. And when these models do generate non-Western music, they flatten it: a NeurIPS 2025 study of 73,000 AI-generated tracks across 147 languages and 79 countries found models systematically collapse regional genres — ghazal, opera — toward Western pop and rock.

Now hold that against where the tools are actually used. On Suno, the US is only about 19% of traffic, with Russia, Germany, Brazil, and Japan next — over 80% of usage comes from outside the United States. Broader GenAI adoption skews the same way: the US was only ~15% of ChatGPT traffic in 2025, behind India and ahead of Brazil and Indonesia. The result is a feedback loop: the heaviest users of AI music are being handed the models least fitted to their genres, so whatever "low quality" a Western critic perceives is partly an artifact of training bias — not of the creators.

It is worth being precise about what "slop" measures. The figure that launched the narrative — Deezer's report that about 44% of daily uploads are AI-generated — also notes those tracks are only 1–3% of actual streams, with up to 85% of their streams flagged as fraudulent. That is an upload-and-fraud metric on a Western platform: it counts what gets dumped into a royalty system, and it cannot see a person in Jakarta or Recife making a song. And when the same company ran a blind test, 97% of 9,000 listeners across eight countries could not distinguish fully AI-generated tracks from human-made ones. If almost no one can hear the difference, "slop" is not a property of the audio. It is a status judgment about provenance, made loudest in the markets whose monetization is stagnating, about music being created in the markets that are growing.

Continuity, not disruption

The cleanest way to read all of this: AI music did not invent a new creator class. It removed the last cost barrier for one that was already winning. The independent, non-elite, non-English creator is exactly who broke streaming open — over half of all artists earning more than $10,000 on Spotify come from countries where English isn't the first language; 70%-plus of Brazilian royalties go to independents. The dominant non-English genres were themselves bottom-up: funk built on cheap digital tools, arabesk on migrant audiences, dangdut across every class.

Producing one song the traditional way costs roughly $500 to $5,000 — studio time, a producer, mixing, mastering. AI tools drop that toward zero, and they drop it for a base that is already mobile-first and already creating: Indonesia alone has 356 million mobile connections; Latin America's mobile economy is worth $550 billion, 8.2% of regional GDP. The story isn't a machine flooding the West with garbage. It is a Vietnamese teenager, a Brazilian church group, and a Turkish wedding singer getting a studio for the price of a prompt — and writing in their own language.

That is the premise behind Lacuna, and the reason the product runs in nine languages rather than one. The world does not make music in English. Neither should the tools.