The first 15,343 songs on MuziAI: who made them, what, and in which language
Real numbers from 15,343 songs users made on MuziAI: 88% brought their own lyrics, almost half are in English, pop is one third. How we counted and the limits.
By the MuziAI team. The numbers come from measurements we ran ourselves while building MuziAI; drafts were written with AI assistance, and every figure was checked against our measurement logs. Updated November 3, 2026.
When we opened MuziAI, we expected most people to type an idea and let the AI write the lyrics from scratch. We were wrong. After 15,343 finished songs, 88% of them use lyrics the user brought, and only 12% use lyrics written by the AI. This post is the full count from the site's database as of Nov 3, 2026: who makes songs, in which genres and languages, how long they are, and what these numbers cannot tell you.
What we counted, and what we did not read
Every number below is a count of labels from the site's song table. Only finished user songs are included (songs that have an audio file); the sample songs we made ourselves are excluded, and so are failed or pending songs. A user's song lives on a private page reachable only by its link, so we did not read the lyrics or ideas of any song for this post: we only counted genre, lyrics language, voice type, length, and whether the lyrics were written by the user or by the AI.
| Metric | Value |
|---|---|
| Finished user songs | 15,343 |
| People who made them (estimate, see limits) | 7,518 |
| Period | Sep 19 to Nov 3, 2026 (real use from Sep 30) |
| Lyrics brought by the user | 13,505 songs (88.0%) |
| Lyrics written by the AI from an idea | 1,838 songs (12.0%) |
| Songs with the user's own ABC sheet music | 789 songs (5.1%) |
| Average length | 191 seconds (3 min 11 s) |
Songs per day: a peak of 575, now about 400
The site opened on Sep 19, but the first ten days produced only 56 songs, mostly tests. Real use started on Sep 30. The busiest week was Oct 12 to 18 with 4,026 songs, 575 a day on average. The next two weeks dropped to 425 and then 404 a day, and the last seven days ranged from 337 to 441. We have not measured why volume fell, so we will not guess here.
Songs are made almost evenly around the clock: the quietest hour (04:00 UTC) has 492 songs and the busiest (12:00 UTC) has 811, a ratio of 1.65. If users were concentrated in one time zone, this curve would have a deep night-time dip; it does not, which matches users coming from many languages.
Almost half the lyrics are in English, Vietnamese is 3.1%
This was the biggest surprise for a site built in Vietnam. English lyrics make up 47.1% (7,227 songs), followed by Portuguese at 10.4%, Korean at 7.6% and Hindi at 6.2%. Vietnamese is ninth with 479 songs, 3.1%. The English interface also produced 62.3% of all songs.
Three lyrics languages stand out because the site had no interface in them for most of the period: Russian (691 songs) and French (578 songs) never had their own interface, and Spanish (593 songs) only got one on Oct 29. People writing in these languages used the English interface to have their own lyrics sung.
Genres: pop is a third, children's music is fourth
| Genre | Songs | Share |
|---|---|---|
| Pop | 4,986 | 32.5% |
| Ballad | 1,795 | 11.7% |
| Rap / Hip-hop | 1,748 | 11.4% |
| Children's | 1,029 | 6.7% |
| Rock | 990 | 6.5% |
| Acoustic | 968 | 6.3% |
| EDM | 577 | 3.8% |
| Cinematic | 563 | 3.7% |
| R&B | 491 | 3.2% |
| Vietnamese folk | 450 | 2.9% |
| Country | 438 | 2.9% |
| All other genres | 1,308 | 8.5% |
Children's music is fourth with 1,029 songs, ahead of rock: songs for kids are a bigger need than we expected. Bolero, a genre closely tied to Vietnamese listeners, has only 138 songs (0.9%), in line with the small share of Vietnamese lyrics.
The top genre changes with the lyrics language. For English, Portuguese, German, Hindi, Russian and Vietnamese lyrics, pop comes first. Korean lyrics are the exception: ballad comes first (382 songs against 319 for pop). Portuguese lyrics have acoustic in second place (233 songs), Russian lyrics have rock second (103 songs), and Vietnamese lyrics have ballad second (102 songs, after 152 for pop).
Mood: happy outnumbers sad almost two to one
Happy is the most common mood with 4,026 songs (26.2%), followed by powerful at 15.0%, sad at 14.0%, romantic at 13.6% and energetic at 12.0%. Happy, energetic and powerful together make 53.3% of songs; sad, lonely and nostalgic together make 21.5%. Peaceful (2.1%) and lonely (1.5%) are the least chosen moods.
Voice and song length
| Voice | Songs | Share |
|---|---|---|
| Female | 6,429 | 41.9% |
| Male | 5,769 | 37.6% |
| Male and female duet | 2,005 | 13.1% |
| Instrumental | 640 | 4.2% |
| Not recorded | 500 | 3.3% |
Instrumental music only became available on Oct 5, so counted from that date it makes up 4.5% of songs (638 of 14,191).
The average song is 191 seconds long. The 2:30 to 3:30 range is the most common (30.2%), followed by 3:30 to 4:30 (25.9%). Songs under 1:30 make up 11.6%; songs of 4:30 or more make up 14.7%. The longest song is 360 seconds (6 minutes).
How many people come back
| Songs per person | People | Share |
|---|---|---|
| 1 song | 5,224 | 69.5% |
| 2 to 4 songs | 1,783 | 23.7% |
| 5 to 19 songs | 459 | 6.1% |
| 20 songs or more | 52 | 0.7% |
Seven in ten people made one song. 30.5% made two or more, and 52 people made 20 or more, with a maximum of 145. The number of «people» here is an estimate; read the limits below before quoting it.
AI-written lyrics: who asks for them most
The share of songs with AI-written lyrics differs by the interface people used: lowest in the Korean interface (6.0%), highest in the Hindi (19.1%) and Vietnamese (16.6%) interfaces; the English interface sits at 12.2%. In every interface, most users still bring their own lyrics.
Limits of these numbers
- «People» is an estimate. The site has no accounts. One «person» is one hashed network address (the raw address is never stored), with IPv6 addresses grouped by /64. Several people sharing one address (an office, a mobile network sharing IPs) count as one; one person switching networks counts as several. The true number can be off in either direction.
- Genre and mood have two sources. They are the user's choice when the user picks one; when left at «Any», the system labels the song after reading its style description. 42 songs have no genre label.
- Lyrics language is the language the user selected when creating the song, not a language detected from the audio.
- «Lyrics brought by the user» includes both lyrics they wrote and lyrics copied from elsewhere. We do not read lyrics, so we cannot separate the two.
- These are counts, not quality ratings. This table does not say which songs are good, or which language the AI sings best. Those questions have their own measurements in the other research posts on this blog.
What this means if you are making a song
- If you already have lyrics, you are doing what 88% of users do: paste them into the «Write your own lyrics» tab, pick a genre and voice, and create. If you do not have lyrics yet, the AI lyrics tab takes a one-line idea.
- The most common length is 2:30 to 4:30. For an ad jingle or background music, choose the short length from the start.
- To hear styles before choosing, open the sample song library: every sample shows the full settings that made it and a button to reuse that style.
Try it now on MuziAI
Free, no sign-up required. Share an idea, and our AI lyrics generator writes and sings a complete song.
Create Song