The AI Song Report, October 2026: What People Ask AI to Sing About

Key findings

• Own words, often. 45.4% of requests came with the person's own lyrics; 52.9% were described in words; 6.1% had the Instrumental switch on. Over half of all people (52.5%) brought their own lyrics at least once.
• Descriptions are paragraphs, not prompts. When people described a song instead, the median description was 68 words. Only 3.6% used five words or fewer.
• One in six is a gift. 17.5% of requests were for an occasion or a person: a birthday, a wedding, a partner, a parent, a best friend, an apology, a memorial.
• Mothers before birthdays. Songs about or for a mother made up 6.1% of all requests, more than birthday songs (4.3%), romantic songs (3.6%) and songs about or for a father (3.3%).
• Pop and rap lead. Pop was named in 18.0% of requests and rap or hip-hop in 11.3%, ahead of rock (5.5%), R&B (4.8%), dance and electronic (4.8%) and country (3.2%).
• English is the minority. Only 26.7% of people sang mainly in English. Next came other Latin-alphabet languages (21.1%), Spanish (11.8%), French (8.8%) and Persian or Urdu (6.1%), and 8.3% of people sang in more than one language.
• Three minutes and change. The median song ran 3:13; 36.4% ran three to four minutes and 23.4% ran past four. Songs from people's own lyrics run longer at the top end: 31.6% passed four minutes, against 18.1% of songs made from a description.
• About a minute. From pressing Create to both finished takes took a median of 1:04, and 85.5% of requests were done inside two minutes. The weekly median fell from 2:48 in the first days of September to 0:56 in the week of 5 October.

What people make songs for

Most requests are not for an occasion at all: they are songs people want to hear, about their own lives, in their own genre. But one in six is made for someone, and the people they are made for are close family more often than the calendar would suggest.

A request counts toward an occasion when the person's own words name it: in the description, a style they typed or a title, in any of the site's languages (cumpleaños, mãe, Hochzeit, 生日 and so on). Lyrics count only when the word comes back at least twice, because one passing "my mother said" does not make a song for a mother.

Not shown: songs for a wife, a husband or a best friend, graduations, memorials, apologies, funny songs, diss tracks and jingles. Each was too small or too concentrated in a few people to pass the rule above. They are in the data; we just won't put a number on them yet.

The genres people ask for

Genre is counted only from what people asked for musically: the description, a style they typed, a genre they picked from the menu. Never from lyrics, where "rock me" and "let's dance" are lines, not requests. A request can name more than one genre, and many name none, so the column does not add up to 100%.

Lo-fi, worship, metal, jazz and lullabies are asked for too, but each fell short of the people rule in this window.

Own lyrics or a description

Tunely has two ways in: describe the song and let it write the lyrics, or paste your own and have them sung word for word. The split is close to even, and it is recorded at the moment of the request, not guessed from what the text looks like afterwards.

• 45.4% of requests brought the person's own lyrics; 52.9% described the song; the remaining 1.7% were instrumentals typed as a style alone.
• 52.5% of people brought their own lyrics at least once.
• Descriptions had a median of 68 words; 3.6% were five words or fewer.
• 20.1% of requests said how fast the song should go in the person's own words (a BPM, "slow", "upbeat"), not counting the tempo menu.
• 6.1% of requests had the Instrumental switch on.

The languages people sing in

Language is read from the lyrics of the song that came back, so it is the language actually sung, whether the person wrote the words or described the song in that language. Because a few people make many songs in one language, we count people here, not requests: each person once, by the language most of their songs are in.

Base: people with at least one sung song whose lyrics could be identified. 8.3% of them sang in more than one language. Italian, Chinese, Japanese and Korean appear too, but none of them passes the people rule yet, so we give them no number. "Other" groups Latin-alphabet languages outside English, Spanish, Portuguese, French, German and Italian.

How long the songs are

Lengths are measured from the audio files of the songs that came back (each request returns two takes), and checked against a separate file-by-file measurement: the two agreed to the second.

Across every song, 15.2% ran under two minutes, 24.9% two to three, 36.4% three to four and 23.4% over four. The longest songs come from people's own lyrics: the length follows the words, and a long lyric sheet makes a long song.

How long it takes

Time is measured from pressing Create to both finished takes, on requests from signed-in accounts (a visitor's request can be rewritten later when they sign up, which would blur its timing). A separate log of finished generations, which includes visitors, gave the same median within a second.

• Median 1:04; a quarter finished within 0:52 and three quarters within 1:28.
• 85.5% finished inside two minutes.
• By week (Monday to Sunday): 2:48 for 2 to 6 September, then 1:31, 1:09, 1:01 and 0:58, and 0:56 in the week of 5 October.

Method

Data. Every song request made on Tunely from 2 September 2026 00:00 UTC to 13 October 2026 00:00 UTC, as recorded at the moment of the request: whether the person brought lyrics or a description, the Instrumental switch, and the exact text and choices they submitted. Tunely is a web app for making songs with vocals from a description or from your own lyrics.

Who is counted. Customer accounts and visitors without an account. Our own test, staff and demo accounts are excluded by an explicit list. A person is their account; a visitor is their browser, or the account they later signed up with.

The people rule. A share is published only when at least 20 different people are behind it and no one person makes up more than 25% of it. Categories that fail are named but not given a number.

Categories. Fixed word lists per category, written before measuring, in the site's eight languages (English, Spanish, Portuguese, French, German, Chinese, Japanese, Hindi) plus a few others where the words are distinctive, among them Arabic, Russian, Turkish, Italian and Urdu. Occasions count the person's own words, and lyrics only when the word recurs. Genres count only what was asked for musically, never lyrics.

Language. Detected from the lyrics of the returned song: by script first (Arabic, Cyrillic, Devanagari, Han, kana, Hangul), then by words that belong to one language only. Arabic script with Persian or Urdu letters is reported as Persian or Urdu. Latin-alphabet languages outside English, Spanish, Portuguese, French, German and Italian are grouped as "other".

Lengths and timing. Lengths come from the audio files. Timing comes from the request's own record, cross-checked against the generation log.

What we don't publish. Totals of requests, songs or people, and anything a person wrote. Every figure here is a share, a median or a percentile.

• Limits. These are the people who use Tunely, not all music listeners or all AI music users; the site's languages and its gift pages shape who arrives. Word lists miss requests that describe an occasion without naming it, so occasion shares are floors, not estimates of everything. Six weeks include no major holidays, which would raise some occasions.

How to cite and reuse

The figures are free to quote and reuse under CC BY 4.0: credit Tunely and link to this page. Every published figure is in the CSV (aggregates only). For questions about the method, or for a figure we didn't publish, write to hello@tunelyai.com.

Tunely (2026). The AI Song Report, October 2026: What People Ask AI to Sing About. Data: song requests on Tunely, 2 Sep–12 Oct 2026. https://tunelyai.com/en/blog/ai-song-report-october-2026