What 123,000 AI-Generated Songs Reveal About How People Actually Make Music With AI

The State of AI Music in 2026

AI music generation is no longer experimental. Suno, the market leader, generates 7 million songs per day — enough to replicate Spotify's entire 126-million-track catalog every two weeks. In November 2025, an AI-generated country song hit #1 on Billboard's Country Digital Song Sales chart for the first time. And 60 million people used AI to create music in 2024 alone, according to the IMS Business Report.

But the macro numbers obscure something more interesting: what are all these people actually making? What do they write in their prompts? What emotions are they expressing? What works, and what doesn't?

We analyzed 123,347 AI-generated songs created by 32,303 users on Neume over 13 months to find out.

1. Love Is the #1 Thing People Want AI to Sing About

The word "love" appears 29,646 times across all prompts — more than any other word by a wide margin. "Heart" follows at 13,867, then "feel" (13,063), "happy" (10,402), and "forever" (10,197).

People aren't using AI music generators to make experimental noise or test the technology. They're making love songs, heartbreak anthems, and songs for the people they care about. The emotional core of AI music creation is overwhelmingly romantic and personal.

The top themes by prompt content, ranked by how many songs mention them:

Theme	Songs	Avg Plays per Song
Family (mother, father, baby, child)	59,520	1.39
Worship & Gospel	24,204	1.50
Death & Grief	13,302	1.64
Birthday	10,192	1.15
Friendship	9,896	1.51
Motivation & Hustle	8,346	1.47
Depression & Anxiety	8,109	1.74
Love & Romance	7,824	1.48
Money & Wealth	7,248	1.51
Gaming	6,951	1.64
School & College	5,902	1.43
Party & Celebration	5,117	1.46
Diss & Roast	4,671	1.29
Breakup & Heartbreak	2,270	1.67
Wedding & Marriage	1,240	1.65

Source: Neume platform data, keyword analysis of 72,638 prompts

Family, worship, and grief are the top three categories — deeply personal, emotionally significant occasions that traditional music can't always serve in a personalized way.

2. Birthday Songs and Diss Tracks: The Use Cases Spotify Can't Serve

Two of the most distinctive use cases we found are birthday songs (10,192 songs) and diss tracks (4,671 songs). These are inherently personal — you can't find a birthday song for your specific friend named Sarah who loves cats on any streaming platform. And you certainly can't find a diss track about your coworker Dave who keeps stealing lunches from the office fridge.

But they behave very differently:

Birthday songs are high-volume, low-engagement utility. 93% of users who start with a birthday song never make anything else. They come, generate the song, share it once (average 1.15 plays), and leave. These users create an average of 2.94 songs total and almost never return the following month (1.3% return rate).

Diss tracks are social currency. They get an 85% like rate — the highest of any theme — and the fewest dislikes (0.012 per song). 13.7% of diss track creators go on to make songs in completely different genres. They create an average of 3.08 songs and are 3x more likely to diversify than birthday users.

The difference: birthday songs are a transaction. Diss tracks are entertainment. Both are valid, but they require fundamentally different product strategies.

3. Users Aren't Prompting — They're Composing

Perhaps the most surprising finding: 36% of all prompts are over 1,000 characters long. The average prompt is 145.8 words — a full paragraph. Users aren't writing "make me a happy song." They're writing complete lyrics with verse/chorus structure, production notes, and artist references.

And it pays off. Prompt length directly correlates with engagement:

Prompt Length	% of Songs	Avg Plays
Under 20 characters	6.5%	0.44
20–50 characters	12.1%	0.73
51–100 characters	11.7%	0.81
101–200 characters	11.3%	0.94
201–500 characters	10.5%	1.34
501–1,000 characters	12.2%	1.34
1,000+ characters	35.9%	1.59

Source: Neume platform data, 119,240 songs with prompts

Songs with detailed prompts including structure markers like [verse], [chorus], and [bridge] get 2.6x more plays and 3x more likes than short descriptions.

The implication: the best AI music comes from people who are already writers. The AI amplifies their existing creativity rather than replacing it. High-performing prompts read like poetry. Low-performing prompts read like instructions.

This aligns with the broader industry pattern. 10.6 million words were written in prompts on our platform alone — equivalent to roughly 106 novels. Across the industry, the SACEM/GEMA/Goldmedia study of 15,073 music creators found that 51% of creators under 35 already use AI for music — and they're using it as a creative tool, not a replacement for creativity.

4. Most AI Music Is Made to Be Created, Not Consumed

The play distribution of AI-generated music follows an extreme power law — steeper than anything seen on traditional streaming platforms.

Nearly half of all AI-generated songs are never played — not even once, not even by the creator. The median song receives a single play. Over 97% of songs receive fewer than 50 plays. And fewer than 0.01% of songs ever achieve what you might call meaningful traction.

Source: Neume platform data

This is the defining paradox of AI music: the act of creating the song is the product, not the song itself. For most users, the value is in the generation — seeing their words transformed into music — not in repeated listening.

This mirrors what's happening on streaming platforms at large. Of Spotify's 202 million tracks in 2024, 50 million had zero listeners and 175 million had fewer than 1,000 streams. The long tail of music — human or AI — is extraordinarily long.

And the gap is even more dramatic at the industry level: Deezer reports that AI-generated tracks now make up 28–39% of daily uploads to streaming platforms, but account for only ~0.5% of actual streams. AI music is being created at an unprecedented rate, but the overwhelming majority is personal, ephemeral, and never intended for an audience beyond the creator and their immediate circle.

5. AI Music Is a Global Language

AI music creation isn't just an English-language phenomenon. Non-English words appear frequently across prompts, with several languages showing strong representation:

Language	Sample Words	Relative Prevalence
French	dans, pour, avec, amour	██████████ (1st)
Spanish	para, amor, como, porque	████████░░ (2nd)
Filipino/Tagalog	ikaw, bawat, puso, mahal, dahil	██████░░░░ (3rd)
German	mein, dein, liebe	█████░░░░░ (4th)
Korean	(via K-pop tags and prompts)	███░░░░░░░ (5th)

Source: Neume platform data, word frequency analysis of all prompts. Note: raw word counts can be skewed by prolific individual creators; rankings reflect overall frequency, not unique users.

Filipino users are creating love songs, worship music, and personal dedications in Tagalog. French and Spanish — the world's most widely spoken languages after English and Mandarin — lead overall. Korean shows up primarily through K-pop-style tags and prompts rather than full Korean lyrics.

This aligns with broader industry data: fans of regional genres and non-Western music styles express the strongest interest in using AI to make music. AI music tools are democratizing music creation globally — particularly in markets where professional studio access is limited.

6. What AI Does Well (and What It Can't Do Yet)

We tracked like/dislike ratios across all genre tags (minimum 20 reactions) to find where AI music delights — and where it disappoints.

Genres AI excels at (highest like rates)

Genre	Songs	Like Rate	Avg Plays
Rap vocals	87	100%	11.37
Country pop	103	95.2%	2.76
Black gospel	173	94.3%	1.62
Indie folk	300	93.9%	1.25
K-pop	898	93.9%	1.59
Love song	1,091	88.9%	1.64
Bilingual	576	88.2%	1.97

Genres AI struggles with (lowest like rates)

Genre	Songs	Like Rate	Avg Plays
Stadium anthem	110	31.3%	1.37
Marching band	84	33.3%	1.13
Power ballad	107	37.5%	1.82
Sports anthem	231	43.8%	1.16
Orchestral	707	46.3%	1.08
Christmas	1,988	54.8%	1.08

Source: Neume platform data, 9,252 total reactions across 123,347 songs

The pattern is clear: AI excels at intimate, vocal-driven music — rap, country, gospel, indie, K-pop. These genres are built on voice and simple instrumentation (guitar, piano, beat). The AI can deliver a convincing, emotionally resonant performance.

AI struggles with music that requires complex production — orchestral arrangements, stadium-filling anthems, symphonic metal, power ballads. These genres demand layered instrumentation, dynamic range, and precise multi-instrument coordination that current AI models can't reliably deliver.

In other words: AI music's superpower is intimacy, not grandeur. It's a personal songwriting tool, not a replacement for a symphony orchestra.

This is consistent with broader perceptions. A Deezer/Ipsos study of 9,000 participants found that 97% of listeners couldn't distinguish AI from human music in blind tests. But when told a track was AI-generated, listeners rated it as significantly less moving — a perception bias that vanishes when the music is personal and emotionally relevant to the listener.

7. Name-Dropping Works: How Artist References Shape AI Music

Referencing a known artist in your prompt — "in the style of Taylor Swift," "like Kendrick Lamar" — produces measurably better results:

	Songs	Avg Plays
References an artist	7,651	1.58
No artist reference	111,589	1.16

36% more plays when an artist is referenced. The AI performs better with a stylistic target.

The most-referenced artists and how their songs perform:

Artist	Songs	Avg Plays	Notes
Conan Gray	10	43.80	Small sample, huge engagement
Gracie Abrams	15	29.27	Indie pop resonates
Taylor Swift	179	3.58	High volume + high plays
BTS	604	1.84	Most referenced overall
TWICE	956	1.47	Largest K-pop presence
Eminem	493	1.19	High volume, average engagement
Drake	210	0.87	Below average

Source: Neume platform data, regex matching across 72,638 prompts

Taylor Swift is the gold standard: 179 songs with 3x average plays. Her narrative, emotionally-driven style translates exceptionally well to AI generation. K-pop groups (BTS, TWICE, BLACKPINK, BABYMONSTER) dominate by volume — the K-pop fanbase is a massive driver of AI music creation, as fans create songs in the style of their favorite groups.

Interestingly, Drake and NBA YoungBoy references perform below average — their production-heavy styles may not translate as well, or users referencing them may be less engaged with the output.

8. Depression Songs Get the Most Replays

The theme with the highest engagement per song isn't love, or birthday, or party music. It's depression and anxiety — 8,109 songs with an average of 1.74 plays and the highest average likes (0.096 per song).

Breakup songs are second (1.67 avg plays), followed by wedding songs (1.65) and death/grief (1.64).

People replay songs about pain. They use AI music as a form of emotional processing — writing out their feelings, hearing them sung back, and listening again. This is consistent with research on music and emotional regulation: music serves a functional emotional purpose, and AI makes it possible to create perfectly personalized emotional music on demand.

Meanwhile, birthday songs (1.15 avg plays) and Christmas songs (1.08 avg plays) are the lowest-engagement themes. They're created for a moment, shared once, and done.

The emotional spectrum of AI music engagement:

Depression/Anxiety  ████████████████████████████████████  1.74 avg plays
Breakup/Heartbreak  █████████████████████████████████░░  1.67
Wedding/Marriage    ████████████████████████████████░░░  1.65
Death/Grief         ████████████████████████████████░░░  1.64
Gaming              ████████████████████████████████░░░  1.64
Friendship          ██████████████████████████████░░░░░  1.51
Worship/Gospel      ██████████████████████████████░░░░░  1.50
Love/Romance        █████████████████████████████░░░░░░  1.48
Diss/Roast          █████████████████████████░░░░░░░░░░  1.29
Birthday            ██████████████████████░░░░░░░░░░░░░  1.15

9. The Industry Context: Where AI Music Is Heading

Our platform-level data sits within a rapidly evolving industry:

The market is growing fast. The generative AI in music market was valued at $569.7 million in 2024 and is projected to reach $2.79 billion by 2030, a 30.5% CAGR. AI music generation startups raised $250 million in equity funding in 2025 alone.

Suno dominates. With a $2.45 billion valuation, $300 million in ARR, 2 million paid subscribers, and 7 million songs generated daily, Suno produces in two weeks what Spotify's entire catalog took decades to accumulate.

The legal landscape is stabilizing. After the RIAA sued Suno and Udio in June 2024, Warner Music settled and signed licensing deals with both companies in November 2025. Suno will launch fully licensed models in 2026. The U.S. Copyright Office has clarified that purely AI-generated content cannot be copyrighted, but AI-assisted works with meaningful human authorship can.

Musicians are worried — and also using it. 79% of musicians worry about AI competition, and creators could lose up to 24% of revenue by 2028 according to a CISAC study. Yet 87% of producers already use AI tools in their workflow — for mastering, arranging, generating loops, and creating visuals.

Fraud is a real problem. Deezer reports that AI tracks make up 28% of daily uploads but only 0.5% of streams, and up to 85% of those AI streams are fraudulent.

10. What We Learned: Key Takeaways

For creators using AI music tools:

• Write detailed prompts with full lyrics and [verse]/[chorus] structure — it produces dramatically better results (2.6x more plays)
• Reference an artist to give the AI a stylistic target (36% more plays)
• Stick to genres AI handles well: rap, country, K-pop, gospel, indie folk
• Avoid genres that need complex production: orchestral, symphonic, stadium anthems

For the music industry:

• AI music is primarily a personal expression tool, not a Spotify competitor. 43% of songs are never played. Users create for themselves and a small circle — not for audiences.
• Birthday songs and diss tracks are entirely new use cases. They represent personalized, occasion-driven music that traditional platforms can't serve.
• The Filipino, French, Spanish, and German-speaking markets are actively creating AI music in their own languages. Localization matters.
• Depression and grief songs get the most replays. AI music is being used as emotional therapy. This has implications for mental health, music therapy, and platform responsibility.

For researchers and analysts:

• The gap between AI music creation (millions of songs/day) and consumption (0.5% of streams) is the defining tension of this era. People want to make AI music far more than they want to listen to it.
• Prompt engineering for music is a real skill. The 36% of users writing 1000+ character prompts are essentially composers using AI as an instrument — a fundamentally different behavior than the "make me a song" casual user.
• Engagement with AI music follows an extreme power law more severe than traditional streaming. The long tail is even longer.

Methodology

This analysis is based on first-party data from the Neume AI music platform, covering the period February 2, 2025 through March 10, 2026.

Dataset:

• 123,000+ songs (98.4% completed successfully)
• 32,000+ unique creators
• 72,000+ prompts analyzed
• 71,000+ lyrics analyzed
• 9,000+ reactions (76.6% likes, 23.4% dislikes)

Song metadata includes: title, tags, provider, play count, creation date, prompt text, lyric text, explicit flag, like/dislike counts, remix status, and user attributes.

Industry statistics are sourced from public reports, press releases, and company announcements, cited inline throughout the article. All Neume data is queried directly from the production database.

Limitations: Play counts reflect in-platform plays only and do not include external shares. Like/dislike data is only available from August 2025 onward. Language detection is based on keyword matching, not formal NLP classification. Theme categorization uses regex patterns and may undercount some categories.