44% of New Songs Are AI-Generated: Why the Music Industry Is Moving Towards "Identity Display"

Before playing music, check whether the song was performed by humans or created by generative AI.

Such behavior might become a new habit when using streaming services in the near future.

Led by the International Federation of the Phonographic Industry (IFPI) and the Recording Industry Association of America (RIAA), organizations of independent music companies, the Grammy Awards organization, and unions of actors and singers have announced a common labeling system for recordings using generative AI.

The participating organizations proposed two types of labels: "AI-Generated" and "AI-Assisted."

The system is currently voluntary, but it is intended to be widely adopted in the global digital music market, not just in specific countries or by certain distribution services. There is also a plan to update the content in line with changes in technology and legal systems.

The goal is not to uniformly exclude music made with AI. It is to clarify how songs were created and to enable listeners to make choices based on their own values.

What Is the Difference Between "AI-Generated" and "AI-Assisted"?

The first label, "AI-Generated," is used when all or the main parts of the creative elements of a recording are made by generative AI.

For example, this applies when the lead vocals of a song are AI-generated, when the main instrumental performances are by generative AI, or when the entire song, including singing and accompaniment, is created based solely on textual instructions.

The second label, "AI-Assisted," refers to songs where human creation or performance is central, but generative AI is used for some expressions.

In this classification, it is assumed that the main vocals and instruments are performed by humans. Generative AI might be used for supplementary voices, sound effects, timbres, or editing tasks.

The important point is that not all works that use AI to some extent are treated as "AI-made music."

In modern music production, machine learning technologies are introduced in various processes, such as noise reduction, source separation, pitch adjustment, mastering, creating temporary vocals, and generating timbres.

It is not realistic to categorize songs written, composed, and performed by humans in the same category as songs automatically generated from short instructions. However, it is also not easy to determine from which point in the production process a work is "human-centric," and two labels alone cannot classify all works.

This system is not a finished form drawing a clear boundary between humans and AI, but rather the first common language to explain the rapidly complicating reality of music production.

Why Is a Display System Needed Now?

The main reason the music industry is rushing to respond is the speed at which AI-generated songs are increasing.

According to the French music streaming service Deezer, as of April 2026, about 75,000 fully AI-generated songs were being delivered daily to the service. This accounts for about 44% of newly uploaded music, amounting to over 2 million songs per month.

However, just because there are many uploads does not mean AI music is overwhelmingly popular compared to human music.

On Deezer, AI-generated songs account for only about 1-3% of total plays. Moreover, about 85% of plays recorded for AI-generated songs were judged to be due to fraudulent manipulation and were excluded from monetization.

These figures indicate that the issues surrounding AI music are not merely about the artistic debate of "Can AI create better songs than humans?"

With generative AI, a large number of songs can be created with much less cost and time than before. Furthermore, if play counts are inflated using automated accounts, there is potential for illicit revenue from streaming services.

If a large number of posted songs fill search results and recommendation sections, there is a risk that human artists and works produced with time and effort will become harder to discover.

Therefore, AI labeling is attracting attention not only as an explanation to listeners but also as basic information for detecting fraudulent plays, rights processing, revenue distribution, and managing recommendation functions.

Streaming Services Seek "Accurate Metadata"

The industry group DIMA, representing music streaming companies, has also shown interest in the new system.

DIMA includes operators such as Amazon, Apple Music, Spotify, TIDAL, and YouTube. The group explained that it is important for creators, rights holders, and distribution companies to provide accurate and prompt information to streaming services.

The key here is "metadata."

Metadata refers to information attached to audio sources, such as song titles, artist names, lyricists, composers, and rights holders. The new system envisions adding AI usage status to this metadata, passing information from creators to distribution agents, streaming services, and finally to listeners.

If the AI-used parts are accurately registered, labels can be displayed on app playback screens, and services can use this information for recommendations, monetization, fraud detection, and rights processing.

However, if creators or distributors, who are the starting point of information, do not report accurately, the system will not function.

If a poster who wants to avoid unfavorable treatment hides their use of AI, it is difficult to completely identify the production process from the finished audio alone. Conversely, if there is excessive reliance on detection technology, there is a risk of mistakenly judging human-made songs as AI-generated.

It is necessary to combine self-reporting and technical inspections with production history, rights information, and data at the time of distribution.

Deezer Already Displays AI Songs

Even before the announcement of a common system for the entire industry, Deezer had been advancing the detection and display of AI-generated music.

The company not only labels songs judged to be fully AI-generated but also excludes them from recommendation functions and playlists created by the editorial team. Furthermore, plays judged to be fraudulent are excluded from royalty calculations.

This approach does not completely ban AI music itself but aims to inform listeners of its origin and prevent illicit profit-making and excessive inflow into recommendation systems.

Spotify is also working on initiatives to demonstrate artist credibility from a different angle than AI labeling.

Announced in April 2026, "Verified by Spotify" is a system that reviews artist profiles and displays certification if they meet criteria related to authenticity and reliability.

This is different from the current label indicating AI use for each song. However, in an era where fictional singers and unauthorized postings by AI are increasing, it shares the goal of making it easier to confirm "who the artist is" and "whether the audio is approved by the artist."

In the future, there may be a move to combine artist-level identity verification with song-level AI usage display.

Transparency Welcomed on Social Media

On social media and in AI music creator communities, there are opinions welcoming transparency regarding the new system.

A representative viewpoint is that "listeners should be able to decide what they listen to."

If someone listens to a song without knowing it was made with AI and later learns that fact, the feeling of being "deceived" might outweigh the quality of the work. It is pointed out that disclosing AI use from the start makes it easier to maintain trust between creators and fans.

Users who post AI-generated music also express opinions such as "I clearly state that my work is AI-generated" and "listeners have the right to know the production method."

There is also a call for information disclosure to distinguish between works that use unauthorized voices resembling real singers and those that use AI as a creative tool after rights processing.

It is not the use of AI itself but the concealment of it, or the unauthorized use of others' voices or works, that provokes backlash. Ensuring transparency is seen as beneficial for protecting creators who use AI legitimately.

Criticism That "Two Types Are Too Broad"

On the other hand, there are also opinions that "two types of labels are insufficient" regarding the new system.

For example, there are works where humans write the lyrics and consider the direction of the melody, but AI generates the singing and accompaniment. Such works might be classified as "AI-Generated" along with works where both lyrics and music are entirely left to AI.

Conversely, even if a song is sung and performed by humans, using AI-generated sound effects in part would classify it as "AI-Assisted." The boundary between general digital music production and creation by generative AI may not be correctly conveyed to listeners.

On social media, there are proposals to display AI usage for each production process, such as vocals, instruments, lyrics, composition, mixing, and mastering.

The idea is that a system displaying only simple icons on the usual playback screen, with detailed information available by opening it, confirming "the vocals are human," "part of the accompaniment is AI," "the lyrics are human," is fairer.

In fact, the current labels target the recorded sound itself, and at present, the use of generative AI in lyrics, composition, music videos, and jacket images is excluded.

Even if the audio is performed by humans, if the lyrics and jacket are entirely AI-generated, the current display does not reveal the overall production picture.

The next challenge after implementing the system will be how far to expand the scope of the target.

The Potential for AI Labels to Cause Bias Against Works

On social media, there are concerns that labeling might prevent works from being fairly evaluated.

Someone might enjoy a song when listening to it without any information, but lower their evaluation upon seeing the "AI-Generated" label. In such cases, the judgment is based on the category of production method rather than the content of the music.

In discussions about AI work labeling, there are repeated concerns that simply adding a label might reduce views and reactions.

If creators who declare AI use transparently suffer disadvantages, while those who do not declare and make their works appear human-made gain more plays and revenue, the system risks becoming one where honesty is penalized.

To make the labeling system function fairly, it is necessary not only to distinguish those who declare but also to have mechanisms to verify undeclared works and address false declarations.

Additionally, caution is needed to ensure that the "AI-Generated" label is not treated as a warning for low quality, illegality, or plagiarism.

AI works created with the permission of rights holders are not the same as works that replicate the voice of a real singer without consent. Whether AI is used and whether there is rights infringement are fundamentally separate issues.

Labels are meant to convey production methods and do not automatically determine the quality or legality of a work. Without clarifying this point, labels risk functioning as de facto exclusion marks rather than information providers.

Can the Same Standards Be Applied to Both Major and Individual Creators?

In the AI music creator community, there are doubts about whether the same standards can be applied to both major labels and individual posters.

There is concern that if a major artist uses AI as production assistance, it will be labeled "AI-Assisted," while an individual using a generative AI service to create a song will be labeled "AI-Generated," with judgments varying based on affiliation or fame.

If the criteria remain ambiguous, classifications might differ by distribution company. If one service labels a song as "AI-Assisted" and another as "AI-Generated," it could cause confusion for both artists and listeners.

It is also essential to have procedures for creators who receive incorrect labels to verify the basis and request corrections.

If aiming for a global common system, it is necessary to align not only the design of the display but also the classification criteria, data recording methods, verification procedures, and mechanisms for objections.

The Question Is Not Just "AI or Human"

In discussions about AI music, the focus often shifts to the rivalry of "which is superior, AI or humans."

However, the issues that the current labeling system aims to address are not limited to that.

Who wrote the lyrics? Who performed? Whose voice is used? Were the works used in AI model training authorized? Who will the revenue be distributed to? And is the information conveyed to listeners accurate?

A single song involves multiple processes, including production, performance, recording, editing, distribution, and rights processing. With generative AI entering each of these processes, the traditional "artist name" and "lyricist/composer name" alone can no longer sufficiently explain the origin of a work.

What will be required in the future is not a simple binary choice of "human-made" or "AI-made," but a system that allows verification of the work's history.

If listeners want to judge based on simple displays, they can look at icons. If they want to know more, they can check which processes used AI and who holds the rights. Such phased information provision is ideal.

An Era Where "Ingredient Labeling" Is Needed for Music

Some people value the fact that music was performed by humans, while others want to evaluate it solely based on the song itself, regardless of the production method.

Some want to avoid AI music, while others actively seek it. Due to physical, economic, and technical circumstances, some people who previously found music production difficult use AI as a means of expression.

It is impossible to decide which value is correct.

Therefore, instead of mixing works with hidden production methods in the same market, it is important to provide the necessary information and allow listeners to make their own choices.

The proposed labels do not constitute a rule to ban AI music. Amidst the rapidly blurring boundaries between humans and AI, they are a first step in explaining the origin of music.

However, trust will not be established simply by preparing two marks.

The system only makes sense when accurate reporting, common metadata, technical verification, preservation of production history, fair classification, procedures for objections, and measures against unauthorized use and fraudulent plays are