Listeners interact much less deeply with music attributed to AI than with music attributed to a human – even when the music is definitely human-composed, a brand new peer-reviewed tutorial examine has discovered.
The authors say their findings recommend “truthful attribution can have actual penalties for the way music is perceived and understood” – touchdown in the midst of an energetic {industry} debate over obligatory AI music disclosure.
The peer-reviewed paper, which you’ll be able to learn in full right here, revealed within the journal Cognitive Analysis: Ideas and Implications in March, was authored by Sarah H. Wu of Stanford College and Kevin J. Holmes of Reed Faculty.
Varied music streaming providers have been rolling out their very own AI labeling methods in current months.
Apple Music launched its Transparency Tags system in March, asking labels and distributors to flag AI use on the level of supply.
Spotify adopted final month with the beta launch of AI Credit in tune credit, equally counting on labels and distributors to self-disclose AI use.
In late April, Spotify went additional by introducing a brand new “Verified by Spotify” badge, with the streaming large saying that profiles “that seem to primarily signify AI-generated or AI-persona artists” wouldn’t be eligible for verification.
“Within the AI period, it’s extra necessary than ever to have the ability to belief the authenticity of the music you hearken to,” Spotify mentioned in an accompanying weblog submit.
Deezer has gone additional nonetheless – independently detecting and tagging AI music on the platform degree, and now reporting that round 75,000 totally AI-generated tracks are uploaded to its service day by day, making up roughly 44% of every day deliveries.
Earlier this month, music supervisor Frederic Schindler made the case in an MBW op-ed for an industry-wide “Music Details” disclosure protocol modeled on FDA vitamin labels, with “AI Generated” listed as one in every of 4 obligatory origin classes.
The Wu and Holmes paper was primarily based on two preregistered research involving 399 US contributors, who listened to instrumental music clips and reported whether or not they imagined a narrative whereas listening – a phenomenon the researchers describe as “narrative listening”.
Within the first examine, contributors heard six human-composed items – together with works by Beethoven, Mozart, Debussy and Ravel – with out being advised who or what had composed them.
The extra strongly listeners believed a given piece was computer-generated, the much less probably they had been to think about a narrative – and the much less partaking the tales they did think about had been, in keeping with Wu and Holmes.
“FALSELY FRAMING AI-GENERATED ART AS A HUMAN CREATION MAY ELICIT A GREATER SENSE OF MEANING – BUT AT A COST TO HUMAN CREATORS, DEPRIVING THEM OF CREDIT AND COMPENSATION FOR THE WORK FROM WHICH AI PRODUCTS ARE DERIVED.”
SARAH H. WU AND KEVIN J. HOLMES
Within the second examine, the researchers used eight items – 4 human-composed, 4 generated by AI software program AIVA – and labeled each both “Composer: Human” or “Composer: AI” whereas it performed to the participant.
The “AI”-labeled items elicited fewer and fewer partaking imagined narratives than the “Human”-labeled items – no matter who or what had really composed the music, the Stanford and Reed researchers reported.
That suppression was nominally stronger when utilized to precise AI compositions – suggesting listeners had been additionally choosing up on acoustic markers of AI on high of the labels themselves, in keeping with Wu and Holmes.
“Attributing music to AI is related to – and might engender – an impoverished listening expertise, devoid of the psychological narratives that unfold because the composer’s musical decisions information the listener’s creativeness,” Wu and Holmes wrote.
The authors mentioned the label impact gave the impression to be pushed by listeners ascribing much less communicative intention to items marked as AI-made. “Labeling music as AI composed, in truth or in any other case, could lead listeners to deduce that the music lacks that means or depth,” Wu and Holmes wrote.
“Attributing music to AI is related to – and might engender – an impoverished listening expertise, devoid of the psychological narratives that unfold because the composer’s musical decisions information the listener’s creativeness.”
SARAH H. WU AND KEVIN J. HOLMES
These findings land alongside parallel analysis from Kiel and Hamburg economists Jana Friedrichsen, Julia Schwarz, and Michel Clement.
Their three-study working paper, not but peer-reviewed, was summarized in ProMarket final Monday (Might 4), and located that listeners’ willingness to pay for AI-generated music drops when its AI origin is disclosed – an impact primarily pushed by pop listeners.
“Customers can solely make knowledgeable decisions if artists and music platforms are clear about the usage of AI,” Friedrichsen, Schwarz and Clement wrote.
The Wu and Holmes paper opens with a reference to The Velvet Sunset, the “band” that surpassed 1 million month-to-month Spotify listeners in 2025 earlier than its operators confirmed the music was AI-generated.
The authors describe one in every of The Velvet Sunset‘s songs as making “for an satisfactory road-trip soundtrack,” including: “A lot AI-generated music could go undetected as a result of it’s designed to blur the distinctive qualities of human-composed works, yielding a form of algorithmically curated simple listening.”
For Wu and Holmes, that under-detection comes at a value to human creators.
“Despite the fact that AI methods can produce works that look or sound spectacular, audiences could interact with them in a reasonably shallow method, lacking the human contact that makes artwork really feel significant,” Wu and Holmes wrote.
Wu and Holmes added: “By the identical token, falsely framing AI-generated artwork as a human creation could elicit a higher sense of that means – however at a value to human creators, depriving them of credit score and compensation for the work from which AI merchandise are derived.”
That threat to human creators was illustrated earlier this 12 months by the case of Murphy Campbell, a North Carolina people musician who found AI-cloned covers of her songs uploaded to her personal Spotify profile, earlier than a copyright troll claimed possession of her official YouTube recordings by way of gamma-owned distributor Vydia.
The dimensions of that hurt extends far past Campbell‘s case. Sony Music Leisure revealed on the launch of the IFPI‘s World Music Report 2026 in March that it had requested streaming platforms to take down greater than 135,000 songs created by fraudsters utilizing generative AI to impersonate its artists.
Dennis Kooker, Sony Music‘s President of World Digital Enterprise and US Gross sales, mentioned the deepfakes trigger “direct industrial hurt to official recording artists.”
In a February MBW op-ed, IFPI CEO Victoria Oakley and RIAA CEO Mitch Glazier wrote that generative AI has “industrialized” streaming fraud.
Efficiency rights group ASCAP, in the meantime, has been calling for transparency round AI use in musical works since 2023, when its board adopted six AI rules together with a name to differentiate AI from human-generated works.
Concluding the paper, Wu and Holmes wrote: “For music to encourage our interior storyteller, it helps to know there’s a human thoughts behind it.”Music Enterprise Worldwide
