News & Press

Earlier this year we put generative AI to the test to see if it could out-create a human composer when given specific musical briefs. The results (featured in Forbes) showed that while our human composer proved superior, AI’s creations rated highly for musical appeal. Its primary weakness, however, was emotional accuracy — inconsistently conveying specific emotional attributes in its creations. With rapid advancements in generative AI music platforms, we decided to perform this experiment again, only this time with pop music. Can AI write a better song than Taylor Swift?

Given Taylor’s dominance of the pop music scene over the past decade and her ability to crank out hit after hit, it may be more appropriate to refer to this study as “Machine vs. Machine,” however, we felt that her music would provide the quintessential challenge for AI. Working with leading sonic testing agency SoundOut, we conducted a study whereby we reverse-engineered five of Taylor’s biggest hits into creative briefs and gave them to a popular AI platform (Suno), the output of which was then tested against the originals for appeal and emotional accuracy.

Methodology
We began by selecting hits from Taylor’s last five albums: Cruel Summer (Lover, 2019), Evermore (evermore, 2020), Red (Red, Taylor’s Version, 2021), Anti-Hero (Midnights, 2022), and So High School (The Tortured Poets Department, 2024). These tracks were run through OnBrand, SoundOut’s proprietary platform mapping 240+ emotional attributes to each. Creative briefs (see Appendix) were designed for each song using the strongest rated attributes along with musical style and direction, and then provided to Suno to produce comparative tracks. No mentions, audio files, or references of any kind to Taylor Swift were utilized in the creative briefs or AI prompts.

Further, the five Suno tracks were filtered through Jammable to produce five additional tracks with a Taylor-esque vocal style — no changes were made beyond the lead vocal. All fifteen tracks (Taylor’s, Suno’s, Jammable’s) were blind-tested for appeal with over 200 US-based human respondents. We also ran the Suno and Jammable tracks through OnBrand to see how well they represented the emotional attributes in each brief. Listen to our Suno and Jammable tracks and explore OnBrand’s emotional analysis, here.

Musical Appeal
It turns out, AI can indeed write a hit song, so much so that two of the Suno/Jammable tracks outperformed Taylor in terms of appeal — Forever Young (71% appeal to So High School’s 38%) and Winter’s Embrace (65% appeal to Evermore’s 19%). Taylor’s three other songs: Red, Anti-Hero, and Cruel Summer all outperformed their Suno/Jammable counterparts by an average of 37%.

Our expectation was that the Jammable tracks (those with the Taylor-esque lead vocal) would have higher appeal than the Suno versions given the brand equity of Taylor’s vocal style. However, the results proved otherwise with the Jammable tracks being the least appealing for all but Winter’s Embrace, which begs the question, “Is the artist or the song more important to listeners?”

In August 2024, a SoundOut study of over 1,000 US consumers revealed that 54% felt that the music, regardless of singer, is most important when adding music to their playlists. Only 4% felt their musical choices were purely artist-based.

That said, 87% felt that getting to know the artists behind the music is an important part of their music enjoyment, and when asked if they would happily add quality AI-created music to their playlists, just over half (57%) agreed. This suggests that musical discovery is driven in large part by the music, with additional value generated by relating to the artists. The case holds for brands as well with 54% of respondents stating that use of AI music in ads would have no impact on their respect for the brand.

This, coupled with the results from our tests, demonstrates that when it comes to musical appeal, it is not as critical who creates it — good music is good music. While tying the music back to an artist’s personality, motivations, and stories is a value-add, it is not essential to the enjoyment of music.

“The majority of consumers already seem willing to embrace and consume AI music if the song is sufficiently appealing, and it therefore appears that the artist connection is potentially disposable. It is not that the artist is unimportant, just that the song is much more important.”

— David Courtier-Dutton, CEO, SoundOut

While these findings are not ideal for musical artists, the case is different for brand music. Sonic branding involves accurate emotional representation of brand values, personas, and creative briefs, and is not as easily replaced by AI.

Emotional Accuracy
While AI can write catchy tunes, it continues to be highly inaccurate when representing specific emotional attributes, keeping with our Human vs. Machine, Round 1 findings. SoundOut recommends a minimum accuracy score of 80% for a song to suitably represent a given set of emotions — as is the case when trying to find music that fits a brand’s personality and values. Despite being given the top six emotional attributes for each Taylor Swift song, AI was unable to produce an accuracy rating above 38% (Suno’s Broken But Dancing), and no higher than 10% for eight of the ten AI tracks. Had this been for a brand, none of the AI tracks would be deemed acceptable.

AI further demonstrated a lack of emotional complexity in the attributes it did rank for. Unlike Taylor’s songs, each of which had dozens of attributes above the 80% threshold, AI’s tracks only had one or two in that range. Not only that, all ten AI tracks scored highly for a similar set of attributes: Simple, Pure, Peaceful, and Tough, none of which were included in the given briefs — Simple scored above 90% for all AI tracks.

Is this some type of ghost in the machine or something else? More than likely it is a reflection of the technology’s adolescence, and perhaps provides us with the greatest window into gen AI’s current “emotional capacity.” Although its algorithm is capable of producing clever and appealing music, AI cannot with any regularity or purpose, express true human emotion in its creations, and certainly not with any depth, such as with compounded emotions.

“With sonic branding, our composers combine a great deal of analysis, artistry, and subjectivity to get the right musical feel for our clients. We’re not magicians, but having the ability to directly relate to the emotions being expressed goes a long way in the composing process. Our experience with AI is that while it is a great inspirational tool and can produce music we like, it has not demonstrated the emotional complexity to dependably produce impactful brand music.”

— Chad Cook, President of Creative & Marketing, SAM

Our Perspective
It would appear that the answer to the question of how you top a prolific songwriter like Taylor Swift at her own game, is to introduce an advanced computer algorithm into the contest. It is extraordinary how quickly AI has advanced in the music space, and with two AI tracks outperforming Taylor’s originals, this experiment confirms that AI can indeed write a hit song (with the right musical inputs and instruction). But is this the haymaker that knocks out human-produced music?

Overall, we found the creative process with AI to be unpredictable. Despite our detailed creative briefs for each Taylor Swift song, Suno often had its own interpretation of musical genres, styles, and musical terminology. Briefs were revised multiple times to assist the platform in achieving a desirable result, however, each click of the ‘regenerate’ button felt like a shot in the dark — no two results were alike. It was like communicating with a talented composer who doesn’t speak the same language.

Ultimately, we were able to produce some impressive (and effective) results, but not without a time-intensive process. It should also be noted that the output from Suno was not edited or enhanced in any way (other than swapping out the lead vocals to produce the Jammable versions). To be considered a final mix for our sonic branding clients, these pieces would require additional editing, including arranging, high-fidelity re-records of instruments and vocals, as well as mixing and mastering.

Considerations for Brands
Results from this study raise several considerations for brands when it comes to AI and sonic branding. Atop the list of concerns is emotional complexity. Brands are deeply dialed into the values and emotions they wish to present and how they want audiences to feel. While consumers might not judge a brand for using AI music, using music that expresses the proper vibe is critical, especially considering the multitude of brand channels that incorporate sound. With custom music needs, such as expressing a brand in a 2-3 second sonic logo or a longer-form promotional track, marketing teams are likely to gravitate towards a high-touch, human creative process — at least until the emotional accuracy of AI improves.

Secondly are copyright issues with AI source material, which currently has Suno, Udio, and major music labels engaged in a lawsuit. Many agencies and marketing departments are understandably weary of using any music that could potentially lead to copyright and usage issues down the line. Only time will tell how this issue shakes out, but ultimately it is not hard to imagine a generative AI platform emerging that is truly self-taught as a musician would be.

Finally, the promise of lower music production costs and a faster music selection process is attractive to brands, however, as we experienced, getting exactly what you want out of AI can be fiddly, requiring both time and musical knowledge to accurately accomplish. Will brands and/or agencies want to devote this time and energy? And for brands that spend millions creating a desired perception, is music — one of the most influential of the senses when it comes to emotion — the area to cut costs?

AI certainly has the potential to reshape the industry and our creative approach, however, humans are creative beings — musicians will not simply turn off their desire to create new music. Consider brands like IKEA and the furniture industry. While machines can produce attractive, efficient, and cost-effective furniture, there is still demand for custom pieces, and advancements in tools and processes enable craftsmen to create these more efficiently. It is our belief that the results from this study will encourage musicians, perhaps even Taylor Swift herself, to strive for new creative heights using AI’s capabilities as a catapult. AI is here to stay, and musicians, music agencies, and leading brands are best served by exploring how we can embrace it.

Want to hear more? For an in-depth analysis of Taylor Swift’s music and commercial appeal, check out the latest research from SoundOut: The Taylor Swift Index 2024.

---

Russell Boiarsky is Director/Brand Strategy at Stephen Arnold Music, The World Leader in Sonic Branding®, which has created audio branding for leading companies worldwide including CNN, Sony Interactive, Airbnb, The Weather Channel, ESPN and more.

Appendix

The following AI creative briefs were derived from Taylor Swift songs released from 2019 to 2024 and used for the “Human vs. Machine: Round 2” study.

• AI Brief: Cruel Summer (Lover, 2019)
  Lyrics: A song about summer love turned to heartbreak, vulnerable, resilient, progressive, direct, artistic,compelling, witty, independent
  Style: Female lead vocal, electro, pop, A major, upbeat, energetic, half time drum groove
• AI Brief: Evermore (evermore, 2020)
  Lyrics: A song about winter depression and deriving hope from someone special, sadness, loving, upper class, down-to-earth, clear, idealistic, luxurious
  Style: Female lead vocal, piano ballad, strings, Db major, soft, delicate, no drums
• AI Brief: Red (Red, Taylor’s Version, 2021)
  Lyrics: A song about heartbreak and infatuation, reminiscent, strength, fun-loving, movement, small-town, alive, upbeat, sparky
  Style: Female lead vocal, alt-folk pop, E major, 125 BPM, energetic, big rock chorus
• AI Brief: Anti-Hero (Midnights, 2022)
  Lyrics: A song about trying to be oneself amid societal judgment, vulnerable, introspective, enlightening, sensual, secure, articulate, idealistic, assertive
  Style: Female lead vocal, indie pop, E major, 97 BPM, energetic, synth accents, half time drum groove
• AI Brief: So High School (Tortured Poets Department, 2024)
  Lyrics: A song about high school love with a nostalgic 90’s feel, youthful, positive, playful, authentic, family-oriented, trusted, clear, unique, uplifting
  Style: Female lead vocal, 90’s shoegaze, G major, 88 BPM, energetic, feel good vibe