An AI-generated voice fooled radio listeners about as often as it gave itself away, matching a professional voice actor in credibility, warmth, and appeal in a blind listening test. The real divide showed up only after listeners learned which voice they actually heard.
The study of 1,326 weekly radio listeners, conducted by Harker Bos Group’s Crowd React Media in May and June, tested two reads: a station imaging spot built around a gas card giveaway and a Mother’s Day call-to-action reminder. Each script was recorded in a human and an AI-generated version, with the human read performed by professional voice actor Neil Wilson.
Half of respondents, ages 18-45, heard the human version and half heard the AI version, with neither group told which they were getting. Statistical significance was tested using a two-proportion z-test at a 95% confidence interval. The finding lands amid a broader industry debate over AI voice that has largely run on anecdote rather than evidence.
Listeners were asked to guess which voice they had heard before anything was revealed. Fifty-nine percent of the group that heard the human voice correctly identified it as human, compared with 55% of the group that heard the AI voice, mistaking it for human as well. Neither result cleared the bar for statistical significance, meaning listeners were not reliably distinguishing AI from human on sound alone.
Across both clips, human and AI voices scored within a few percentage points of each other on professionalism, authenticity, credibility, energy, and likability, with none of the gaps reaching statistical significance. Overall appeal landed at 60% for the human voice and 61% for the AI voice on the first clip, essentially a coin flip.
Only one attribute produced a statistically significant gap. Listeners who heard the human voice were more likely to describe it as funny, 33% to 26% for the AI version, a seven-point difference. The clip’s punchline, built around a joke about an Optimus Prius, apparently lands differently coming from a human than from AI-generated speech, suggesting comedic timing remains an area where the technology has not fully closed the gap.
Crowd React Media Founder and Harker Bos Group Vice President of Strategy Katie Miller framed the takeaway as a question of disclosure, not audio quality.
Among those who had heard the human voice, 48% said they now viewed the clip more favorably knowing a real person was behind it, and only 4% felt worse. Among those who had heard the AI voice, just 25% felt more favorably after the reveal while 20% felt worse, with the majority reporting no change either way. All three gaps were statistically significant at the 95% confidence level. The performance had been identical. The reaction to the source was not.
Asked more broadly how they feel about AI-generated voices in advertising and media, 44% of the full sample responded positively, 26% negatively, and 30% neutrally. Narrowed to radio specifically, the numbers tilt against disclosure: 33% of listeners said learning a station uses AI voices would make them feel less favorably toward it, compared with 21% who said it would make them feel more favorably. Forty-seven percent said it would make no difference.
Listener comments collected after the reveal traced two distinct reactions. Those who heard the human voice described feeling validated once they learned a real person was behind it. “I wasn’t sure if it was AI at first, but now knowing it’s voiced by a live person, I’m all for it!” one respondent wrote. Another said, “If they stay human, I will keep listening!”
Reactions in the AI cell split further. Negative comments centered on trust rather than sound quality. “It feels like the station is lying to listeners and I do not want to listen anymore,” one listener wrote, while another called the practice “sacrilege to their listening audience.” But a portion of AI cell respondents reacted with interest rather than suspicion. “Knowing it was AI-generated impresses me because the voice quality sounded completely natural,” one wrote, calling the approach “innovative and cutting edge.” Others drew a line by use case, with one respondent saying AI works “for quick little blurbs or commercials” but crosses into unacceptable territory when applied to “a non-air personality.”
The divergence sets up a live case study within radio itself. iHeartMedia has made a public selling point of banning AI-generated on-air talent and content companywide under its “Guaranteed Human” campaign and citing internal research showing 90% of consumers want media created by real humans.
Other broadcasters are moving to embrace the tech. Starting in 2025, Entravision is pairing Jose 97.5 (KLYY) Los Angeles host Geraldine “GeeGee” Guzman with Coyotec, an AI-generated co-host the company brands as the industry’s first Latino AI-powered radio personality, and crediting the pairing with a 75% jump in weekly cume among Hispanic men 25 to 54 between March and April of this year.
