New research shows that radio listeners are having a harder time telling real human voices from AI creations. But the survey also finds a few shortcomings in new artificial voice technology.
A study from research company Crowd React Media tested AI voice technology with approximately 1,326 weekly U.S. radio listeners ages 18–45.
In the test case scenario, the researchers played identical station promos voiced either by a professional voice actor or an AI-generated voice.
Katie Miller, founder of Crowd React Media and Vice President of Strategy at Harker Bos Group, said half the respondents heard the human version, and half heard the AI version, without knowing which they were listening to.
Her conclusion? She said most participants couldn’t tell the difference between real and artificial.
“People who heard the human voice were only slightly more likely to correctly identify it as human,” Miller explained. “People who heard the AI voice were slightly more likely to think it was AI.”
However, she said the margins were small enough that neither group was really identifying the voice correctly. “The AI voice passed the blind test.”
The test criteria included a station imaging spot with a gas card giveaway promotion, and a short call-to-action reminding listeners to call their mothers on Mother’s Day. The human voiceover was performed by a professional voice actor.
You can take the test for yourself here.
The overall appeal of the of the AI and human clips was essentially tied, with the AI version actually edging the human voice by one percentage point in one of the clips, according to the survey’s results.
The only area where the human voice showed a measurable advantage was humor, Miller said. A promotional script built around a joke with a human voice resulted in 33% of listeners describing it as funny compared to only a 26% funny rating for the AI version.
Perception is reality
Crowd React Media, which conducts media research for several major radio groups in the Top 50 markets in both English and Spanish-language radio, found human and AI voices scored comparably on professionalism, authenticity, credibility, energy and likability.
However, people’s perception of the AI generated voices changed once they were told which was which, Miller said.
When human voice listeners found out they’d been listening to a real person, according to the study, 48% felt more favorably about what they’d heard. Only 4% felt worse.
When AI voice listeners found out they’d been listening to an AI, 20% felt worse about it. Only 25% felt better, and a majority felt nothing either way.
Miller, who authored the report, said that’s a 23-point gap in positive lift and a 16-point gap in negative reaction to AI generated voices.
[Related: “Time for Your “AI or Human?” Radio Announcement Quiz”]
Participants were also asked a more broad question about how they felt about AI-generated voices in advertising and media in general. The results showed 44% are positive, 26% were negative and 30% as neutral.
Miller said that overall, the study has a more favorable split than you might expect “given the volume of (radio) industry hand-wringing,” but the negative segment is real and shouldn’t be dismissed.
Trust issues become a bit more evident when panelists were asked would knowing a station uses AI voices change how they feel about it. 33% said they would feel less favorable knowing so with only 21% feeling more favorable. Another 47% said it would make no difference.
Trust and transparency
In her conclusions about the survey, Miller made a list of what the results mean for radio:
- AI voice is technically competitive. In a blind listen, AI performs as well as a human voice actor on almost every measurable dimension. Stations and advertisers worried that AI voices will sound cheap or robotic may be underestimating what today’s technology can do, according to Miller.
- Disclosure is the real variable. The performance gap is negligible. The perception gap after disclosure, Miller said, is substantial. “If you’re using AI voice in production and listeners find out, whether through your own disclosure or through the wrong kind of attention, you’re carrying asymmetric risk,” she said.
More people will feel worse than will feel better.” - Humor is where human voice still has a measurable edge. For imaging built around a comedic premise, the data suggests to keep the human in the booth, Miller said.
- The most negative reactions are about trust and transparency, not sound quality. The verbatims aren’t saying “it sounded fake,” Miller explained. They’re saying “it felt like lying.” That’s a different problem, according to Miller, and it points toward how stations approach disclosure, not just whether to use AI.
- Credibility is worth pricing in. According to Miller: “Yes, hiring a human voice actor costs money. But the post-reveal data shows that credibility with your audience has real value too, and it can erode fast once listeners feel deceived. Before cutting the budget on voiceover, the question worth asking is: at what point does the cost savings stop being worth the credibility risk?”
Miller said that Crowd React Media does not have any AI technology development companies for clients.
