“It read like the sort of thing that I should be implacably against,” wrote industry observer James Cridland about Futuri Media’s RadioGPT, which uses AI to generate customized content for radio stations. “Irritatingly, though: It’s rather good.”
“I really wish the new RadioGPT was awful,” wrote broadcaster and music historian Alan Cross. His panicky-but-admiring response to hearing it was “oh s***”; an air personality I spoke to in the last week said something more easily reprinted when he told me, “I was hoping I’d have a few more years in the business before this happened.” In general, the RadioGPT demo has been far more earthshaking for radio people than Spotify’s recently reviewed addition of an AI announcer.
Futuri isn’t positioning RadioGPT as a replacement for live announcers, but rather against the reality that much of radio is unhosted or minimally hosted already. (On an NAB podcast this week, Futuri’s Daniel Anstandig estimates that more than 50% of radio shows are not live.) “With RadioGPT, there should never be a ‘liner-card’ or ‘sweeper-only’ air shift again. Now everyone can be live and local,” declared last week’s product announcement.
In my overall radio listening, I’ve been noticing radio that sounded AI-aided for at least 6-9 months. The most obvious uses have been in short pieces — weather, traffic, song tags. But I’ve also heard at least one station doing brief topical breaks (of the sort RadioGPT offers) that sounded like they were AI-voiced. I’ve heard several newscasts as well, although one PD assured me that I’d heard a human who was just kind of droning. As with voice-tracking, especially at the outset, the best AI-generated breaks are the ones that you don’t recognize as such, and I’m sure I’ve heard those, too.
I’ve recently expressed some concerns, as ChatGPT took hold in pop culture this winter, about how broadcasters will use AI, particularly the potential to institutionalize generic radio, rather than repair it. With AI for radio now on the market, here are some further questions for the entire industry, prompted by a newly available AI product, but related to the larger usage of AI at radio.
Over the last 20 years, owners have rarely improved radio through economies of scale. How can this be different? I believe that many broadcasters over the last two decades have been sincere in their belief that voice-tracking offered better local radio to smaller markets, not just cost savings. It’s a vision that I’m open to, but don’t often hear. Has national contesting kept radio competitive with Powerball? I still hear a lot more discussions about buying tickets on the way home than listening for hourly code words. The hardest-to-deny example of national radio’s greater clout is the success of certain morning shows, and, ironically, those don’t always offer cost savings on the local level.
To some extent, the excitement around RadioGPT only proves that voice-tracking hasn’t allowed every broadcaster to create a viable local station with seven major-market talents. If broadcasters are sincere in their belief that AI can create better local radio — Anstandig suggests using it to put local talent back in the community, for instance, rather than bogging them down with five other jock shifts — then we need to see that from owners and managers, not just further consolidation.
What is proper disclosure? The RadioGPT demo is upfront about being entirely AI-voiced. That’s what it’s selling. Those somewhat robotic newscasts and weather reports I’ve heard — if they were AI-generated — were not disclosed during my listening. What will disclosure sound like when voice-tracking (“some programs may be pre-recorded”) and even paid spins or other label arrangements (“promotional consideration provided by”) are only briefly and infrequently acknowledged.
The disclosure issue is particularly concerning for newscasts, but there’s also a lot of overlap between brief topical breaks of the sort that constitutes so much of radio’s content now and what might be heard on a newscast. Is the identical “stupid criminal” or “researchers in Denmark” story different if delivered during a newscast or over an intro? Often, it’s shared on a morning show by the host and a morning co-host who also doubles as newsperson.
RadioGPT is an extension of Futuri’s TopicPulse service which, as Cridland notes, is not untouched by human hands. The role of humans in helping create that content makes the use of AI voices not that different from the role of text-to-speech/speech-to-text that we’ve already accepted in our daily lives. As other groups and other vendors provide similar products, that will not always be the case and should be addressed now, including a protocol for human-sounding air talent names (or website bios) and live-sounding endorsements.
Whose voice is it anyway? ChatGPT offers stations the option of using an existing air personality voice. Who makes the ultimate decisions about what is said under a personality’s name and persona? When a host leaves a radio station, there are already questions about the use of their voice on spots or imaging. What must those safeguards be now?
What does fairness between formats look like? Hip-Hop/R&B and Adult R&B radio were the first formats dominated by national personalities. They were the stations most noticeably impacted by PPM-era restrictions on jocks. Hip-Hop radio in particular has never returned to dominance despite the genre’s current impact on the musical landscape. In a format once defined by personality, I am surprised by how many largely unhosted Hip-Hop and R&B stations I hear even in some large markets. AI has the potential both to both address that and exacerbate issues between stations in the same cluster. And in music, there have already been questions about who has input into the content process.
Even if you are open to AI-generated hosting, what jobs should still be filled in person in real time? Nearly three years ago, I wrote that music radio should consider hiring newscasters again. I was responding to the music-radio hosts I heard struggling to discuss a world on fire. I’ve been happy to hear a little more news on music stations, but those newscasts have been mostly a new wrinkle on “rip-and-read,” a pejorative we stopped using only because most music stations eliminated their news. It’s also easy to see unverified AI-generated news becoming an easy target in our age of disinformation.
One of the best possible scenarios for the use of AI at radio is if any cost savings make human-generated news more viable and give radio more opportunity to pursue enterprise news again. The same goes for the prospect of radio again being hosted in overnight — something I consider a positive, but also something that reopens the likelihood of callers who need help. Is that a shift that requires a human?
The purpose in asking them here is not to pose them of any one person, but to all of us. When this story was first published on March 2, I called for the discussion to continue at Country Radio Seminar in Nashville this month, as well as all other upcoming industry conventions. The panel “Let’s Chat: GPT and A.I. DJs” has just been added to the CRS schedule, while Anstandig has been announced as a keynoter for the upcoming Radio Days North America.
Broadcasters, meanwhile, need to make sure they’re making the most of whatever resources they have. As we worry about chatbots taking on sentient personalities, we need to ask if our radio stations have those as well. That discussion will continue in this column.
This story was update March 9 to update the CRS info as well as pose the additional “whose voice” question.
I’ve listened to the demo and in spite of doing things “a human DJ can do- only better” (obviously aimed at the money saving CEO), it doesn’t really seem like a listener oriented service. News items delivered over song intros. Artist information delivered like an encyclopedia article. Where’s the connection with the music? The listener? The station? Much of the content delivered these days on radio suffers from the same issues – it’s not connected with the station, the music or the listener. There are “punchlines” delivered at the end of the news stories, who created those? I would suggest that jokes are NOT created by AI. Just a guess. Human involvement seems to be somewhere in the background. I’m just sayin’.
There should be no doubt that broadcasters are going to use this and use it alot because the focus has been for at least a decade on providing adequate content at the lowest immediate financial cost – and ignoring the longterm costs of losing listeners to other media that provide more than adequate content.
Some day we’ll see a radio version of Kitchen Nightmares where a radio version of Gordon Ramsey does the equivalent of chastising the owners of a failing restaurant for selling reheated frozen foodservice commissary items rather than providing an experience and making it fresh.