What AI Does To The Radio Spot

The radio spot has always been the quickest thing an agency makes. A script, a voice artist, a studio slot and a music bed, often turned round in a day for a retail offer or a city launch. Generative tools make it quicker still. A model can draft the script, translate it, generate a voice read and a backing track, and produce versions for twenty cities, each with its own store address, by the afternoon.
For a medium built on local frequency, that is genuinely useful. It also makes it very easy to sound like everyone else.
What AI makes cheap in a radio spot
Volume and versions. A retailer can now run a different spot for each city, each day of a sale and each language, without paying for each one separately. Station production teams, which make much of the local advertising on FM, can turn round small advertisers faster. A shop that never thought it could afford radio can have a spot.
The voice is part of this. We covered synthetic voice in audio advertising earlier: the generic read is now nearly free, while a recognisable voice has become more valuable. The same split is now happening to the whole spot.
Why cheap radio spots sound alike
Radio ad breaks are crowded. A listener in a car or a kitchen hears several spots in a row, half-attending. The ones that register have a voice, a sound, a joke or a jingle they know. The rest become a blur of offers and phone numbers.
Generated scripts drift towards that blur. Ask a model for a sale spot and it will write the most typical sale spot it can: urgency, a list of discounts, a deadline, a call to action. It is competent and forgettable, and when every advertiser uses similar tools, the whole break starts to sound like one long announcement.
On radio, sounding like everyone else is the same as not being on air.
Where the craft of the radio spot moves
Writing for the ear. One idea per spot. The brand name early and more than once. Sound effects and dialogue used to create a picture, because radio has no screen. These are old disciplines, and generated drafts rarely follow them unless a writer insists.
Owning a sonic identity. A consistent voice, a sound signature or a short musical phrase lets fifty versions still feel like one brand. Without it, versioning multiplies anonymity.
Listening, not reading. Every version should be heard before it airs, especially in languages the reviewer does not speak fluently, where a synthetic read can misplace stress or mispronounce a place name in a way that sounds careless to locals.
And consent for any voice that is cloned or recreated. The radio spot is about to be made in far greater numbers. The ones that work will still be the ones somebody actually wrote.
More from The AI Shift.