A radio station's phone line is the one place where listeners raise their hand. They call to win, to request, to argue, to tip off the newsroom. An AI voice agent for radio stations is worth buying only if it turns more of those raised hands into engagement you can measure and sponsorship you can sell, without putting your licence or your reputation at risk.
This guide is for programme directors, station managers and sales leaders at commercial stations and groups. It covers why the listener line deserves investment now, which formats to hand to AI and which to keep human, the rules to build in before anything goes on air, a pilot scorecard and the questions to ask any vendor.
Why the listener line matters more than it used to
Radio's reach problem is smaller than its revenue problem. Two sets of figures frame it.
- The audience is still there, especially in the car. Edison Research's Infinite Dial 2026, released in March 2026, found that 73% of Americans who drive listened to AM/FM radio in the car in the past month, against 48% for online audio and 37% for podcasts. Among drivers aged 55 and over, the figure was 81% (Edison Research, as reported by Radio Ink, March 13, 2026). The same study shows younger drivers moving fast: 73% of 18-to-34-year-old drivers listened to online audio in the car.
- Core ad revenue is flat without election money. BIA Advisory Services' April 2026 forecast puts US local over-the-air radio revenue at about $10.08 billion in 2026, up 1.8% on 2025, and radio digital at about $2.38 billion, up 5.0%. Strip out political spending and over-the-air revenue is projected to slip 0.3%, while digital grows about 2.0% (BIA, April 9, 2026; radio split reported by Radio Online and Radio Ink).
Read together, the message for a station is blunt. You still have the ears, but selling more spots against the same ears will not grow the business. Growth has to come from formats that make listeners participate, because participation is what a sponsor pays a premium for and what a streaming playlist cannot offer.
The phone line is where that participation already happens. It is also where it leaks. A contest cue can pull in far more calls than a producer can answer, and every caller who gets a busy signal learns that calling is pointless. That is the same capacity trap contact centres face, which we covered in the hidden cost of call overflow, with one difference: in radio, the overflow happens live, in front of the audience.
A listener who calls and never gets through is not a lost call. It is a lost habit.
Which formats an AI voice agent for radio stations should handle
Do not start from the technology. Start from the formats on your schedule, and rank each one by how many listeners it touches and how much damage a wrong answer would do on air. Here is a starting matrix.
| Format or call type | What the AI handles | What stays with your team | Risk if it goes wrong |
|---|---|---|---|
| Contest and game entries | Answers every call and message, reads the approved rules, records entries and eligibility answers, runs trivia rounds | Winner selection, prize disputes, rule changes | High: regulated, public |
| Song requests and dedications | Takes the request, checks it against the playlist, captures a short message for the host | What actually gets played | Low |
| Call-in show screening | Greets callers, captures name, town and point, sorts them into a queue for the producer | Who goes on air, and when | Medium: editorial |
| AI co-host segments | Joins the conversation when the host invites it, keeps talking with listeners by phone and WhatsApp off air | The host leads; the station approves scripts and can stop the format | High: brand and trust |
| Listener feedback and news tips | Records and tags the feedback, flags urgent news tips to the newsroom | Verification and any reply | Medium |
| Advertiser enquiries | Captures business name, budget window and campaign dates, books a call with sales | Pricing, packages, negotiation | Low, if no prices are quoted |
| Complaints, legal, on-air incidents | Takes details and routes immediately | Everything else | High: keep human |
Two rows deserve emphasis. Contest entries are the volume play: they turn a ringing switchboard into a list of engaged listeners. The AI co-host is the growth play: it lets a show talk to many more listeners than a host and a producer ever could, and it gives sales a new, sponsorable segment. It is also the format where getting the guardrails wrong costs the most.
Rules to build in before anything goes on air
Most of the risk in radio AI is not technical. It is about honesty with listeners and doing exactly what you said on air. Put each of these rules in the system's logic, not in a prompt the model may or may not follow. We explain why that distinction matters in Confidently Wrong. What follows is general information for US and EU stations, not legal advice.
The disclosure rule is not a formality. In April 2025, it emerged that CADA, a Sydney station owned by ARN, had run a four-hour weekday show with an AI-generated host called Thy for about six months without telling listeners (Global News; Mediaweek). The backlash came from listeners and from the voice-acting community, and it centred on one word: disclosure. If you are building trust in a new format, read our guide to building trust in AI interactions before you write the first script.
Listeners forgive an AI that says it is an AI. They do not forgive finding out.
A pilot scorecard that sales can use
Radio pilots fail when nobody agreed in advance what success looks like. Pick one show and one format, record a baseline for two to four weeks, then compare the same numbers during the pilot. Every input below comes from your own systems; there is no industry benchmark that fits every market, so do not borrow one.
| Metric | Where it comes from | Why it matters |
|---|---|---|
| Calls answered ÷ calls offered during the segment | Phone system call records | Shows how much participation you were losing to busy signals |
| WhatsApp messages per segment | Messaging logs | Captures listeners who will not call but will type |
| Unique and returning participants | Entries de-duplicated by phone number | Returning participants are the habit you are trying to build |
| Sponsor-ready moments | Entries in sponsored games, mentions read, opt-ins collected with consent | Turns engagement into something sales can package and price |
| Answer quality | Weekly transcript review by the producer | Share of conversations with no wrong, off-brand or off-rules answer |
| Team time | Producer estimate before and after | Minutes freed from screening and data entry |
Agree the go or no-go threshold before launch. For example: the format continues if answered calls and returning participants both rise and the transcript review finds no rule-breaking answer. Pricing the sponsorship is a sales decision based on your own rate card, and the pilot gives sales the participation data to defend it.
How a six-week pilot runs
Our interactive AI radio formats overview describes how we run radio pilots at Unicall. The structure is simple and it is the one we would recommend to any station, whoever the vendor.
- Weeks 1 and 2: choose and build. The station and the vendor choose one show and one format, create the voice and character for the station, and run internal tests. The station brings the show, the host and editorial sign-off.
- Weeks 3 to 6: on air and measured. The format goes live and the team tracks engagement: calls, WhatsApp messages and returning listeners, against the baseline.
- End of week 6: shared review. Both sides review the scorecard and decide whether to continue, adjust or stop. The station stays in control throughout and can stop the format at any time.
Six weeks is short enough to protect the schedule and long enough to see whether listeners come back, which is the number that matters.
Buyer checklist: what to ask any vendor
If you are comparing voice AI providers more broadly, our comparison hub applies the same ten buyer criteria to each vendor, using public information only.
FAQ
What does an AI voice agent do for a radio station?
It answers the listener line on every ring, takes contest entries and song requests, collects feedback and qualifies callers for the producer. Some stations also put it on air as an AI co-host that the human host invites into the conversation. Editorial decisions, winner selection and anything sensitive stay with the station's team.
Does an AI co-host replace the radio presenter?
It should not. The formats that work keep the human host in charge: the host decides when the AI joins the conversation, the station approves the scripts and rules, and the AI introduces itself as an AI. The value is extra capacity, such as talking to more listeners and running games all day, not a cheaper presenter.
Do radio contests run by AI still have to follow FCC contest rules?
Yes. In the US, 47 CFR 73.1216 requires a licensee to fully and accurately disclose a contest's material terms and to conduct it substantially as announced, on air or on the station's website. Automation does not change who is responsible. Keep the rules in one approved source that both the website and the AI agent read from, and keep winner selection out of the model. This is general information, not legal advice.
Should listeners be told they are talking to an AI?
Yes, every time. In the EU, Article 50 of the AI Act requires AI systems that interact directly with people to be designed so that people know they are dealing with an AI, an obligation that applies from 2 August 2026. Outside the EU, the reputational case is just as strong: in 2025 an Australian station faced a public backlash after running an AI-voiced host for about six months without telling listeners.
How should a radio station measure an AI voice pilot?
Set a baseline before launch and track the same engagement metrics during the pilot: calls answered versus calls offered, WhatsApp messages, unique and returning participants, and sponsor-ready moments such as entries collected for a sponsored game. Add a quality measure, for example the share of transcripts reviewed with no wrong or off-brand answer, and agree the go or no-go threshold before the pilot starts.
Unicall.ai builds and operates done-for-you AI voice agents, including interactive formats for radio stations, such as an AI co-host and phone and WhatsApp games, where your station stays in control of what goes on air.