How an AI commercial actually works (the spot, not the celebrity clone)

The search box sold you a celebrity

Type commercial voice over, radio commercial, or 30 second commercial and the pages that come back are mostly the same product: clone the announcer from last year's campaign, or pick from a wall of voices tagged ads. Those tools are not wrong about their own job. They are incomplete if the next thing you actually need is a timed take you can still rewrite on Thursday.

A spot is copy plus a clock. Fifteen seconds. Thirty. A disclaimer you did not want and a price the legal team changed at 4 p.m. Collapsing that into a celebrity soundalike is convenient until the offer changes, the station asks for a warmer read, or your counsel asks who that voice is supposed to be.

The product page for this job is Advertising. This post is the craft version: the three jobs hiding under one search, what we return as a file, and what the FTC actually says. How to write the spoken column already lives in how to write a voiceover script. Mid-rolls on a recorded show live in how an AI podcast actually works.

Three jobs, not one generate button

People type the same phrase for three products. Mixing them up is how a 15-second radio bed gets written like a testimonial, or a legal rewrite gets treated as a talent booking.

A 15-second spot is still a script

The Advertising worked example is a 15-second radio spot: three spoken lines, a cue on the first two, a plain close. House math on that page is about 250 characters for fifteen seconds. Treat it as an estimate. Speed varies by voice. If the legal line has to land in two seconds, generate that line as its own file.

Write for the ear. If a sentence is hard to say in one breath, cut it before you generate. We already wrote that fight in how to write a voiceover script. This post assumes you have a spoken column, not a brand manifesto pasted into the box.

Split the spot the way a session would split. Hook. Offer. Disclaimer. Close. One 40-second generate is how you get a take you cannot patch when legal changes the last clause.

Test the read before the booking

The expensive moment in a spot is usually the argument about the read, held after someone already booked. The cheap moment is the afternoon the copy lands. Render the same line in five voices. Play them in the room. Let the take shape the brief.

That is the whole point of the Advertising page: many reads of one line, on a flat allowance, with commercial rights on the export. You are not auditioning a person. You are auditioning a file. If none of them land, the copy is the problem, not the roster.

When the offer changes at 4 p.m., regenerate the line. Do not wait for a booth. The file that ships is the file you can still rewrite.

Same line, two engines

The usual failure is directing a draft, or drafting a keeper.

Write the words on Kokoro until they are actually the words. It is the unlimited draft engine: a clean, plain read, fast enough that rewriting a price is cheaper than arguing about the take. When a line has to sound like a person leaning in, move that line onto the engine that matches the job.

For a spot that needs a cue in brackets, Gemini Flash is the engine on this roster that acts [warm] or [confident]. The Advertising worked example is that split: same three lines, different cues, client hears the difference instead of imagining it. The short definition of a cue is what is a TTS cue. The longer argument is cued TTS versus flat reads.

A hook that needs a persona more than a cue can start on Grok Voice: five English personas, a confident characterful take. MAI Voice 2 is the roster member with real style and speed controls if the station wants the same voice a notch slower. Zonos holds American and British voices when the market is the accent, not the emotion.

Beds under a spot are a second export. Generate the voice here, generate a quiet instrumental in Sound and Music, mix them in the editor you already use. Duck the music under the voice. Do not ask the music model for the announcer.

The FTC pages, not the rumor

Two different rules get mashed into one internet sentence. An endorsement is a claim a listener might treat as someone else's opinion. A roster announcer reading copy you wrote is a read. They are not the same form.

As of November 2026, the FTC's Endorsement Guides still say endorsements have to be honest, and the plain-language FAQ is the page most shops actually mean when they say the Guides. A generated announcer on a script your team wrote is not an endorsement by a customer. We are not the FTC. Read their page. Verify before you rely.

As of November 2026, the FTC Consumer Reviews and Testimonials Rule (in force since October 21, 2024) bans reviews and testimonials that pretend to come from someone who does not exist, and names AI-generated fake reviews as an example. The same Q&A says there is no blanket ban on AI avatars in marketing. A fake testimonial is still a fake testimonial, whether a person typed it or a model did. Do not generate a customer who never used the product. Acceptable use draws the same line on impersonation. Same caveat: the live page wins.

We will not help you imitate a competitor's announcer, a celebrity, or any real person without consent. Voice cloning is not a live path here. When it ships it will need an attestation. Advertising that needs a real person's voice still needs that person.

Where it still fails

A fluent take can still be the wrong file. These are the misses we see when people treat the search phrase as a spec.

• Writing the spot like a landing page, then blaming the voice for running long.
• One 40-second generate for a cut that needed a hook file and a disclaimer file.
• A disclaimer with no pause, then blaming the engine for eating the legal line.
• Cues on a plain-read engine, then blaming the voice for a flat take.
• Asking the studio to mix the bed, book the station, or traffic the cart.
• A clone of a real announcer or a fake customer (we do not offer that, and acceptable use would not either).

Honest limits

We generate files. We do not sit inside Pro Tools, Adobe Audition, or a station traffic system. The output here is audio you own: commercial use, no watermark, same as the rest of the studio (you own what you make here).

No live clone path. No in-app DAW. No cartwall. No caption file. The console API tab is still labeled soon; we will not pretend a public render endpoint exists.

Cue-following is Gemini Flash only. The other engines on the roster read plainly. We strip brackets so they do not speak the direction. We do not pretend they performed it.

A generation has a character cap. Split a long campaign. Do not paste a two-minute manifesto into one box and call the truncation a voice problem.

Premium generation sits inside a flat monthly allowance (published free-tier premium budget: 10,000 characters, about 10 minutes by house math). Kokoro drafting stays unlimited so you can rehearse a price change without burning keeper budget on every micro-edit. Paid plans raise the flat allowance; they do not put you on a per-character anxiety meter.

Voice cloning is not a live path today. When it ships it will be English-first with a consent attestation. Until then, roster voices and personas are the honest path.

Cut one this week

The shortest proof is a spot you already owe, not a roundup of vendors.

• Pick a 15-second radio bed or a 30-second digital. Count it.
• Write the spoken column only. If a sentence is hard to say, cut it.
• Draft the words on Kokoro until they are the words.
• Move the keeper onto Gemini Flash if the line needs a cue, or Grok Voice if it needs a persona.
• Export two or three reads. Play them in the room. Pick one.
• If legal changes the last clause, regenerate that clause. Do not stretch the take.
• Listen once without looking at the page. If a word is wrong, fix the page, not the voice picker.

Open the studio, not the clone

If you came here from a commercial voice over or radio commercial search, the honest product is a spoken file you can hear this afternoon. Open Advertising, or start in Text to Speech if you already have the lines.

When you want the full studio, create a free account. Written to be heard includes the timed spot, as long as we do not lie about what that file is for.