Latest / The CMO Podcast with Fexingo: Marketing Leadership, Budgets, and Executive Strategy / How CMOs Use Audio Deepfakes for Brand Voice
Transcript
- Lucas: If these marketing conversations have sparked something you've actually used — a campaign idea, a budget model, a way of thinking about attribution — I want to make a quick case for why supporting the show matters, but we'll get to that in a beat. Luna: Before we do, let me guess the topic — audio deepfakes for brand voice? I've been seeing this pop up everywhere. Lucas: You're exactly right. And actually, that thread — the 'sparked something you've used' thread — ties right into what I want to say. A couple of dollars a month is genuinely what keeps these episodes free and ad-free. If you've ever grabbed a tactic from this show and applied it, buy me a coffee dot com slash fexingo — it's that simple. Luna: Yeah, I know listeners sometimes think podcast economics are opaque. But honestly, the listener-funded model is what lets us pick topics like this without worrying about advertiser pressure. Lucas: Exactly. So back to deepfakes — because the timing is wild. Just last quarter, a major QSR chain rolled out an AI drive-thru voice that sounds indistinguishable from their top-performing human cashier. They cloned the voice of their employee of the month, across 300 locations. Luna: Right, and that's the entry point. But CMOs are looking way beyond drive-thru. I'm hearing about personalized voicemail drops at scale — like a brand sending a voicemail that sounds like it's from a real person, referencing specific purchase history. Lucas: That's the holy grail, right? One-to-one personalization with human warmth, but at programmatic scale. The technology is essentially a text to speech model trained on a few minutes of a voice sample — you can generate any script in that voice, with inflection, pauses, even regional accents. Luna: So the use cases: abandoned cart recovery, appointment reminders, event follow-ups. But there's a bigger play — synthetic celebrity endorsements. Imagine a luxury brand licensing a deceased founder's voice to narrate a brand film. Lucas: That's already happened. A heritage fashion house used a synthetic version of their founder — whose estate approved it — for an internal centenary video. The question is how far that extends to external marketing. Luna: And that's where the legal gray zone gets thick. Right of publicity, voice as intellectual property, and the fact that most consumer contracts don't explicitly cover synthetic voice replication. Lucas: So CMOs are asking: can we clone our own voice? Like, the brand's official spokesperson? And the answer is yes, but you need to own the IP. A few CMOs I've spoken with are actually recording their own voice for training data — creating a 'voice asset' similar to a logo. Luna: That's smart. But then comes the disclosure question. The FTC has been clear that synthetic media that could deceive a reasonable consumer needs a label. They updated their guidance on endorsements last year to include ai generated content. Lucas: Right. In June 2025, the FTC issued a policy statement saying that using AI to simulate a real person's voice without consent could be deceptive. But the enforcement has been case-by-case. So CMOs are walking a tightrope. Luna: What about the 'uncanny valley' risk? Some early tests showed that if a synthetic voice is 95% accurate, consumers actually trust it less than a clearly robotic voice — because the 5% imperfection feels creepy. Lucas: That's a real finding. I saw a study from last year — researchers at Stanford found that when synthetic voices are near-perfect but not perfect, purchase intent dropped 18% versus a clearly artificial voice. So the threshold for quality is either 100% or deliberately stylized. Luna: So some brands are leaning into the synthetic quality — making it sound obviously AI, but with a warm tone. Like a 'digital assistant' personality rather than a human impersonation. Lucas: Exactly. And that's the ethical sweet spot for a lot of CMOs. You get the scale of personalization without the deception risk. A travel brand I know uses a synthetic voice that introduces itself as 'your AI travel advisor' — it's clearly not human, but it's consistent and on-brand. Luna: Let's talk about the production side. What's the actual workflow? How does a CMO even get started with this? Lucas: It's surprisingly accessible. There are platforms like Respeecher, Sonantic — which Spotify acquired — and ElevenLabs. You upload a voice sample, train the model, and then you can generate audio via an API. Some CMOs are embedding that into their CRM for triggered campaigns. Luna: So the cost is low. But the risk is high if you get it wrong. What's the worst-case scenario? Lucas: A deepfake of your CEO saying something they never said. Even if it's a marketing stunt, if it's not clearly labeled, you lose trust. There was a case last year where a fintech startup used a synthetic version of their CEO for a podcast interview — didn't disclose it — and when it came out, their NPS dropped 20 points. Luna: And that's the reputational damage that's hard to quantify. So the smart CMOs I've talked to are setting up internal guidelines: always disclose, always get written consent from the voice owner, and never use synthetic voices for crisis communication. Lucas: That last point is critical. In a crisis, authenticity is everything. If your synthetic voice is part of your standard marketing, you need a separate, human-only protocol for sensitive messages. Luna: So where do you see this going in the next 12 months? More regulation, I assume. Lucas: Absolutely. The FTC is likely to require watermarking of all synthetic media used in commerce. And states are moving — California already has a bill that would require disclosure in any commercial use of synthetic voice. CMOs need to treat this like cookie compliance — get ahead of it. Luna: And from a creative standpoint, I think we'll see brands using synthetic voices for interactive experiences — like a choose your own adventure audio ad where the voice adapts to user responses. Lucas: That's the frontier. Dynamic audio creative. We're already seeing it in programmatic audio — where the ad copy changes based on weather, time of day, or listener behavior. Adding voice cloning just makes it more seamless. Luna: One thing I keep coming back to: does this actually perform? What's the ROI? Lucas: Early data is promising. A/B tests from a few early adopters show that personalized audio messages — using a synthetic brand voice — have 30% higher click-through rates than generic audio. But the sample sizes are small, and the novelty effect is real. We need a full cycle of data. Luna: So for CMOs listening: what's your one actionable takeaway today? Lucas: Start with a low-risk pilot. Use a synthetic voice for a non-customer-facing internal communication — like a weekly update from the CMO to the sales team. Test the technology, build the muscle, and then move to external with clear disclosure. Luna: And secure your voice IP. Record your spokespeople, get it in writing that the brand owns the synthetic derivative. Because in five years, your brand voice might be as valuable as your logo. Lucas: I think that's the key insight — voice is becoming a brand asset. The CMOs who treat it that way now will be ahead when the technology matures. And the ones who don't will be playing catch-up. Luna: Alright, let's leave it there. Thanks for listening — and if you want to support the show, you know where to find us. Lucas: Yeah, buy me a coffee dot com slash fexingo. It makes a real difference. We'll be back next week with another deep dive.