Latest / Business Podcasting: Production Strategy / Should your podcast use an AI voice?

Transcript

Should your podcast use an AI voice? · Business Podcasting: Production Strategy

• 0:00
  [Synthetic version of Lam Hoang's voice]
• 0:07
  This is Business Podcasting: Production Strategy, produced by WorkHacker.
• 0:13
  I'm your host, Lam Hoang, co-founder of WorkHacker.
• 0:16
  You're hearing a synthetic version of my voice.
• 0:19
  Each episode I take one question about making a podcast for your business,
• 0:23
  tell you what I would do and how solid the evidence is.
• 0:32
  Today's question: should your podcast use an AI voice?
• 0:36
  Two in three weekly podcast consumers would react badly if
• 0:40
  their favorite host started using an AI version of themselves.
• 0:44
  That's Edison Research's question from early 2026.
• 0:47
  It's about a host people already know.
• 0:50
  So does your podcast need one?
• 0:52
  No, it doesn't.
• 0:54
  Already make a show?
• 0:55
  If you speak well and you have the time, keep recording it the traditional way.
• 1:00
  If you have something to say and you're not making
• 1:02
  a podcast yet, a synthetic voice is a great option.
• 1:05
  It cuts the time it takes to put a show together by a lot.
• 1:09
  Interviews and co-hosted chats are the exception: keep recording those.
• 1:13
  Even if you record yourself, a clone can fix small spots.
• 1:17
  It can do voiceovers too.
• 1:19
  And if you do use one, make it a clone of your own voice.
• 1:22
  Say so up front.
• 1:23
  And have a person listen to each episode first.
• 1:26
  Here's what you'll leave with.
• 1:28
  Which of those cases fits you, and whether listeners can even tell.
• 1:32
  Then how to disclose it, how to make a good clone and what stays your job.
• 1:37
  First: which case is yours?
• 1:39
  Already have listeners who know your voice?
• 1:41
  That two in three from the start was about a favorite host changing.
• 1:45
  The cost is real.
• 1:47
  But I think it falls hardest on shows people love.
• 1:50
  Nobody has measured a new show that's open about its synthetic voice from day one.
• 1:55
  No show yet?
• 1:56
  Then your question is time.
• 1:58
  In the episode on starting a podcast,
• 2:00
  the test was timing one episode from start to finish.
• 2:03
  Do you have those hours every week?
• 2:05
  If not, a synthetic voice changes the answer.
• 2:08
  It takes recording sessions and retakes off your week,
• 2:11
  but the writing and the listening stay on your list.
• 2:14
  No study has counted the hours it saves.
• 2:17
  And which parts suit a clone?
• 2:19
  It fits best where you write every word.
• 2:22
  Think of a solo show, or a scripted segment.
• 2:25
  In an interview or a chat with a co-host, the live exchange is the show.
• 2:29
  Even there, a clone can handle fixes.
• 2:32
  At least one audio editor lets you type a
• 2:34
  corrected word and patch it in with your clone.
• 2:37
  That one is English only, so check yours.
• 2:40
  Second: will your listeners even notice?
• 2:43
  You're hearing one right now.
• 2:45
  Judge for yourself.
• 2:46
  And in short clips, many people can't tell.
• 2:49
  Crowd React Media ran a blind radio test in May and June 2026.
• 2:54
  More than half who heard the AI voice took it for a human one.
• 2:58
  Listeners rated the two voices about the same on credibility and likability.
• 3:02
  The human voice won on just one thing: it sounded funnier.
• 3:06
  Those were short radio spots, not podcasts.
• 3:09
  Told it was AI, most felt nothing either way.
• 3:12
  And the rest?
• 3:13
  One in five felt worse, but slightly more felt better.
• 3:17
  The researchers' verdict: knowing mattered more than the sound.
• 3:21
  And finding out later can break trust.
• 3:24
  In Edison's Evolving Ear study of 2025 data,
• 3:27
  some listeners said they'd feel cheated or robbed.
• 3:30
  That's if they learned a host they thought was human was AI.
• 3:34
  Third: how do you tell them?
• 3:36
  Start with what Apple requires.
• 3:38
  As of October 2026, its rules name synthetic voices outright.
• 3:43
  Say it in the audio.
• 3:44
  And in the episode and show details.
• 3:47
  Where in the audio?
• 3:48
  At the start, in one plain line.
• 3:51
  Not saying it up front is worse.
• 3:53
  Anyone listening has already given the show a chance.
• 3:56
  So say it, and move on.
• 3:58
  What should the line say?
• 4:00
  A 2024 study tested AI labels on news headlines with about five thousand people.
• 4:06
  People trusted the labeled headlines a little less.
• 4:09
  They assumed the AI had written them entirely, with no person supervising.
• 4:14
  Told the AI had only polished the wording, people dropped the penalty.
• 4:18
  Told it picked the topic and wrote it all, they kept it.
• 4:22
  That was text, not audio.
• 4:24
  So I'd say what the AI does and who's in charge.
• 4:28
  If your clone reads a script you wrote, say exactly that.
• 4:32
  Fourth: if you go ahead, how do you make a good one?
• 4:35
  We'll cover AI workflows in depth in future episodes.
• 4:39
  This one is about the voice.
• 4:41
  The providers' rule comes first: clone your own voice,
• 4:45
  or one you have explicit permission to use.
• 4:47
  Then comes the recording you train it on.
• 4:50
  One of the big voice companies publishes a guide to making a professional clone.
• 4:54
  It says the clone copies everything it hears.
• 4:57
  Noise and echo included.
• 4:59
  So use the best microphone you can, in your quietest room.
• 5:03
  Keep the same setup and the same distance throughout.
• 5:06
  What about your delivery?
• 5:07
  Providers warn that a flat reading makes a flat clone.
• 5:11
  So give it the delivery you want back.
• 5:14
  How much audio?
• 5:15
  For a professional clone, that company's minimum is half an hour.
• 5:19
  It recommends two to three hours.
• 5:21
  A quick clone needs far less, but it's the less faithful kind.
• 5:25
  And the cost?
• 5:26
  Start with a monthly subscription.
• 5:28
  The first paid tier is enough to begin.
• 5:31
  But one major provider puts its professional clone a tier up.
• 5:34
  Free plans generally won't cover a business show,
• 5:37
  because what you make on them can't be used commercially.
• 5:40
  And if you outgrow the first tier, you're probably getting a lot of value from it.
• 5:45
  Next, the model.
• 5:47
  Stay current with the one you're using.
• 5:49
  At some providers, moving your clone onto a new model is a step you take yourself.
• 5:54
  One major provider released a new model in late September 2026.
• 5:59
  It warns your clone may sound different on it.
• 6:02
  So after any switch, listen again before you publish.
• 6:06
  And keep your raw voice recordings.
• 6:08
  They're how you train the next model.
• 6:10
  At one major provider, you can't export a clone at all.
• 6:14
  Its advice: keep the original audio.
• 6:17
  Companies change course, too.
• 6:19
  One well-known voice company had shut down by early January 2026.
• 6:24
  Another had stopped selling voice AI to new customers by mid-August 2026.
• 6:29
  Last, what still has to be you?
• 6:32
  The voice is the easy part.
• 6:34
  In 2025 the Podnews editor James Cridland put
• 6:38
  the problem on the words rather than the voice.
• 6:41
  Were they checked by a person?
• 6:42
  His own example: if he had a cold, he might use his AI clone.
• 6:47
  It would still be his own voice and his own words.
• 6:50
  The biggest thing is a person in the loop.
• 6:53
  Someone listens to every episode and makes sure every citation is real.
• 6:57
  Build a workflow that makes that check happen.
• 7:00
  A second AI hunting for mistakes helps too.
• 7:03
  In April 2026 an AI-made show put out an
• 7:06
  episode that ended with the AI tool's refusal.
• 7:09
  Its last words were "I'm sorry, I can't assist with that."
• 7:13
  It came from one company's mass-produced shows.
• 7:16
  A person listening would have caught it.
• 7:18
  Make every render a good one.
• 7:21
  One provider says to listen hardest for names and specialist terms.
• 7:25
  In October 2026 one major provider updated its terms.
• 7:29
  Reviewing its output is now your job.
• 7:32
  Some things stay yours, however good the tools get.
• 7:36
  Your show's positioning, its direction and its taste.
• 7:40
  The subtleties AI doesn't know to pick up on.
• 7:44
  Without your context and shaping, AI gives you its default style.
• 7:48
  That's close to what it gives everyone.
• 7:51
  And a cheaper voice doesn't excuse a thin script.
• 7:55
  So here's the rule.
• 7:56
  Already record well and have the time?
• 7:59
  Keep recording, and let a clone handle fixes.
• 8:02
  Something to say and no show yet?
• 8:05
  A synthetic version of your own voice can get it made.
• 8:08
  Either way, disclose it at the start and have someone check every episode.
• 8:13
  Your next step is a small test.
• 8:16
  Pay for one month, and make a quick clone from a couple of minutes of your voice.
• 8:20
  Then have it read a page you wrote, and listen.
• 8:24
  A free prompt for your own AI to check your
• 8:26
  voice setup comes with the written version.
• 8:29
  That's the episode.
• 8:30
  Thanks for listening.
• 8:32
  The written version is on workhacker.com, with every source.
• 8:36
  And if you want a podcast like this for your business, WorkHacker makes them.
• 8:40
  See you next time.