Latest / Business Podcasting: Production Strategy / Which podcast production steps can AI do well?

Transcript

Which podcast production steps can AI do well? · Business Podcasting: Production Strategy

• 0:00
  [Synthetic version of Lam Hoang's voice]
• 0:07
  This is Business Podcasting: Production Strategy, produced by WorkHacker.
• 0:12
  I'm your host, Lam Hoang, co-founder of WorkHacker.
• 0:16
  You're hearing a synthetic version of my voice.
• 0:18
  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: which podcast production steps can AI do well?
• 0:37
  If your show is in English on Apple Podcasts,
• 0:39
  each new episode already gets an automatic transcript.
• 0:43
  Spotify does the same once it counts your show as eligible.
• 0:46
  So which steps does AI do well?
• 0:49
  As of this fall, a lot of them.
• 0:51
  It makes transcripts and chapters, cleans up the audio and cuts the ums.
• 0:56
  It drafts show notes and first scripts, and it finds research leads.
• 1:00
  It can even voice the show, as the episode on AI voices covered.
• 1:04
  But every step needs your judgment.
• 1:07
  Two tests this year found links that likely
• 1:10
  never existed, and claims no source made.
• 1:13
  So check its drafts against their sources, and judge the audio by ear.
• 1:17
  What the show is for, its point of view and the final listen stay yours.
• 1:22
  Here's what you'll leave with.
• 1:24
  What to check in the steps already done for you, and in the editing.
• 1:28
  Then how far to trust AI's drafts, whether it saves time and what you keep.
• 1:34
  First: what do the platforms already do for you?
• 1:37
  As of October, Apple also makes chapters for shows in English.
• 1:41
  You can switch them off, or supply your own.
• 1:44
  Spotify adds chapters for eligible shows too.
• 1:46
  For these steps, the job is checking theirs.
• 1:49
  How good are those transcripts?
• 1:51
  A 2026 benchmark tested transcription systems on long recordings.
• 1:56
  The best one still got about one word in fourteen wrong.
• 2:00
  Those were interviews and talks, not podcasts.
• 2:03
  And the ums weren't counted.
• 2:05
  So read a transcript as a first draft.
• 2:07
  Look hardest at the names.
• 2:09
  Apple asks you to put host and guest names in the episode description.
• 2:14
  That way its transcripts spell them right.
• 2:16
  Early this year, researchers tested speech-to-text
• 2:19
  systems on recordings of people saying street names.
• 2:22
  Averaged across the systems, the error rate was forty-four percent.
• 2:27
  Those were single names said alone,
• 2:29
  by American speakers from many language backgrounds.
• 2:32
  In the same study, a little extra training helped a lot.
• 2:36
  That was for speakers whose first language isn't English.
• 2:39
  YouTube dubs eligible videos into other languages by default.
• 2:43
  Its help page warns that dubs can trip on names, idioms or jargon.
• 2:48
  You can't edit a dub.
• 2:49
  But you can hold dubs for review before they go out.
• 2:53
  Second: can AI clean up your audio and cut the ums?
• 2:57
  One documented case shows both sides.
• 3:00
  A New Zealand university producer made a documentary podcast with AI tools.
• 3:04
  His study came out this year.
• 3:06
  Three of his interviews had background noise.
• 3:09
  An AI noise remover improved them.
• 3:11
  But the cleaned audio still lacked the depth
• 3:14
  and richness of professional vocal recordings.
• 3:16
  His best results paired the tool with his own engineering know-how.
• 3:20
  Cutting the ums has its own risk: clipping the words around them.
• 3:25
  One editing tool's own help page shows it.
• 3:27
  Its setting skips any filler it can't cut cleanly.
• 3:31
  It also lets you review each cut before you make it.
• 3:34
  Listen to the cleaned audio before you publish.
• 3:37
  Then listen at every cut.
• 3:39
  AI clip pickers have no independent test either.
• 3:42
  So check their picks.
• 3:44
  Third: can AI handle the research, the show notes and the script?
• 3:48
  The drafts read well, and that's the catch.
• 3:51
  A 2026 study had AI turn documents into podcast scripts.
• 3:56
  The models it tested often added things the documents didn't say.
• 4:00
  A separate check and repair step made them more faithful.
• 4:03
  Show notes from your transcript are that kind of draft.
• 4:06
  Research has the same weak spot.
• 4:08
  Another 2026 test checked the links that AI research tools cite.
• 4:13
  Three to thirteen percent likely never existed.
• 4:16
  With a link checker, the models cut broken links to almost none.
• 4:20
  How well depended on the model.
• 4:22
  And a working link still has to say what the AI claims it says.
• 4:26
  Use a chatbot for first leads in unfamiliar fields, not as your main research tool.
• 4:31
  Every factual claim needs independent checking.
• 4:34
  For a script, work back and forth with it.
• 4:37
  Even then, keep the editorial calls yourself.
• 4:40
  Your listeners may be relaxed about AI research.
• 4:43
  Edison Research's yearly report asked American
• 4:46
  weekly podcast consumers in early 2026.
• 4:49
  About seven in ten approved of AI doing a show's research.
• 4:53
  For AI-written scripts, it was about half.
• 4:56
  That's approval of an idea, not a test of the work.
• 4:59
  Most also see AI as a threat to podcasts' credibility.
• 5:03
  Whatever listeners accept, check every draft against its source.
• 5:07
  Editing, transcripts and research each get their own episode.
• 5:11
  Fourth: will any of this save you time?
• 5:15
  Nobody has measured it step by step.
• 5:17
  One 2025 survey compared whole episodes.
• 5:21
  A podcast-software company asked independent
• 5:24
  podcasters from its own audience about their hours.
• 5:27
  Those using AI reported taking about as long per episode as those who weren't.
• 5:32
  The company itself says there were too many variables to be sure.
• 5:36
  Its reading: AI won't necessarily save you time.
• 5:40
  That's not the same as saying it doesn't.
• 5:43
  The episode on AI voices said a synthetic voice cuts the time a show takes by a lot.
• 5:49
  This survey can't test that.
• 5:50
  Only three in a hundred used one.
• 5:53
  So what's the case for handing steps over?
• 5:55
  The podcast consultant Paul Colligan argued in
• 5:58
  2025 that the point isn't to automate your show.
• 6:02
  It's to do what used to cost too much time or money.
• 6:05
  A transcript of every episode is that kind of step.
• 6:09
  So is a dub in another language.
• 6:11
  Count what AI lets you add, not just the hours.
• 6:15
  Last: what should stay yours?
• 6:17
  A show you want to stand out still takes a great deal of judgment.
• 6:21
  Without your shaping, AI falls back on a default.
• 6:25
  That's where the episode on AI voices left off.
• 6:28
  A 2026 study gave one AI podcast tool different kinds of documents.
• 6:33
  The tool built the shows on the same fixed template.
• 6:36
  It also turned other languages and cultures into a standard American default.
• 6:41
  So be explicit before AI starts.
• 6:44
  Give it the show's structure, your sources and your point of view.
• 6:48
  In September, a well-known business podcaster had
• 6:51
  AI turn one of his own articles into a script.
• 6:54
  He added a story of his own, and a clone of his voice read it.
• 6:58
  The facts held, because they came from his company's own document.
• 7:01
  But his editor asked if he was sick.
• 7:04
  To the editor, it wasn't his usual style.
• 7:07
  The Washington Post tested AI-made personal
• 7:09
  podcasts before launching them last December.
• 7:12
  Its testers failed most of the scripts,
• 7:15
  though they'd been told to fail anything in doubt.
• 7:18
  It launched anyway, as a beta.
• 7:20
  Within two days, staff flagged invented and misattributed quotes.
• 7:25
  So a person has to listen to every episode.
• 7:29
  And they need the power to hold it back.
• 7:31
  AI can already help you do a lot of the work of making a show.
• 7:35
  So where do you begin?
• 7:38
  I'd start AI on the steps where a mistake is easy to catch and cheap to fix.
• 7:43
  That's transcripts and chapters, then clean-up and cuts.
• 7:47
  Then let it draft what you'll check against a source.
• 7:50
  That's show notes, research leads and a first script.
• 7:53
  Keep the brief, the point of view and the final listen.
• 7:57
  All of this is as of fall 2026.
• 8:00
  Claims about what AI can't do wear out fast.
• 8:04
  Hiring a producer instead?
• 8:06
  A 2026 survey of about fifty podcast agencies
• 8:09
  found more than nine in ten use AI somewhere.
• 8:13
  So ask which steps a tool does, and what a person checks them against.
• 8:18
  Before your next episode, open the automatic transcript of your last one.
• 8:22
  In Apple Podcasts, you touch and hold the episode.
• 8:25
  Then tap View Transcript.
• 8:28
  Check every name and number against what was said.
• 8:31
  And that's it for today's question.
• 8:33
  Thanks for listening.
• 8:34
  The written version is on workhacker.com,
• 8:36
  with every source, and a place to send us your questions.
• 8:40
  And if you want a podcast like this for your business, WorkHacker makes them.
• 8:44
  See you next time.