how to create a podcast with ai comes down to one choice: let the ai do the talking, or let it do everything except the talking. the first route turns a document into a show read by synthetic voices. the second keeps your voice and hands the rest to the ai: the questions, the edit, the levels, the show notes and the feed. if the podcast is meant to build trust in you or your business, the second route is the one that works, and it is what this guide walks through.
the short version in underway: set up an episode, pick just my voice, choose audio interview or podcast with ai co-host, and talk. the questions are asked out loud, the answers are cut into a numbered episode with the questions kept in, and the audio goes to your podcast feed, from which apple podcasts and spotify pull it. no script, no editor, no studio.
how to create a podcast with ai: two routes
most guides on page one describe a pipeline: research with a chatbot, write a script with it, turn the script into speech with a voice generator, clean it up in an editor, publish through a host. every step can be automated, and some tools, google's notebooklm, jellypod and wondercraft among them, do all of it in one press. that is a real option, and it is worth being clear about what it gives you.
| the ai talks | you talk, the ai does the rest | |
|---|---|---|
| what goes in | a document, a link or a topic | your answers to questions asked out loud |
| whose voice | synthetic voices | yours, with the host's questions |
| what it knows | what is in the source | what happened to you, which nobody else can say |
| time to make | minutes, with no recording | as long as you talk, then the cut is made for you |
| good for | turning reading into listening | a show people come back to because of who is in it |
a generated show can be useful as a summary. what it cannot do is introduce you. listeners decide whether to trust a business by hearing the person behind it, and an episode with no person in it gives them nothing to hear. so the rest of this page is the second route.
how to create a podcast with ai that still sounds like you
the hard part of a podcast was never the microphone. it is knowing what to say, saying it to nobody, then spending an hour cutting out the ums. an ai can take all three of those off you without saying a word on your behalf. here is the whole process, step by step.
step 1: decide what the show is about
write one line about what you are documenting. not a pitch: a sentence, like building a ceramics shop from ten customers to a hundred, or the first year of a design studio. there is no list of show types to choose from, because the line is more accurate than any list.
then record the first episode, episode zero. it is the opener every series has, and it covers what you are building, who it is for, what problem it solves and what people do instead today, where it honestly stands, the hardest part, and what comes next. you can practise first, and keep only the take you want. after that, every episode's questions are written knowing what episode zero said.
step 2: choose how you will be asked
when you set up an episode, the first screen asks what kind it is. under just my voice there are three ways to make a podcast:
- audio interview. the host asks questions written from what you picked, one at a time, out loud. you see them before you start and can change any. you answer for as long as you like; it says a short word when you stop and asks the next one, with no press from you. it uses nothing extra on any plan.
- podcast with ai co-host. a live conversation that talks back. no list: it follows where you go, digs in when you say something worth digging into, and you end it when you are done, up to thirty minutes a call. it uses co-host time from your plan.
- a person asks. a co-founder or a friend asks instead, in the room or by a link, and the app keeps quiet.
a topic is optional. pick one from what to film, the list of things your series thinks are worth an episode, or type your own words, or pick nothing and the questions come from your day. each topic you pick becomes one question, so one topic is a short episode and four is a longer one.
creating a podcast with ai needs no studio. a parked car is a quiet room. an ai illustration, not a customer.
step 3: record
find somewhere quiet and put headphones in if you have them. on a laptop it runs in the browser; on an iphone it runs in the app. start, and talk the way you would to a person across a table. you do not need to plan your answers, and you should not read them. a read answer sounds read.
say i'm done when you have finished, or just pause, and the host moves on. that's enough for the episode ends it early. nothing is drawn on screen while you talk, so a phone in your pocket on a walk is as good as a desk.
step 4: let the ai edit it
this is where most solo podcasts die, and it is the part an ai is best at. underway cuts the episode on full sentences, in the order you said them, with filler taken out. it never moves an answer to make it punchier and never adds a word you did not say. the questions stay in, spoken in your series' voice, which is what makes it sound like an interview rather than a voice note with gaps.
the result is one levelled file. you can play it on the episode page, strike a line you would rather not keep, and the cut updates. a voice episode is also drawn as a video, your episode card with the waveform moving to both voices and the words captioned, so the same podcast can go to youtube, tiktok, instagram, linkedin and x.
a real screen from underway: a finished audio episode, cut and ready to post.
step 5: publish to apple podcasts and spotify
a podcast is an rss feed. apple podcasts, spotify and the rest subscribe to the feed and pull every episode from it, so you only set it up once. on the series plan every series with an address has a feed, shown in publishing with copy feed and the steps for each platform. submit the address from their own dashboards, and from then on every episode with audio lands on both.
each episode also comes with its words: show notes, a youtube description, a linkedin post, an x thread and captions, all written from what you actually said. read them before they go; they are yours to change.
what ai should and should not do in your podcast
- it should ask. a good question is the difference between a ramble and an answer, and an ai can write one from what it knows about your week.
- it should edit. filler, false starts and long pauses are mechanical, and nobody should spend an hour on them.
- it should write the words around the episode. show notes and captions are summaries of what you said, which is what a model is good at.
- it should not speak for you. no cloned voice reading a script, no answers you did not give.
- it should not reorder the story. an episode that plays in the order it was said sounds honest; one rearranged for a hook sounds produced.
how often to publish
as often as you say you will, and no more. underway asks what you commit to: weekdays, every day, or a number a week. the streak counts against that, not the calendar, and a pause keeps it whole when you are away. episodes are numbered, not days, so a missed week leaves no gap anyone can see. once a day, whenever suits you, is what it is built for, and a short episode counts as much as a long one.
what it costs
every new account starts with a 7-day free trial of any plan, $29.99 a month after that, which includes the audio interview, the edit and thirty minutes of co-host time a month. the podcast feed is part of series, $79.99 a month, which also has two hours of co-host time. the audio interview never uses co-host time.
can i create a podcast with ai without recording my voice?
yes, with a generator that reads a script in synthetic voices, such as notebooklm or jellypod. underway does not do this: every episode it makes is your own voice answering.
do i need a script to create a podcast with ai?
no. the questions are written for you and asked out loud. you answer them, and the episode is cut from what you said.
what equipment do i need?
a phone or a laptop. headphones help, especially with the co-host, because on a speaker it can hear itself.
how long should each episode be?
as long as your answers. each topic you pick is one question. one topic makes a short episode, and the co-host runs up to thirty minutes a call.
how do i get my ai podcast on spotify and apple podcasts?
submit your series' feed address once from their own dashboards. on the series plan every episode with audio is added to the feed, and both platforms pull it from there.
the host that talks back, explained: podcast with ai host.
when you would rather ask yourself the questions: how to interview yourself for a podcast.
Originally published at https://underway.life/how-to-create-a-podcast-with-ai
Top comments (0)