Making a podcast involves five stages — idea, script, recording, editing, publishing — and AI meaningfully shortens three of them. The honest boundary: AI does not decide your show's angle, and it does not have the conversations that make an interview worth listening to. What it does is compress scripting, cleanup, transcription, show notes, and publishing admin. Here is the full workflow with the tools' real limits at each stage.
Stage 1: Idea and script
ChatGPT (Freemium) is the planning workhorse: episode outlines, interview questions, segment structures. Verify any factual claims it suggests — its answers can contain confident-sounding errors. Perplexity AI (Freemium) grounds research in cited answers, useful for guest background. For the script itself, Video Script AI (Freemium) drafts and outlines — though it is script-only, with no production included.
Stage 2: Recording
Riverside (Freemium) records remote podcasts and interviews with local capture, avoiding the compression that video-call audio suffers. Its limits: the free plan caps recording hours and video resolution, and remote recording needs good internet and a quiet room. Podcastle (Freemium) combines remote recording, AI editing, and voiceover segments — heavy editing sessions can strain older computers, and the free tier caps recording time.
Stage 3: Editing
Descript (Freemium) transcribes and edits audio like a document — the fastest path from raw recording to cut episode. Complex music or multitrack editing still suits traditional DAWs better, and some users need a VPN for reliable access. Auphonic (Freemium) then handles leveling and loudness compliance; the free tier caps processing time per month, and processing happens in the cloud. Krisp (Freemium) removes background noise — very loud or chaotic noise can still leak through.
Stage 4: Show notes and extras
Castmagic is Paid and generates show notes, social clips, and transcripts — AI show notes need review to match your show's voice, and there is no free tier. Whisper is Open Source for subtitle generation, meeting transcripts, and custom pipelines — larger models need substantial GPU memory to run comfortably.
Stage 5: Publishing
Alitu is Paid and automates episode publishing, audio cleanup, and show setup — with no free tier, so you pay before testing the workflow. Listnr (Freemium) covers article narration and podcast hosting with limited free word and voice quotas.
A full workflow that chains them
- Outline the episode with ChatGPT; research the guest with Perplexity.
- Record with Riverside or Podcastle.
- Cut and clean with Descript, then level with Auphonic.
- Generate show notes with Castmagic or a manual pass.
- Publish with Alitu or your existing host.
Common mistakes
Skipping the leveling pass — inconsistent loudness loses listeners; recording in an untreated room and relying on AI to save it; paying for show-note tools before your show has ten episodes; and over-editing out the personality.
FAQ
Can AI edit a whole episode? It cuts and levels, but creative editing is still yours.
Which recorder is best? Riverside for quality; Podcastle for all-in-one.
Is Descript enough? For spoken-word shows, mostly yes.
Do I need Castmagic? Only once publishing volume justifies the Paid plan.
Can AI handle remote guests? Riverside and Podcastle both record remote interviews locally.
What about transcripts? Descript transcribes as part of editing; Whisper is Open Source for custom pipelines.
How long until launch? A first episode is realistic in a weekend with this stack.
Pricing at a glance
ChatGPT, Perplexity AI, Riverside, Podcastle, Descript, Auphonic, Krisp, and Listnr all offer free tiers with real caps — recording hours, processing minutes, transcription volume. The Paid tools, Castmagic and Alitu, automate the show-notes and publishing stages that only matter once you are producing regularly. For a first season, the free stack covers every stage. The first paid upgrade most podcasters make is publishing automation, because admin fatigue — not editing — is what actually kills shows around episode ten.
What determines whether anyone listens
Not the tools. A podcast survives on a clear premise, consistent publishing, and a host with something to say. AI compresses the production work — and production work is the reason most shows quit — but it cannot supply the premise or the personality. The creators who use this stack well publish more of the conversations they were already having, just without the three-hour edit that used to follow. Treat the tools as the reason your show survives episode fifteen, not as the reason anyone tunes in.
