The short answer: AI is useful for podcast scripting in the preparation stages — structuring an episode, writing a cold open, drafting questions, and turning show notes into a promo. It is much weaker at the part listeners actually stay for, which is the specific, slightly messy way a real person tells a story. Script an outline, not a monologue, and you will sound better than if you script every sentence.
| Episode part | Script fully? | Why |
|---|---|---|
| Cold open | Yes | Short, and worth getting tight — it decides whether people stay |
| Host intro | Loosely | Needs to sound the same every week, but not read |
| Segment transitions | Bullets | Over-scripted transitions are the clearest tell of a written script |
| Interview questions | Yes, plus follow-ups | Prepare more than you need; the best moments are usually follow-ups |
| Ad read | Yes | Required wording is often fixed by the sponsor |
| Stories and opinions | No | This is the part listeners came for |
Write for the ear, not the page
The most common mistake is generating prose that reads well and sounds wrong. Spoken English differs from written English in specific ways worth instructing the model about:
- Shorter sentences. A listener cannot re-read a clause.
- Signposting. "Three things went wrong" tells the ear what to expect; written prose often omits it.
- No nested parentheses or long lists. Both are hard to follow aloud.
- Contractions and plain words. "We did not" written out will be read as "we didn't" and look wrong on the page.
- Read it out loud before recording. This catches most problems faster than any amount of editing on screen.
A rough guide for length: most people speak at around 130-150 words per minute. A 20-minute segment is roughly 2,600-3,000 words of script. If AI gives you 5,000 words for a 20-minute slot, you are going to rush.
A structure that holds up
- Cold open, 30-60 seconds. The most interesting thing in the episode, before any introduction.
- Intro, under 30 seconds. What this is and why it matters today.
- First segment. Establish the problem before any solution.
- Middle. The substance. If it is an interview, this is mostly questions.
- Ad read. Placed at a natural break, not mid-sentence.
- Close. One takeaway and one next step. Not a summary of everything.
Using AI for interview prep
This is the highest-value use. Paste a guest's bio, their last few published pieces and the episode angle, then ask for: ten open questions, five follow-ups per question, one question that gently challenges their stated position, and three places where their public work is thin. That last one is where good interview questions come from, and it takes far longer to do by hand.
Show notes and promotion
Once the episode exists, the repurposing work is mechanical and AI does it well: chapter timestamps from a transcript, a 150-word description, five quote cards, a short promo script and an email blurb. Give it the transcript and ask for each output separately rather than all at once — combined requests tend to produce shallow versions of all five.
What do people ask about AI writing tools for podcast scripts?
Can AI write podcast scripts?
It can script the structural parts well — cold opens, intros, questions, ad reads. It is weaker at stories and opinions, which are the parts listeners stay for. Outline rather than script every sentence.
How do I choose an AI for podcast scripts?
Both Claude and ChatGPT handle this. Choose based on which one tracks your show's voice over a long prompt, and paste a transcript of your own speaking as the style reference rather than a written sample.
How long should a podcast script be?
Most people speak at roughly 130-150 words per minute, so a 20-minute segment is about 2,600-3,000 words. If the draft is much longer, you will end up rushing.
Should I write for reading or speaking?
Speaking. Use short sentences, signpost what is coming, avoid nested clauses and long lists, and read the script aloud before recording.