The quickest way to make a generated episode sound generated is to have one voice explain a topic from start to finish. Nothing interrupts, nothing is challenged, no one asks the obvious question. It is a document read aloud.
The quickest way to fix it is to have more than one person in the room, and to make them genuinely different from each other.
Cast size changes the format, not just the voice count
A podcast's cast is a roster of characters, and the number you put on it changes what kind of episode gets written.
- One host produces a narrated explainer. Appropriate for a briefing, a lecture, or a focused monologue — and the script is written for solo delivery rather than being a two-hander with one part deleted.
- Two hosts is the conversational default: one tends to drive, one tends to question, and the material gets examined rather than delivered.
- Three or more becomes a panel. This is where disagreement gets interesting, because positions can genuinely differ rather than just alternating.
What a character actually carries
A character is more than a name and a voice. Each one has attributes that shape how they speak and what they say:
- Expertise. What this person actually knows. An economist and a historian looking at the same source pull out different things, and the script reflects that.
- Intellect. How the character reasons — whether they go straight to the abstraction or work from concrete examples.
- Emotion and personality. Warmth, scepticism, enthusiasm, dryness. This is most of what makes a voice recognisable across episodes.
- Voice. Per language, as discussed in the languages guide. A character keeps their identity across every language the show is produced in.
Characters live at the account level, so a host you design once can appear in several shows, carrying the same persona and voice each time.
Give them something to disagree about
The single highest-leverage thing you can do is stop designing two hosts who agree.
Two enthusiasts produce an episode where every point is affirmed. An enthusiast and a sceptic produce one where every point is tested. The second is more useful to a listener and dramatically more listenable, and it costs nothing but a different personality setting on one character.
The same applies to expertise. A practitioner and a theorist. Someone close to the subject and someone encountering it fresh — the second one asks the questions your audience is actually thinking.
Speaker dynamics
Beyond the individual characters, a show has dynamics that govern how the conversation runs — who leads, how much the hosts interrupt one another, whether the register is formal or loose. This is separate from character design on purpose: the same two characters can host a tight formal briefing and a rambling discussion, and you should not have to rebuild them to switch.
Format sits alongside this. The same cast and the same sources can produce a lecture, an interview, a discussion, a panel debate, or a solo monologue, and can be informational, persuasive or analytical in stance.
Hosts that hold up under questioning
There is a practical test for whether a cast is well-designed: put them in a live conversation. Live Talk lets you interrupt the hosts mid-episode and ask them something, and they answer in character, in their own voices.
A well-designed cast handles this gracefully — the sceptic stays sceptical, the expert answers from their domain. A cast of interchangeable narrators produces answers that could have come from any of them, which is the same flatness that made the episode dull in the first place, just harder to ignore.
A starting recipe
If you are building a first cast and want something that works:
- Two hosts.
- One with deep expertise in the subject; one bright, curious and not an expert.
- Opposed temperaments — one warm and enthusiastic, one dry and questioning.
- Clearly distinct voices. If listeners cannot tell who is speaking with their eyes shut, nothing else matters.
- Generate one episode, listen, and adjust a single attribute at a time.
That last point is worth emphasising. Changing four things at once tells you nothing about which one helped. Change the sceptic's temperament, regenerate, and listen to what moved.