September 24, 2026
How to write an audio drama that hooks: 12 rules of writing for the ear
Hook, dialogue, narrator, sound effects, silences, music and cliffhangers: the concrete rules to write an audio fiction people listen to until the end, with or without AI.
Frank HoubreFounder of Imaginode
Writing for the ear is not writing for the eye
A listener cannot reread or see: everything they need to understand comes through a voice, a sound or a silence, and they drop off as soon as they get lost.
A novel can describe a face over a whole page; a film script can let the image speak. An audio fiction has neither. It has voices, sounds, music and time. The rules below come from radio, from the episodic audio series that dominate fiction apps, and from what we learned having AI write and produce episodes in the Drama Studio.
They apply whether you write by hand or with AI. In the second case they matter even more: a language model left to itself writes for reading, and it is up to you to impose the ear.
1. Everything must be heard
If a piece of information is neither said nor heard, it does not exist for the listener.
A character pulling a weapon, another turning pale, a letter placed on the table: to the ear, none of it happens unless someone says it or a sound shows it. "Put it down. Put that knife down." beats a stage direction no microphone will ever record.
Reread your script with your eyes closed, or better, listen to it performed: every time you have to imagine a gesture to follow the scene, a line or a sound is missing.
2. Hook in the first twenty seconds
The first seconds decide everything: an abnormal situation, a question or a danger, not a setting.
In the Train de Minuit pilot, the conductor finds a ticket dated 1987 before the first minute is over. Opening on a description of the station would have lost half the listeners. Start in the middle of an action and let the context arrive in small touches.
It is also the first score the studio's AI script editor looks at: the hook. Below 7 out of 10, rewrite the opening before anything else.
3. Have characters name each other
By ear, you recognize a voice, not a name: characters must call each other by name, especially early on.
"Claire, have you seen the time?" introduces two characters in one line: the one speaking and the one spoken to. In the first scenes, use first names more often than you would on the page, then space them out once the voices are established.
Also avoid names that sound alike. Leah and Leo, Mark and Max get mixed up far more by ear than on paper.
4. One voice, one way of speaking
Each character should be recognizable by their words as much as by their timbre: rhythm, register, verbal tics.
An inspector who speaks in curt questions, a grandmother who ends her sentences with a proverb, a teenager who interrupts everyone: these signatures let the listener follow a three person scene effortlessly. Write them into each character sheet, as the studio's Bible agent does for every role.
On the voice side, pick distant timbres for characters who share scenes: different genders, ages, low and high voices. Two voices that are too close turn a dialogue into a confusing monologue.
5. Short lines that answer each other
One idea per line, sentences that cut each other off and reactions: that is what makes it feel like a real scene.
On the page, a paragraph of dialogue works; by ear, past two sentences the listener loses track of who is talking. Cut it up, have the other character react, leave sentences unfinished when tension rises.
For total length, keep one benchmark in mind: a calm fiction voice says about 14 characters per second. A 3 minute episode therefore fits in about 2,500 spoken characters, around sixty short lines. It is the number the studio gives its AI writer, because a script that is too short is the most common mistake.
6. Use the narrator sparingly
The narrator moves time forward and changes place; if they comment on emotions, the fiction becomes an audiobook.
"Quarter past midnight. Gare de Lyon. The last train to Nice waits on platform three." Three sentences, and the listener knows where and when they are. That is all a narrator should do. As soon as they explain what a character feels, they steal the scene from the actors.
In a 3 minute episode, five or six narrator lines are enough. Beyond that, ask yourself whether a line of dialogue or an ambience would say the same thing.
7. Let the ambience tell the place
A well chosen ambience replaces a whole description: a few seconds of a station platform at night say more than a paragraph.
Every location in your series deserves a continuous ambience: rain on a platform, the murmur of a café, wind in a forest, the hum of a train. It plays in the background throughout the scene, and a change of ambience is enough to signal a change of place.
Write it as a list of concrete sources, not as a mood: "light rain on a platform, distant announcements, a train idling" rather than "an ominous atmosphere". It is also the form sound generators understand.
8. A sound effect must carry the story
A sound effect is not decoration: it must tell the listener something, otherwise it wears them out.
The door slamming as the suspect runs, the phone buzzing during a confession, the ticket being handed over: these sounds move the scene forward. Footsteps in a corridor, three times in a row, only slow it down.
In our tests, an AI writer left to itself placed eleven sound effects in two minutes. We now cap it at about one per forty seconds, and the episodes play noticeably better.
9. Silence is a tool
One second of silence before a reveal is worth all the dramatic music in the world.
On radio, silence is an event: the listener leans in. Place a short pause before an answer that matters, after a reveal, or when a character hesitates. One to two seconds is enough, two or three times per episode.
The opposite is also true: an unintended silence between two lines stitched together breaks the scene. That is why it is better to have the lines of a scene performed together rather than recorded one by one.
10. Music frames, it does not fill
A recognizable theme at the opening, a short sting at the ending, and the music keeps quiet under the dialogue.
The music theme is the series signature: listeners should recognize it from the first notes of every episode. Compose it once, as an instrumental, and reuse it everywhere.
During scenes, the music drops well under the voices, or stops. A dramatic sting of a few seconds is enough to underline the episode's ending. More music does not add tension, it takes it away.
11. End every episode on a question
The last line must make the next episode a must: a reveal, a danger, a pending decision.
"I need to find that ticket tomorrow" is a lukewarm ending; "This ticket... it has my name on it" is a real one. The whole episodic audio series format rests on this mechanism: listeners come back because they cannot leave the question hanging.
Write the ending before the rest of the episode, then work back to the opening. The script editor's "ending" score is the one to watch on every episode.
12. Keep continuity
A character cannot know what they have not learned: over a season, it is the mistake listeners forgive least.
Keep a log, episode by episode: what happened, what each character learned, which threads are open or closed. It is exactly the memory the Drama Studio records after every episode it writes, and gives back to the writer before the next one.
These twelve rules are quick to test: write a pilot, have it performed and listen to it in one go. Our guide to creating an audio series with AI walks through the full production, from bible to mixed file, with the real costs of an episode.