Halfway through a game, the players begin connecting clues. An argument develops around a character’s motives. The host brings out an AI-generated video prepared for the group.
The video might be beautiful. Will it deepen the discussion, or stop everyone to watch something that has little to do with their reasoning?
This is an imagined scene. It captures a question we wanted to address when we began exploring AI for scripted entertainment.
In these experiences, a group brings a story to life together. Players read, reason, and inhabit characters; a host attends to pace and emotion. At performance-led venues, actors and physical spaces also carry the story. AI can produce words, voices, and images. It does not automatically know when to enter.
In June, we prepared a research checklist. Character videos, evidence footage, physical props, and character epilogues were among the possibilities. Spaces for a venue, a game, and a pilot moment remained unfilled. We had many things we could make, and still needed to understand how they would fit an actual session.
Find a moment that leaves room for the players
Across a session, the work changes.
Before arrival, a venue needs to explain a story without revealing its secrets. At the opening, the host establishes a shared world. During play, the players need space to form their own judgments. After the debrief, content can become something to take home.
A character video might establish context at the opening and compete for attention during an intense discussion. Generated material must also respect confirmed clues and endings. The more convincing an image becomes, the more easily an error could alter a player’s judgment.

We would rather start with one moment. A character keepsake collected after the debrief is one example: the ending is confirmed, permitted material is checked, and the player chooses whether to receive it. Even if a person assembles the first version, the whole path from source material to a worthwhile object becomes visible.
That makes a modest question easier to answer: did this add something to the experience?
The host holds the other half of the answer.
How many extra taps does the new material require? How does play continue if generation is unfinished? Who resets a prop? Our checklist placed these questions alongside player experience because waiting, operation, and upkeep all consume hosting time.
We want to observe these details before choosing a form for AI. It might be an opening prepared in advance, a card after the session, or a design we try and decide to abandon.
The unit of research becomes a specific collaboration: a defined input, someone responsible for checking it, a suitable moment to use it, and a way to keep the story moving when things go wrong.
That is the starting point June left us.


