Writing Interactive Stories That Are Worth Replaying

A regular AI chat goes wherever the conversation drifts. A Story Mode story has a shape. The chapters are authored. The milestones are placed. The world exists before the player opens the door. Everything between those anchors is generated fresh, by the AI, every time someone plays.

The medium is unusual: part visual novel, part tabletop, part chat. Done well, the same story can be played three times and feel like three different stories. Done poorly, it's a railroad in disguise - the player notices the rails by chapter two.

This guide is about what separates the two.

The four parts of a Story Mode story

In Reverie, authoring a story means filling out four things:

Chapters - the narrative containers. Each chapter has its own emotional focus and tension. "The Discovery," "The Investigation," "The Confrontation."
Milestones - the moments that matter. Defined story beats the AI works toward without forcing. "First kiss." "Truth revealed." "Betrayal." When one triggers, the player feels narrative weight.
Data panels - the structured state. Relationship levels, clue counts, resources, health. The AI reads from and writes to these as the story progresses; low trust starts producing guarded dialogue, high affection produces openness.
Player identities - the roles a player can adopt. Same story, different identity. Detective vs. suspect vs. the victim's best friend. Genuinely different stories on the same skeleton.

These are the load-bearing pieces. Get them right and the AI does the rest. Get them wrong and no model in the world will save the playthrough.

Chapters: write the tension, not the plot

The most common mistake new authors make is writing chapter descriptions like episode summaries:

Chapter 2: The player goes to the warehouse, finds the briefcase, and meets the antagonist.

This tells the AI what should happen. That sounds like a feature; it's actually a problem. The AI will make those things happen, regardless of what the player is doing, and the player will feel it.

Write the tension instead:

Chapter 2: The clue points to the warehouse, but going there alone is reckless. The protagonist is short on allies and shorter on time. Trust is the resource that's running out fastest.

Now the AI has a gravity, not a script. The warehouse might come up; it might not. The player might bring an ally, refuse to go, or call the police instead. All those paths can satisfy the chapter, because what the chapter is about is trust and time pressure, not visiting a specific location.

This is the single biggest lever you have. Author what the chapter is about, not what happens in it.

Milestones: anchor the moments that need to land

Milestones are not achievements and not checkpoints. They're the moments you wrote this story to deliver. The first kiss. The reveal. The breaking point. The scene a reader would remember six months later.

A good milestone has three properties:

It's earned. It depends on something - a data-panel threshold, a chapter, an earlier milestone. The AI won't trigger it before the conditions are met. Skipping straight to "first kiss" on message three is impossible by design.
It's specific. "Things go well between them" is not a milestone. "She admits she's been lying about her sister" is.
It's optional. Not every playthrough should hit every milestone. The ones the player misses become the reason to replay.

Use 4-7 milestones for a chapter-length story. Fewer feels under-designed; more turns the story into a checklist the AI is racing through.

Data panels: the cheapest way to make a story feel reactive

A data panel is just a small set of numbers or flags the AI reads at every turn and writes back to as the story moves.

Affection (Mira): 0-100, starts at 20
Trust (Detective Park): 0-100, starts at 50
Clues found: list, starts empty
Time of day: morning / noon / evening / night
Player wounded: bool, starts false

That's it. That's the whole feature.

What it buys you: a character whose dialogue changes as their affection shifts, without you having to write three versions of every scene. The AI sees the numbers in context and adjusts. A 15-affection Mira speaks in clipped sentences; an 80-affection Mira touches your arm without thinking.

The trap: making the panel too complicated. Five fields is plenty. Twelve fields and the AI starts averaging across them and none of them feel meaningful. Pick the variables that change behavior and ignore the rest.

Player identities: the cheapest way to make a story replayable

This is the secret weapon. Most authors think "replayable" means branching paths and alternate endings. Those are expensive to design. Player identities are cheap and produce a stronger effect.

Same story. Same chapters. Same milestones. Different person playing it.

The detective gets information first but is distrusted by half the cast.
The suspect knows what really happened but can't say so without making it worse.
The victim's friend has emotional standing nobody else does, but no investigative authority.

These aren't difficulty modes. They're entirely different experiences of the same events. The same scene plays out completely differently when the AI knows who you are.

Two identities is good. Three is excellent. Four starts to dilute. Don't go past four unless you have a strong reason.

Designing for replay specifically

The question to ask yourself, repeatedly, while authoring: what will be different the second time?

Things that make the second playthrough feel new:

A milestone the player missed last time.
A data-panel state they didn't see (what does a 95-trust Detective Park sound like, when last time he topped out at 40?).
A second identity with access to different information.
A chapter tension that resolves differently because the player approached it differently.

Things that make the second playthrough feel like a rerun:

Required scenes that always happen in the same way.
Milestones that always trigger.
Data panels that never reach extremes.
One identity, one optimal path.

Audit your story against these. If the second playthrough doesn't have at least three sources of variation, the story isn't replayable, no matter how good the prose was the first time.

What Story Mode is not

Setting expectations correctly avoids a lot of frustration:

Not branching paths. The structure is authored; you don't write multiple endings. The variation comes from generated content, milestone selection, and identity, not from forking trees.
Not turn-based. There are no game mechanics beyond narrative progression. Data panels track state; they don't impose combat or skill checks.
Not multiplayer. Currently single-player. If you author with co-op in mind, that's authoring for a feature that doesn't exist yet.
Not a chat with extra steps. Don't write a story when what you actually want is a long-form chat. They're different surfaces; the wrong one is always frustrating.

How this stacks with the rest of Reverie

Character writing - story characters use the same authoring as chat characters. Sharp characters with contradictions and voice rules carry stories better than generic ones.
Memory - in-story memory still applies; characters remember within a playthrough. Data panels are the durable state that persists across the arc.
Pacing - chapter tension and milestone placement are pacing tools at the story level. The same instincts apply at a larger scale.
Free - Story Mode's core features (chapters, milestones, identities, data panels) are free for all users. Premium provides faster generation, not feature access.

The takeaway

Author the skeleton. Let the AI generate the flesh. Write tension, not plot. Pick milestones that deserve to be moments. Keep data panels small. Make identities different enough that the same story is genuinely a different story when you switch.

A story worth replaying isn't a story with more content. It's a story whose shape has room for the player to leave a different mark each time.