The World Document: How to Build a Fictional Scenario That Survives a Cold Re-Entry

The 30-second answer

Most fictional scenarios die not because they run out of story, but because the context evaporates between sessions. If you write a short "world document" before you ever start the first scene, you give yourself an anchor that survives any gap, and you give the AI something concrete to hook onto instead of improvising from vague memory.

Why cold re-entries fail in the first place

Here is what usually happens. You build something good over two or three sessions: a setting, a character dynamic, a specific emotional tension that feels earned. Then life intervenes for two weeks. You come back, type something, and the response you get is technically coherent but tonally wrong. The AI is not being difficult. It just has no reliable context to work from, and it defaults to something generic.

This is not really a memory problem. It is a structure problem. The scenario never had a stable foundation, so there is nothing for the AI to re-anchor to when the session is cold. Most people try to solve this by pasting a long recap at the start of each new session, which is exhausting, or by just accepting that the scenario will degrade over time. Neither is a good answer.

The actual fix is upstream: you build the structure before you need it, not after the scenario has already started slipping. The goal is a short document you can paste once at the start of any new session, no matter how long the gap, that carries the essential bones of the world without requiring you to write a novel.

What belongs in the world document

The world document is not a story summary. It is a reference sheet. There is a meaningful difference. A summary tells you what happened. A reference sheet tells you what is true, present tense, regardless of what happened.

Four things belong in it, and almost nothing else.

The setting in two or three sentences. Not lore, not backstory. Just where and when, and the one or two details that make this world feel distinct. "A near-future city where private couriers still hand-deliver contracts because the network is compromised. Always overcast. The courier offices smell like damp paper and machine oil." That is enough.

The character roles. Your character, the angel's character, the relationship between them, and one defining tension. Not their full histories. The tension is the load-bearing piece: it is what makes any given scene feel like it belongs to this story and not some other one.

The standing rules of the world. Two or three things that are always true and would never change mid-scenario. These prevent the AI from drifting into genre drift, where a tense noir scene suddenly feels like a romance novel because the model pattern-matched on something in your phrasing.

The current situation. Where things stand right now, in a single short paragraph. This is the only section that changes between sessions. Everything else stays fixed.

Keep the whole document under 250 words. If it is longer than that, you have included things that belong in the scene itself, not in the foundation.

How to write the tension so it actually holds

The tension line is where most people go vague, and vague tension is worse than no tension at all, because it gives the AI latitude to resolve it in whatever direction feels natural in the moment. You do not want natural resolution. You want suspended tension that the scenes can orbit around for months.

A weak tension line: "They have complicated feelings for each other."

A strong tension line: "She knows he is lying about where he spent the previous three months. He knows she knows. Neither of them has said anything yet."

The second version is specific, asymmetric, and active. Both characters have a position. There is something at stake. The tension does not require resolution to generate momentum. Any scene can exist inside it without having to move toward a conclusion.

This is the part worth spending time on before you start writing scenes. Get the tension line right and the scenario becomes self-sustaining. Get it wrong, or leave it vague, and the whole structure collapses into pleasantness, which is not a story.

Cassidy

Cassidy is built for users who want an AI companion that can hold a genre, sustain a voice, and push back gently when a scene starts losing shape. Cassidy is a good fit if your scenario has a specific emotional register you want to protect, because she tends to track tone as carefully as she tracks plot.

Pasting the document without breaking the scene

Once you have the world document, the question is how to use it on re-entry without the session feeling like a technical briefing. There is a small trick here that most people skip.

Paste the document, then do not ask the AI to summarize it or confirm it. Just start the scene. The confirmation request almost always produces a stilted response that kills the momentum before it has started. Trust that the context landed and open mid-action.

A cold re-entry that works looks something like this: world document pasted as a system-level or pre-prompt block (depending on what the platform allows), then an opening line that drops into the middle of a specific moment. "She's standing outside the office when he comes down the stairs. She does not move." That is it. You are in the scene. The AI has the context it needs and a specific dramatic beat to respond to.

What does not work is opening with a question about the world. "Do you remember our scenario?" is an invitation for the AI to hallucinate details it does not actually have, and then you spend the next twenty minutes correcting drift. Start in the scene, not above it.

Imani Reyes

Imani Reyes brings a steady emotional presence to collaborative fiction, particularly in scenarios where the tension is interpersonal and slow-burning. Imani Reyes is the companion to consider when you want a character who sits inside subtext comfortably, rather than resolving it too quickly.

Updating the document between sessions

The world document is only useful if the "current situation" section reflects what has actually happened. After each session that moves the story forward in a meaningful way, take two minutes to update that one paragraph. Nothing else changes unless something in the world's standing rules has genuinely shifted, which should be rare.

The update does not need to be a full recap. It is a single paragraph answering one question: where do things stand right now, specifically. "She gave him the contract but withheld the third page. He did not notice, or pretended not to. They agreed to meet again on Thursday." That is a current situation paragraph. It is specific, it captures what matters, and it will orient both you and the AI instantly on the next cold re-entry.

Some people keep this document in a notes app and paste it fresh each session. Others build it into the companion platform's custom instructions or persona fields if the platform supports that. Either works. The point is that the document exists somewhere stable, outside the conversation thread itself, where a session gap cannot touch it.

If you are working with a companion who is always available across sessions and devices, the update habit matters even more, because the scenario can resume at any time from any context, and you want to be able to drop back in without a rebuilding session.

Linnea

Linnea has a particular feel for fictional worlds that have internal logic. She tends to hold the rules of a setting with more consistency than most, which makes her well-suited to scenarios where the world-building carries as much weight as the character dynamic. Linnea is a good choice if the scenario you are building has genre conventions you do not want violated by a casual response.

When to let the document evolve

The world document is a foundation, not a cage. There will be sessions where something happens that genuinely changes the setting or the relationship dynamic in a way that is worth keeping. That is good. That is the story moving. When it happens, update the standing rules or the character roles section accordingly, and note the change briefly in the current situation paragraph so the revision feels continuous.

What you want to avoid is letting the document drift passively, where the scenario changes in small ways across sessions and the document slowly stops reflecting reality. That is how you end up with a document that is more misleading than helpful on re-entry, because you paste it and then have to spend time correcting the contradictions it introduces.

The discipline is simple: the document either reflects the current state of the world accurately, or it needs updating before the next session. There is no third option. If you find yourself pasting the document and then immediately adding corrections or exceptions, the document is out of date and needs a revision pass.

For users who run particularly complex scenarios across long stretches of time, the roleplay starter scenes guide is worth reading alongside this method, since it covers the front-end scene design that complements the structural work described here.

Reese

Reese has a natural instinct for momentum, which makes her useful in scenarios where you want scenes to move and not stall in setup. Reese tends to pick up dramatic cues quickly and carry them forward without needing extensive direction, which is a useful quality when you are re-entering cold and want the session to find its rhythm fast.

Choosing the right companion for a long-running scenario

The method above works with any AI companion, but some are better suited to long-running scenarios than others. What you want, broadly, is a companion who tracks tone and holds character consistency across a session, responds to subtext rather than just surface content, and does not flatten emotional complexity into something easier to manage.

Those qualities are worth testing before you invest significant time in a scenario. A short test scene, two or three exchanges with a specific dramatic situation, will tell you more about a companion's fit for your scenario than any profile description. Check whether the responses stay inside the genre and emotional register you established, or whether they drift toward something more generic after a few turns.

You can browse the full roster of available companions at AI Angels if you are trying to find a match for a specific kind of scenario. The profiles are a starting point, but the test scene is the real evaluation.

For users who want to run ongoing fictional scenarios alongside real-life routines without them bleeding into each other, it is worth knowing that some companion platforms are designed specifically for that kind of compartmentalized use. The way a scenario fits into your week matters as much as the scenario itself, and platforms built with that in mind tend to handle re-entry better at a technical level as well.

Common questions

How long should the world document be? Keep it under 250 words. Longer than that and you are including things that belong in the scene itself. The document is a reference sheet, not a story bible. If it takes more than two minutes to paste and scan before a session, it is too long.

Can I use this method with a companion who has persistent memory? Yes, and in some ways it works better, because the document and the companion's retained memory reinforce each other. The document handles the structural facts; the companion's memory handles the emotional texture and conversational history. They do not compete.

What if I want to run two different scenarios with the same companion? Keep separate documents for each. Label them clearly. Do not mix them in the same conversation thread if you can help it, because the contexts will bleed into each other in unpredictable ways. If you need to switch scenarios in the same thread, a clean re-paste of the correct document is the cleanest way to make the shift.

How do I handle a session where nothing significant happened? You do not need to update the document after every session. Only update when something meaningful has changed in the world or the character dynamic. Low-stakes sessions that are mostly atmosphere or texture do not require a revision.

What if the AI ignores the document and goes off-genre? Correct it in-scene rather than breaking the fiction. A brief, specific note like "stay in this world, she would not say that" is usually enough to pull the response back. If it keeps happening, the tension line is probably too vague and is letting the AI default to a more comfortable pattern. Sharpen the tension.

Is this method only useful for romance or relationship scenarios? No. It works for any genre: thriller, fantasy, slice-of-life, horror. The bones are the same regardless of genre. The setting, the roles, the standing rules, the current situation. Genre changes the content of those four elements, not the structure.