Asking the AI Narrator a Question Is Not a Move

Asking the AI narrator a question is not a move, and the difference matters more than it sounds. A move is something your character does; a question is something you want to know. Typed into the same box, the second becomes the first — your character says the words out loud, the room hears them, and the story now contains a person who stopped to ask who Mira’s brother was.

• A move changes the world; a question changes only what you know.
• Typed as a move, a question usually becomes spoken dialogue, and the story records it as something that happened.
• Parser interactive fiction separated the two decades ago: the QuestJS player’s guide states that generally no turn is taken for a meta-command.
• Most questions readers want to ask are already answered in the story’s memory record, which costs nothing to open.
• Fabledrift, an interactive fiction app for Android, lets you ask the narrator a question without advancing the story.

Why does asking the AI narrator a question derail a scene?

It derails the scene because the narrator has to read your input as something, and the only thing on offer is an action. “Who is she again?” arrives as a line your character speaks. Someone answers it in voice. The scene you were in now has an awkward beat in the middle of it, and the beat is yours, not your character’s.

The second cost is quieter and lasts longer. A spoken question is an event, and events go into the story’s record next to everything else that happened. Three scenes later the narrator still knows your character was confused about the letter, and it may write that confusion into how other people speak to you.

What parser interactive fiction settled long ago

Text adventures sorted this out by sorting commands into two piles: commands about the world, and commands about the game. The QuestJS player’s guide puts it plainly — “Meta-commands are commands about the game itself, rather than the game world” — and adds that generally, no turn will be taken for a meta-command. SAVE, UNDO, HINT and VERBOSE sit outside the fiction. Nothing in the room hears them.

The same guide notes that a turn passes only when a command is both understood and successful. That is a useful test to carry into AI stories, where no parser is enforcing it for you. Before you type, ask whether what you are about to write should cause time to pass. If the answer is no, it is not a move, and it should not go where moves go.

Register	What you are doing	Should time pass?
A move	Something your character does or says	Yes
A look	Attention paid to something in front of you	Yes
A question to the narrator	Something you, the reader, want to know	No
A note to the narrator	A fact you want carried forward	No

The fourth row is worth separating from the third. A question asks for what is already recorded; a note changes what is recorded. Both live outside the scene, and confusing them produces the familiar failure where a reader argues with the narrator for four turns instead of writing down the correct name once.

Which questions belong in the story and which do not

If the answer is something a person standing in that room could find out, ask it inside the story. If the answer is about the story’s machinery — what is written down, how much of an earlier thread is still live, why a scene went the way it did — ask it as an aside. That single test resolves most cases.

What you want to know	Where it belongs	Why
How something present looks, smells or sounds	In the story, as a move	Attention is an action your character can pay
Whether a plan would work	In the story, as an attempt	A failed attempt is still a scene worth reading
A name or detail you were told and forgot	The memory record, then an aside	The record already holds it, in writing
What the narrator believes about a character now	An aside	The answer describes the story’s state, not the room
Why the last scene went the way it did	An aside	The reason is authorial; nobody in the fiction knows it

The first two rows are the half that rewards practice. An input the narrator can use names a thing and does something to it; one it has to guess at gestures at a mood and hopes. The last three are not inputs at all, and treating them as such is how a clean story picks up a scene it did not need.

When the question is better asked inside the story

There is a third option readers forget, and it is often the best one: ask a character rather than the narrator. If you want to know who the woman at the gate was, your character can ask someone who was there. The answer comes back as dialogue, in voice, from a source with reasons of their own to be helpful, evasive or wrong. That is usually better reading than the same fact delivered flat.

This works only for what a character could plausibly know. Nobody in The Emberwood can tell you what the story’s record holds. Nobody on Ironreach Station knows why a scene was written the way it was. Those questions have no in-world speaker, and trying to force one produces a character who suddenly sounds like a manual.

How to phrase the question so the answer is useful

Ask about the record, not about the fiction. “Does anyone know about the letter?” invites the narrator to improvise, and whatever it improvises may become true by being written. “Who currently knows about the letter?” asks it to report what is already there. The first produces a sentence; the second produces a list you can act on.

• Name the thing you mean instead of pointing at it with a pronoun.
• Ask for state: who knows, what is owed, where something was left, what is still open.
• Keep one question to one aside. Two questions come back as one blended answer.
• Ask what the narrator has, not why it did something. The second invites an apology.

Chat-shaped roleplay apps handle all this with brackets: an out-of-character line in parentheses that the model is supposed to read as instruction rather than speech. It works unevenly, because the line still lands in the transcript and is still sitting there ten turns later, tinting everything after it. Having somewhere else to put the question is part of what separates an AI roleplay chat from interactive fiction.

Check the record before you ask

Before asking the AI narrator a question, open the story’s record: most of what readers want to ask has already been answered in writing. The story’s memory record — the facts, people, places and open promises the narrator is handed every turn — is the same material an answer would be drawn from, and reading it yourself is faster and more exact than asking for a summary of it.

It also tells you something an answer cannot: whether the fact is recorded at all. If the name you are hunting for is not in the record, the narrator does not have it either, and an aside will produce a confident invention rather than a retrieval. That gap is the whole of the difference between story memory and a context window, and it is worth knowing which side of it you are standing on before you ask.

The test to run before you type

Read back what you are about to write and imagine a person in the room hearing it. “I search the desk” survives that. “Wait, what floor are we on?” does not — it is a reader’s sentence wearing a character’s clothes. Asking the AI narrator a question belongs in the other channel: send it there, take the answer, and return to a scene that never knew you left. The record stays a record of what happened, and your confusion stays yours.