What is metagaming in roleplay?
Updated
Metagaming is when a character acts on knowledge their player has but the character shouldn't: a spoiler from the card, a secret written in a persona, an aside made out of character. In AI roleplay the model metagames when its character "knows" something it only read in your notes, and you metagame when your character reacts to a twist you saw in the scenario. The fix is keeping player knowledge and character knowledge apart.
Try it now, free, no account neededStart a story with Detective Rosa QuinnThe case is buried. She brought a shovel. You're in.Player knowledge vs character knowledge
The term comes from tabletop games. A player who has read the monster's weaknesses in the rulebook, and has their brand-new fighter suddenly reach for silver, is metagaming: the character never had a way to learn that. In roleplay the same split applies. You can know the butler did it; your detective has to find out.
It matters because secrets, reveals and misunderstandings are some of the best fuel a story has. Once a character acts on something they couldn't know, the reveal you were building towards has already happened, just badly.
Why AI characters metagame
A model gets one big prompt: the character card, your persona, any memory notes and the recent chat, all at once. Nothing in it is marked "the character can't see this". So if your persona says "secretly a spy for the enemy court", a character may greet you with a knowing look and ask who you report to, three messages in.
| Where a fact sits | Who reads it | Metagaming risk |
|---|---|---|
| Your persona or About you | The model, every reply | High for secrets |
| Story memory notes | The model, every reply | High unless marked hidden |
| An OOC aside in the chat | The model, while it stays in view | Medium |
| Something said in the scene | Everyone present | None, that is fair play |
Keeping a secret from the character
You have two good options. The cleanest is to leave the secret out of every note and simply play it: the model can't leak what it never saw, and you reveal it when the moment comes.
If the AI needs to know so it can drop hints or react properly later, label it as hidden and say what would expose it:
- Hidden from Caelan: Alex is reporting to the old king. Caelan must not suspect unless Alex slips up or is caught with the letters.
That is a natural fit for Prince Caelan Vire, whose new spouse arrived from the enemy court three days ago. On UnsavedAI, About you and Story memory (⋯ menu) are both read on every reply, so treat anything you put there as something the model will use. If a reply leaks a secret anyway, Edit it out at once; left in the chat, the leak becomes part of the story.
When the player is the one metagaming
Users do it too, usually by accident. The card says the friendly florist is an assassin, and your character starts eyeing her knives in scene one. Or you guess the culprit from how the greeting is written and your detective accuses them with no evidence.
Mysteries suffer most. With Detective Rosa Quinn or Ophelia Marsh, who knows every secret in a house where the lord lies dead, let your character earn each clue on the page. A useful habit is asking yourself "how would my character know this?" before acting on a hunch. Detective mystery roleplay has more on keeping a case fair.
Meta that helps the story
Not every use of outside knowledge is a problem. Steering the plot out of character is meta by design, and it works: (OOC: let's have someone recognise me at the ball) tells the model what you want without your character knowing it is coming. Knowing tropes is meta too, and it helps you set up payoffs. The line to hold is simple: you can use what you know to shape the story, but your character only acts on what they have learned in it. What OOC means covers the notation.
Questions people ask
Is metagaming the same as godmodding?
No. Godmodding is taking control of outcomes or of someone else's character. Metagaming is acting on knowledge the character couldn't have. A message can do both.
Why did the character know my persona's secret?
Because it was written in text the model reads every reply. Remove the secret, or label it as hidden from that character and say what would expose it.
Can I tell the AI a twist without the character knowing?
Yes. Put it in a note labelled as hidden, and phrase it as something to build towards rather than something the character believes.
