AI Roleplay Formatting Conventions: OOC, Asterisks, Brackets, and Other Symbols Explained

Flat illustration of oversized asterisks, double parentheses, square brackets and quotation marks beside two chat bubbles, one with a wavy line for an action and one with a straight line for speech

Last updated: September 28, 2026 · Reading time: 9 minutes

Summary. In text roleplay, a small set of symbols separates what a character says from what a character does and from what the player wants to say outside the story. Asterisks (*like this*) or italics mark actions; quotation marks mark speech; double parentheses (((like this))) or square brackets ([OOC: like this]) mark out-of-character messages. These conventions predate AI roleplay by decades and are understood by most AI companion platforms because the language models behind them were trained on text that uses them. This page lists the symbols, the common abbreviations, the two main writing styles, and how AI characters interpret each.

Quick reference: roleplay symbols

Symbol	Name	What it means	Example
`text`	Asterisks	An action, gesture, or expression performed by a character	`she closes the book and looks up`
`_text_` or italics	Italics	Same as asterisks on platforms that render formatting; also used for thoughts	He wasn't sure that was true.
`"text"`	Quotation marks	Spoken dialogue	`"You're late again."`
`((text))`	Double parentheses	Out-of-character (OOC) message from the player	`((can we skip to the next morning?))`
`[text]` or `[OOC: text]`	Square brackets	Out-of-character message, or an instruction to the AI	`[OOC: keep replies shorter please]`
`(text)`	Single parentheses	Out-of-character on some platforms; ambiguous, avoid if `(( ))` is available	`(brb)`
`//text`	Double slash	Out-of-character, common on forums and Discord	`//I'll reply tomorrow`
`{text}`	Curly braces	Rare; used by some platforms for hidden instructions	`{the door is locked}`
`~text~`	Tildes	Sing-song or playful tone; informal, not universal	`~good morning~`
`...`	Ellipsis	A pause, hesitation, or trailing off	`"I... didn't mean that."`
`—` or `--`	Dash	Interruption or abrupt stop	`"Wait, I can—"`
`CAPS`	Capitals	Shouting	`"GET DOWN!"`
`> text`	Greater-than	Narration or player action in text-adventure style	`> open the drawer`
`-text-`	Hyphens	Action markers in some older communities; largely replaced by asterisks	`-nods-`

The first five rows cover more than 90% of what you will see in AI roleplay. The rest are worth recognising but not worth adopting unless the platform or partner already uses them.

Speech: quotation marks or plain text

Spoken dialogue is written either in quotation marks or as plain text, depending on the style. In script style (see below), plain text is assumed to be speech and only actions are marked. In paragraph style, speech is wrapped in double quotes exactly as in a novel, and everything outside the quotes is narration.

Script style:
Fine, I'll go. *grabs her coat*

Paragraph style:
"Fine, I'll go," she said, grabbing her coat.

Both are valid. The one rule that matters is consistency inside a single conversation: if speech is sometimes quoted and sometimes not, the AI has to guess, and it will guess wrong more often.

Actions: asterisks and italics

Action text describes what a character physically does, and it is conventionally wrapped in asterisks. The convention comes from early chat systems (IRC, AOL Instant Messenger, MUDs) where there was no formatting and *waves* was the only way to show an action. On platforms that render Markdown, a single pair of asterisks turns the text italic, which is why "asterisks" and "italics" now mean the same thing in most AI roleplay.

Three rules keep action text readable:

Third person, present tense is the default: *She sits down*, not *I sat down*. Many players use first person for their own character; either works if it is consistent.
One action per pair of asterisks. *sighs* "Fine." *sits down* is clearer than *sighs and says fine and sits down*.
Do not nest. *she said "*hi*"* breaks rendering on most platforms and confuses the model.

Thoughts: italics, single quotes, or labels

Internal thoughts have no single standard. The three common conventions are italics without asterisks on platforms that distinguish them, single quotation marks ('He's lying,' she thought), and an explicit label (Thought: he's lying). In paragraph style, thoughts are usually written as narration: She was fairly sure he was lying. If a platform uses asterisks for actions, avoid using them for thoughts as well; pick single quotes or plain narration instead.

Out of character: double parentheses and brackets

An out-of-character (OOC) message is written by the player as themselves, not as the character they are playing, and it is marked so that the other party does not respond to it in-scene. The two standard markers are double parentheses ((text)) and square brackets [text] or [OOC: text]. Double slashes //text are common on Discord and forums. Single parentheses are used on some platforms but are ambiguous because characters also use parentheses inside dialogue.

OOC messages are used for four things:

Instructions: ((please write shorter replies))
Corrections: ((my character has brown hair, not blonde))
Story steering: ((let's have someone knock on the door))
Pausing or ending: ((I need to stop here, continue tomorrow?))

The opposite of OOC is IC (in character): any message written as the character inside the story.

How AI characters handle OOC

Most language models recognise (( )) and [OOC: ] as meta-instructions and will follow them without breaking the scene, but this is a learned behaviour, not a guarantee. Three things improve reliability:

Put the OOC note in its own message, or at the very start or end of a message, not in the middle of dialogue.
Use the label OOC: inside the brackets the first time; the model then associates the bracket style with instructions for the rest of the conversation.
Keep the note short and concrete. ((shorter replies, no more than two paragraphs)) works better than a long explanation.

If a character replies to an OOC note in character (for example, the character asks "who are you talking to?"), regenerate the reply and restate the note with the OOC: label. On platforms with long-term memory, such as Varenchat, an OOC correction about a fact ("my character is 32, not 23") is worth stating once clearly, because the correction can be stored rather than needing to be repeated.

Narration and scene-setting

Narration is text that describes the setting, the passage of time, or events that are not one character's action. In paragraph style it is simply unmarked prose. In script style it is often put in asterisks like an action, or introduced with a label such as Scene: or Setting:. Text-adventure style uses a > prefix for the player's commands and plain text for the world's response.

Scene: A train platform, early evening, light rain.
*The 6:40 pulls in three minutes late.*
"You came."

Scene labels are useful at the start of a conversation and after a time skip, because they give the model a clear anchor for where and when the story is.

The two writing styles: script vs. paragraph

Script style (also called chat style or asterisk style) writes speech as plain text and marks actions with asterisks. It is fast, short, and the default on most mobile companion apps.

Paragraph style (also called novel style, literate, or prose style) writes everything as narrative prose, with speech in quotation marks and actions described in full sentences. It is slower, longer, and closer to reading fiction.

	Script style	Paragraph style
Speech	Plain text	In "quotation marks"
Actions	`in asterisks`	Described in prose
Typical length	1–3 lines	1–4 paragraphs
Perspective	Often first person for the player	Usually third person
Best for	Quick back-and-forth, banter, mobile	Slow burn, description, long scenes
Model behaviour	Mirrors short replies	Mirrors long, descriptive replies

The same moment in both styles:

Script:
You didn't tell me you were leaving. *crosses arms*

Paragraph:
"You didn't tell me you were leaving." She crossed her arms, and the 
kitchen felt suddenly smaller than it had a minute ago.

AI characters mirror the style they receive. A character will not stay in paragraph style if the player writes one-line script replies, and it will not stay short if the player writes four paragraphs. The fastest way to change a character's reply length or style is to change your own.

Community terms for length exist as well: one-liner (a single line), semi-lit or semi-literate (a short paragraph), lit or literate (one or more full paragraphs), and novella (several paragraphs per turn). These labels are used to set expectations between human partners and are less relevant with AI, which adapts to whatever length it is given.

Common roleplay abbreviations

Abbreviation	Meaning	Notes
RP	Roleplay
RPer	Roleplayer	A person who roleplays
IC	In character	A message from the character
OOC	Out of character	A message from the player
OC	Original character	A character created from scratch
CC	Canon character	A character from an existing franchise
NPC	Non-player character	A side character controlled by the narrator or AI
PC	Player character	The character a player controls
GM / DM	Game master / dungeon master	The person (or AI) running the world; from tabletop games
POV	Point of view	Whose perspective the scene is told from
AU	Alternate universe	A version of characters in a different setting
WIP	Work in progress
TBC	To be continued	Ends a scene to resume later
FTB	Fade to black	Skipping a scene rather than writing it
TW / CW	Trigger warning / content warning	A heads-up about sensitive content
Slow burn	—	A relationship or plot that develops gradually
Fluff	—	Light, comforting, low-conflict scenes
Angst	—	Emotionally heavy scenes
Hurt/comfort (H/C)	—	One character is hurt and another cares for them
Canon	—	What is officially true in a story or franchise
Headcanon	—	A personal interpretation not confirmed in canon
Retcon	Retroactive continuity	Changing something already established
Timeskip	—	A jump forward in story time
Godmodding	—	Making one's character invincible or all-powerful; considered bad form
Autoing / autohitting	—	Deciding the outcome of an action on another character without their input
Powerplaying	—	Controlling another player's character; overlaps with autoing
Metagaming	—	Using knowledge the character could not have
Bump / Nudge	—	A reminder that a reply is owed (human RP only)
Semi-lit / Lit	Semi-literate / literate	Reply-length expectations; see above
Multi-para	Multi-paragraph	Several paragraphs per reply

Godmodding, autoing, powerplaying, and metagaming come from human-to-human roleplay, where they describe one player being unfair to another. With an AI partner they still matter, in reverse: an AI character that decides your character's actions for you is "autoing," and the standard fix is an OOC note such as ((please don't control my character's actions)).

Platform notes

Conventions are mostly portable across AI roleplay platforms, with a few practical differences.

Platforms that render Markdown (most web apps) will show *text* as italics; platforms that do not will show the asterisks literally. Both are readable; the model treats them the same.
Some platforms reserve square brackets for system features (for example, memory tags or commands). If [text] produces unexpected behaviour, switch to ((text)) for OOC.
On Varenchat, characters follow whichever style the user writes in. The recommended setup is quotation marks for speech, asterisks or italics for actions, and ((OOC: text)) for out-of-character notes. Because Varenchat characters carry conversation details forward across sessions, a formatting preference stated once in an OOC note ("keep replies to two paragraphs") tends to hold without being repeated.
Character greetings show the platform's preferred style. Matching the greeting's format in your first reply is the simplest way to get consistent formatting for the rest of the conversation.

A minimal set of rules

If you remember nothing else from this page, these four conventions cover almost every situation:

Speech in "quotation marks" or as plain text; pick one.
Actions in *asterisks*, third person, one action per pair.
Out-of-character notes in ((double parentheses)), with OOC: the first time.
Match the character's greeting style, and change your own style if you want the character's to change.

Frequently asked questions

What does OOC mean in roleplay? OOC stands for "out of character." It marks a message written by the player as themselves rather than as their character, usually to give an instruction, make a correction, or steer the story. It is typically written as ((text)) or [OOC: text]. The opposite is IC, "in character."

What do asterisks mean in roleplay? Text wrapped in asterisks, such as *she nods*, describes an action, gesture, or expression performed by a character, as opposed to something the character says. The convention comes from early text chat systems where asterisks were the only way to show an action. On platforms that render Markdown, asterisks display as italics.

Should I use asterisks or quotation marks? Use both, for different things: quotation marks for spoken words and asterisks for actions. If you prefer script style, you can leave speech as plain text and use asterisks only for actions. The important rule is to be consistent within one conversation.

Do AI characters understand ((double parentheses))? Usually, yes. Most language models have seen the (( )) convention in training data and treat it as a message from the player rather than the character. Adding the label OOC: inside the parentheses the first time makes this more reliable. If a character responds to an OOC note in character, regenerate the reply and restate the note.

What is the difference between script style and paragraph style? Script style writes speech as plain text and actions in asterisks, in short replies. Paragraph style writes everything as narrative prose with speech in quotation marks, in longer replies. AI characters mirror whichever style the player uses.

What is godmodding? Godmodding is making one's own character invincible, all-knowing, or immune to consequences, or deciding the outcome of events for other characters. It comes from human roleplay etiquette. With an AI partner, the equivalent problem is the AI deciding your character's actions; the fix is an OOC note asking it not to.

How do I change the length of an AI character's replies? Change the length of your own replies first, since characters mirror the player's style. If that is not enough, send a short OOC note such as ((please keep replies to two paragraphs)). On platforms with long-term memory, one clear note is usually sufficient.

This page is maintained by the Varenchat team and updated when conventions or platform behaviour change. To cite: Varenchat, "AI Roleplay Formatting Conventions: OOC, Asterisks, Brackets, and Other Symbols Explained," varenchat.ai/blog/ai-roleplay-formatting-ooc-asterisks, updated September 28, 2026.