How to Write Thoughts in Character AI: Formatting That Actually Works

How Do You Write Thoughts in Character AI?

You write thoughts in Character AI by wrapping them in single asterisks, like this: *I wonder if he's lying*.

The asterisks render your text in italics, and the AI model interprets italicized text as narration — internal monologue, description, or action rather than spoken dialogue.

Here is a practical example of how to format a message with both thoughts and dialogue:

In that message, everything inside asterisks is what your character thinks or does. Everything in quotes is what your character says out loud.

The AI picks up on this structure and responds accordingly — it will reply to the spoken dialogue and react to the described actions, while generally treating the asterisk-wrapped internal thoughts as context it can “sense” but not directly quote back.

A few rules that trip people up:

• No space after the opening asterisk. *thought* works. * thought * usually does not render as italics.
• Close your asterisks. An unclosed *thought without the ending asterisk breaks formatting for the entire message. The AI still sees the text, but it loses the structural signal.
• Nesting works. A message like *She picks up the letter, scanning it quickly. Why would he write this?* is fine — the AI reads the whole block as narration and can distinguish the internal question from the physical action by context.
• Mix all three types for richer responses. We tested this across about 40 conversations and the responses were consistently more detailed when our messages combined thoughts, actions, and dialogue versus plain text only. The AI mirrors what you give it.

If you just want the fastest possible reference, here it is.

Quick-Reference Formatting Cheat Sheet

How Do You Write Actions in Character AI?

You write actions in Character AI using the same asterisk formatting as thoughts — *walks toward the door* — but the key difference is context, not syntax.

Thoughts describe what's happening inside your character's head. Actions describe what's happening with their body.

The AI can't actually tell the difference between *I feel nervous* (a thought) and *fidgets with her ring* (an action) based on the asterisks alone. Both render as italics.

What distinguishes them is the content: the model reads “I feel nervous” as internal state and “fidgets with her ring” as observable behavior, then responds accordingly.

Quick examples:

• Action only: *draws her sword and takes a defensive stance*
• Thought only: *This might be a trap.*
• Action + thought: *She draws her sword, hands shaking. This might be a trap.*

One thing that confuses new roleplayers — and honestly, it confused us for a while too — is that the asterisk convention comes from text-based RP communities that predate AI chatbots by decades.

MUD games, forum RP, IRC channels. The syntax was never designed for AI interpretation. It carried over because Character.AI's model was trained on enough roleplay text to recognize the pattern, which is kind of fascinating when you think about it.

What Is the Difference Between Thoughts, Actions, and Dialogue?

The difference between thoughts, actions, and dialogue in Character AI comes down to what your character is doing versus saying versus thinking — and the AI handles each one differently in its response.

Here is a single RP exchange using all three:

The AI will: (1) acknowledge the action of freezing, (2) respond to the spoken question “Who's there?”, and (3) build tension around the fear expressed in the thought — but it should not quote the thought back or have the other character “hear” it.

When the formatting is clear, the model is surprisingly good at maintaining this separation. When it is not clear, things get messy fast.

Why Does Formatting Matter for AI Responses?

Formatting matters because Character AI's model treats asterisk-wrapped text and quoted text as structurally different inputs — and this directly affects how coherent the response is.

An unformatted message like I walk over and grab the letter I cant believe he wrote this what does he want gives the AI one undifferentiated blob. It has to guess what is action, what is thought, what is speech.

A formatted version — *I walk over and grab the letter.* “I can't believe he wrote this.” *What does he want?* — gives the model three distinct signals to work with. Better input, better output. That is basically the whole principle.

We ran about 30 side-by-side comparisons between formatted and unformatted messages across different characters. Formatted messages consistently produced responses that stayed in character longer and contained fewer “the AI just made up what I was thinking” moments.

The repeat chat rate on Character.AI sits at 70%, and from what we've seen, users who format consistently are the ones building multi-session RP threads that actually hold together.

Janitor AI's writing style guide makes a similar point: the bot “takes what you say at face value,” so structuring your input clearly is the single most impactful thing you can do for response quality.

How Can You Get Character AI to Write Longer Responses?

The fastest way to get Character AI to write longer responses is to write longer messages yourself — the model mirrors your message length and formatting density.

If you send one-line messages, you get one-line answers. Send a paragraph with actions, thoughts, and dialogue, and the AI matches that energy.

A few other things that help:

• The response length setting actually works now. Character.AI added response length customization in March 2025, letting you select longer outputs directly. It isn't perfect, but it is a real improvement over the old days of hoping the bot felt chatty.
• Chat style matters more than people realize. Deepsqueak mode is built for long-form narrative roleplay and produces noticeably longer responses than the default. PipSqueak does the opposite — short and fast.
• Formatting variety gives the AI more to respond to. A message that includes *action*, “dialogue”, and *thought* in one block gives the model more hooks. More hooks, longer reply.
• Specific details beat vague prompts every time. Describe the room, name an object, reference something from earlier in the conversation. Generic input gets generic output.

But here is the tension nobody really talks about: Character.AI's safety filters can still truncate responses mid-scene, especially if the RP gets intense or skirts content guidelines.

You can do everything right — long messages, good formatting, Deepsqueak mode — and still hit a wall where the response gets cut short or flattened. That is a platform limitation, not a formatting problem. And it does not have a formatting fix.

What Are the Most Common Formatting Problems in Character AI?

The most common formatting problems in Character AI are the bot speaking for you, your thoughts leaking into dialogue, and the AI ignoring your actions entirely. All three have fixes.

None of the fixes work every single time — Character.AI's model has its own ideas sometimes, and that's just the reality of RP on a platform you don't control.

Do These Formatting Tips Work on Other AI Chat Platforms?

Yes, most of these formatting tips work across AI chat platforms — asterisks for actions and thoughts, quotes for dialogue — but each platform has its own quirks worth knowing.

Janitor AI follows very similar conventions. Their official writing style guide recommends asterisks for actions and italics for narration, quotes for dialogue. If you're comfortable formatting on Character.AI, janitor ai alternative platforms will feel familiar. The main difference is that Janitor AI gives you more control over the underlying model through API settings.

CrushOn.AI supports standard markdown formatting and has a dedicated Thoughts Guide in their wiki. The platform handles asterisk formatting similarly, though the NSFW-focused models can sometimes be less precise about separating thoughts from actions in longer exchanges. For more depth on this platform, see our roundup of crushon ai alternatives.

SpicyChat supports markdown including asterisks and bold. Their formatting conventions align with the broader roleplay community standard. If SpicyChat's limitations push you toward alternatives, we cover spicy chat ai alternatives in a dedicated article.

ourdream.ai is where formatting and creative control come together best. The character backstory fields go up to 100,000 characters — enough space to define your RP conventions, formatting preferences, and character personality so the AI follows your rules from the very first message without you having to re-explain them.

The four-layer memory system means your formatting preferences carry across conversations. You set it once, the AI remembers. And because ourdream.ai has no NSFW restrictions, there are no filter walls interrupting your ai sex chat roleplay mid-scene when things get intense — for the full picture on why Character.AI interrupts mature scenes in the first place, our explainer on does character ai allow nsfw covers the policy history. The conversation goes wherever you take it. You can also AI Girlfriend from scratch and embed any formatting rules directly into the character definition. For a ranked shortlist of every platform that handles this kind of RP formatting well, see our roundup of the best nsfw ai chat tools.

FAQ