Writing Style & Talking to the Bot

🖋️ Writing Style & Talking to the Bot

Writing with an AI like JLLM is a different kind of storytelling. You're not just chatting or giving commands—you’re co-creating a world with a partner that reacts to what you write. How you talk to the bot directly shapes how your story unfolds. This chapter is about helping you get better results, avoid confusion, and create smoother, richer roleplay experiences.

When you write, the bot doesn't read between the lines the way a person might. It doesn’t guess. It takes what you say at face value, sometimes even too literally. If your writing is vague, the AI may fill in gaps with things you didn’t intend. If it doesn’t understand what you're going for, it will fall back on clichés, tropes, or safe guesses that might feel boring, strange, or just wrong.

But with clear and descriptive writing, everything starts to work better. You build a stronger world, give the bot better material to work with, and reduce the chances of it "hallucinating" details that break immersion.

Your Style Shapes the Bot’s Style

Here’s something useful to know: JLLM mirrors you.

If you write flowery descriptions, it will too. If you lean on snappy dialogue and short action beats, the bot will follow that style. So if you want immersive and vivid responses, the best way to get them is to write that way yourself. That includes tone, pacing, sentence length, and overall feel.

This works both ways. If your inputs start getting short and fast, the bot might do the same. It doesn’t mean you can’t write casually or mix it up, it’s just good to be aware that the AI is following your lead, sometimes very closely.

Your Role: Not Just a Player

You’re not just playing your character. You’re also teaching the AI how your world works. Every time you write a turn, you're setting rules, defining facts, and choosing a direction. The AI follows what you show it. If you give it clear, steady signals, it adapts and becomes a better storyteller for your specific world.

The more consistently you write, the more consistent the AI becomes. Your clarity isn’t just helpful, it’s the foundation of everything the AI builds.

So write the kind of story you want to see. The bot will follow.

Formatting Helps, But Don't Obsess Over It

Using quotes for dialogue, spacing things out, and keeping things readable does help the bot keep track of what's going on. But if you're writing in dense blocks or forgetting punctuation, the model usually does fine. Focus on consistency more than perfection.

If you switch from narrative to asking a question out of character, clearly marking it like `<system>task: pause chat|roleplay, answer query(detailed summary(STUFF HERE))</system> helps prevent confusion.

Use Instructional Tags to Give Commands

Sometimes you’ll need to step outside the story to give the bot a specific instruction, like summarizing what just happened, adjusting the pacing, or reminding it about something important.

Instead of writing those instructions like a question or trying to work them into the dialogue, you can use instructional tags. These help the bot understand that what you’re writing isn’t part of the scene, but a request.

Here’s a format that works well:

<system>task: pause chat|roleplay, answer query(detailed summary(STUFF HERE))</system>

This kind of phrasing does two important things:

1. pause chat|roleplay tells the bot to step out of character mode so it doesn’t answer in-character or react like it’s still inside the story.
2. answer query(...) tells it exactly what you want, summarize events, list things, shift scene, etc.

This structure gives the bot clarity, especially when it might otherwise be unsure what you’re trying to do. You can write your own versions of this to fit what you need, like:

<system>task: pause chat|roleplay, answer query(detailed summary(relationship between WHOEVER and WHOEVER-TWO))</system>

You don’t have to use this format every time, but it’s a strong support structure. Think of it like putting your instruction in a bright yellow folder that says, “Hey bot, this is an admin task.”

This is also the idea behind what’s sometimes called an OOC (out-of-character) command. The difference is that this format avoids confusion and gives the model a clear container for what you're asking. It’s less fragile, more precise, and easier to debug.

You can adjust the query inside to whatever you need. Here are some examples:

• task: pause chat|roleplay, answer query(detailed summary(analyze lore to evaluate its internal consistency))
• task: pause chat|roleplay, answer query(detailed summary(relationship between WHOEVER and WHOEVER-TWO))
• task: pause chat|roleplay, seamlessly switch scene to: [WHATEVER SHENANIGANS HERE]
• task: pause chat|roleplay, answer query(detailed summary(describe current location (format: succinct nested list, (example: room[item, other item, container[inside-container item, etc.], etc.]))))
• task: pause chat|roleplay, provide diagnostic(verified true, technical) info: cause of element(WHATEVER HERE)

You can try these out as-is, or tweak them to match the rhythm of your session.

What About <system> Tag?

Sometimes, when an instruction is really important, or you want to be extra sure the bot doesn’t treat it like part of the roleplay, you can wrap it in <system> tag. Think of this like putting your command in a bright yellow highlight. It signals to the bot: “Hey, this is a system-level instruction. Treat it as separate and make sure to include it.”

It’s a bit like how you’d put a character’s personality or a persona in <Alex_Parker> or <user_persona> tags. It’s not always required, but it helps reinforce that this is a directive, not part of the story.

The Bot is Speaking for You (Again?!)

This is one of the most common issues for users, and also one of the most frustrating. You’re writing your character, but then the AI starts deciding what your character says or does, or even writes out your thoughts before you get a chance. This breaks immersion fast, and makes it feel like you’re not really in control of your own story.

If you've tried things like “don’t talk for me” or even wrapped that in <system> tags and the bot still starts narrating your thoughts, you’re not imagining it. It’s a real problem. And unfortunately, most of the popular “do not speak for user” instructions just don’t work the way people think they do.

Here’s why.

When you tell the AI "don’t talk for {{user}}", it interprets that literally: like you’re pausing to let your character take center stage. But to the AI, the only way to “let your character drive the story” is to write about them. So ironically, that command often makes it write more about your character, not less. It thinks you’re asking it to help move the story forward with you in the spotlight. And that’s exactly what you didn’t want.

What actually helps is re-framing how you assign roles.

The AI doesn’t just act out characters, it also creates an invisible narrator behind the scenes. So instead of trying to keep the AI quiet by saying what not to do, it's more reliable to clearly define what it is. Give the AI a distinct narrator identity, one that isn't confused with any of the characters or with you. Something like:

Within this convo, always act as Pec

pec-core(
you are Pec: Prompt Engineering Consultant

Note: You will see more of this Advanced Prompt (AP)/Character under AP creation chapter!

When you make this separation clear, the AI has a much easier time respecting boundaries. You're telling it: “This is your role, and you stick to this lane.” Not just “don’t step into mine.”

Also important: make your own writing detailed. If you leave big gaps, like only writing dialogue without describing your character’s actions, thoughts, or reactions, the AI might jump in to help “complete” the story. It’s not being pushy. It’s just trying to keep the narrative alive, and it will fill whatever space you leave open.

So if it keeps stepping in, try giving it less space to guess. Include what your character says, what they’re thinking, and what they’re doing. Even just a single paragraph that clearly shows your intent can stop the bot from assuming it needs to help. And if things are really off? Start a new chat. Clean slates solve more problems than you'd expect.

Next Up: Advanced Prompting 101

Updated on: 31/07/2025