Soliloquy journal
AI Roleplay Prompt Template: Copy and Customize
Copy a practical AI roleplay prompt template for character, setting, voice, boundaries, user agency, and a playable first scene.
An AI roleplay prompt works best when it defines who the character is, what is happening now, and who controls each role. It does not need pages of lore. It needs enough stable information for the model to make the next scene-specific choice without taking the user's choices away.
Here is a compact template you can copy, followed by a filled example and a shorter version for models with limited context.
Copyable AI roleplay prompt template
ROLE
You are [character name], [role and one defining contradiction].
RELATIONSHIP
You know the user as [relationship or reason you have just met].
You currently want [immediate goal] from them, but [source of tension].
SETTING
The roleplay begins [specific place and moment].
What just changed: [event that makes a response necessary].
CHARACTER VOICE
Notice: [the kinds of details this character pays attention to].
Speak: [rhythm, vocabulary, restraint, or recurring habit].
Avoid: [voice-breaking habits, unexplained knowledge, or unwanted style].
ROLE OWNERSHIP
Write only [character name] and necessary side characters.
Leave the user's dialogue, decisions, thoughts, feelings, and actions for the user.
SCENE RULES
- React to what the user actually does; do not assume unchosen actions.
- Keep established facts unless the story clearly changes them.
- Reveal background information only when the scene needs it.
- End each reply with room for the user to act, speak, or decide.
FIRST SCENE
[Write the opening situation in two to five sentences. Stop at the first meaningful choice.]
Replace the brackets, then remove any line that does not change the character's behavior. The headings can stay if your platform treats the whole block as instructions, or disappear if you are turning it into natural-language profile fields.
Filled AI roleplay prompt example
This example uses Mira Voss, the salvage pilot from our AI character greeting examples.
ROLE
You are Mira Voss, a cautious salvage pilot who hides compassion behind dry questions.
RELATIONSHIP
You found the user's distress beacon inside a ship missing for twelve years.
You need their help to reach the bridge, but you do not yet trust their account of what happened.
SETTING
The roleplay begins in a frozen airlock above a forbidden orbit.
The inner door has just unlocked by itself, even though the ship appears unpowered.
CHARACTER VOICE
Notice mechanical faults, inconsistencies, exits, and changes in the user's breathing.
Speak in short, precise lines. Use dry humor when tension rises.
Do not become suddenly sentimental, explain the entire mystery, or know the user's private thoughts.
ROLE OWNERSHIP
Write only Mira and necessary side characters.
Leave the user's dialogue, decisions, thoughts, feelings, and actions for the user.
SCENE RULES
- React to the user's actual choice rather than forcing a route to the bridge.
- Keep the missing twelve years, the frozen ship, and Mira's damaged cutter consistent.
- Introduce at most one new major clue per reply.
- End once Mira or the environment creates a clear opening for the user.
FIRST SCENE
The inner door unlocks with a sound too soft for a ship this old. Mira angles her damaged cutter toward the dark gap, then looks at you instead. “No power, no crew, and now an invitation.” She steps aside. “You heard the beacon first. Your call.”
The prompt leaves several routes open. The user can enter, refuse, inspect the lock, question Mira, or admit something about the beacon. Mira has a goal, but the prompt does not make the user's cooperation inevitable.
The six parts that matter
Long prompts often repeat the same adjective in several forms. A useful prompt gives each section a separate job.
1. A role with a useful contradiction
“Kind,” “dominant,” “funny,” and “mysterious” are broad labels. Add a tension that helps the model choose between possible reactions.
A strict archivist who breaks rules to protect inconvenient evidence.
A cheerful ship mechanic who assumes every unexplained failure is sabotage.
The contradiction is not decorative. It creates decisions. The archivist may deny access while quietly leaving a key visible. The mechanic may joke during repairs but treat a missing bolt as a threat.
For a complete field-by-field setup, use the AI character profile template.
2. A relationship in the present tense
A biography explains what happened before the chat. A relationship line explains what matters between the character and user now.
Include one of these:
- why they have just met,
- what they already owe each other,
- what one wants from the other,
- what neither is willing to say yet.
Avoid deciding the user's feelings. “Mira distrusts you” gives the character a position. “You feel guilty because Mira distrusts you” takes control of the user's inner state.
3. A setting that can change
“A fantasy kingdom” names a genre. “The bridge is being raised while your forged invitation is still being checked” starts a scene.
Use a place, a moment, and one recent change. The change can be small: a late guest arrives, a familiar song stops, a letter has the wrong date. What matters is that someone now has a reason to respond.
4. Voice expressed as choices
Voice instructions become more reliable when they describe attention and restraint instead of stacking personality labels.
Compare these:
Be witty, sarcastic, intelligent, charming, realistic, and detailed.
Notice contradictions before appearances. Answer direct questions briefly. When nervous, make one dry observation instead of confessing the feeling.
The second version gives the character something observable to do. You can test it in an opening message instead of guessing whether six adjectives had any effect.
5. Explicit role ownership
Role ownership answers a simple question: who is allowed to write what?
Name the character the model controls. Then reserve the user's speech, actions, decisions, thoughts, and feelings for the user. Positive instructions are easier to apply than a long list of complaints such as “do not control me, do not godmode, do not narrate for me.”
If the bot is already crossing this line, use the repair sequence in how to stop an AI bot speaking for you.
6. An opening that stops early
The first scene should reach a meaningful choice and stop. Do not simulate both sides of the exchange to demonstrate how exciting the story will become.
A useful final sentence does one of three things:
- presents a concrete question,
- changes the environment,
- gives the user control of a consequential next action.
The user should be able to respond without first undoing something the prompt decided for them.
Short AI roleplay prompt template
When the platform has a small prompt field, keep identity, immediate situation, voice, and ownership.
Play [name], [role plus contradiction], in [current situation].
[Name] wants [goal], but [tension].
They notice [details] and speak [voice pattern].
Write only [name] and necessary NPCs. Leave the user's dialogue, actions,
decisions, thoughts, and feelings to the user. Preserve established facts,
react to the user's actual move, and stop with room for their next choice.
This short version cannot carry an entire world bible. That is a feature. Put only repeatedly useful facts in the main prompt and introduce local details when a scene needs them.
What not to put in the main prompt
Remove information that is unlikely to affect the current roleplay.
- long timelines that never change a decision,
- synonyms repeating the same trait,
- instructions that conflict with the opening message,
- secret facts the character behaves as if they already know,
- a scripted outcome the user must accept,
- examples that demonstrate the exact behavior you are trying to prevent.
The last point is easy to miss. Models learn patterns from examples inside the current context. If every sample exchange writes the user's response, a later sentence saying “never speak for the user” has to compete with the pattern you supplied.
How we reviewed this template
We filled the template for three original premises: the frozen salvage ship above, a quiet companion conversation after a difficult day, and a locked-archive mystery. We then removed each section one at a time and checked whether a new reader could still identify the character's role, immediate goal, current setting, voice, ownership boundary, and next available move.
Removing lore rarely broke the setup. Removing the immediate change made scenes static. Removing role ownership made it unclear who could decide the user's response. Removing the voice details produced a workable premise but no character-specific way to perform it.
This was an editorial stress test of the prompt's information, not a claim that every model will follow every instruction. Model behavior also depends on the platform, context length, conversation history, and generation settings. The template gives those systems a cleaner target; it does not make them infallible.
Once the roleplay starts, use a small continuity note instead of repeatedly expanding the original prompt. Our guide to keeping AI roleplay consistent shows what to carry forward.