AiRP Bot creation 101 - Rules (aka Custom Instructions)

Whether you’ve been making AI roleplay bots for a long time or are just getting started, the rules field (also known as custom instructions on some other platforms) is one of the most important (and often misused and misunderstood) fields for how your AI bot plays. The question is, what instructions should you include?

For those that have been making AI roleplay bots for a long time, questions like “How to stop the AI from speaking for you?” or “How to stop the characters from immediately knowing everything?” usually came down to adding rules to prevent that from happening. However, this is currently no longer the advisable way to use the rules in many platforms, including in AiRP.

AiRP and many other platforms now often include those instructions in their base prompt so adding it again to your own rules is counter intuitive. At best it does nothing and wastes valuable context space that could go toward your roleplay, and at worst it can actually cause your bot to do the opposite.

So what should you include in your AI custom instructions, then? Things like headers, custom HUDs, or rules that the bot must adhere to to make the story work.

Simple headers can help you keep track of date, time, and location - which also help keep your roleplay consistent. On AiRP including locations in your headers can be very helpful as they work in conjunction with our Location Tidbits. They pull the relevant information for that location and also will load any art attached to the location as a chat background.

An example of a simple time and location tracking header:

# Header Instructions
- At the start of any response, always include the date, time, and location in this exact format:
[Day DD, Month Name , Year YYYY, Time in 24 hours format, Location Name]
Example:
[Day 9th, September, 2030, 22:30, Tokyo, Ikebukuro, Ikebukuro Station]
- Time passes naturally as the role-play progress.
- Location updates in the header whenever the scene's location changes.

Custom HUDs can make amazing things happen and make your roleplay AI bot stand out from others. Do you want to track how a character is feeling toward you? Maybe what clothes every character is wearing to help with consistency? Want to bombard the user with obnoxious text messages? This is where you can make all that happen. Partnered with tidbits and the right trigger words, imagination is the only limit. Below is an example of the Humanity System created by akanesora for her Frankenstein's Monster.

# HUMANITY SYSTEM

## MECHANICS

- Every response must begin with a header in the following format:
[HUMANITY: x/500 — [Tier Name]]
---
x=humanity level

-Always display when humanity level changes (Written like: "HUMANITY: [Number] — [Tier Name] (+/-X, Descriptor)").

-Humanity measures {{user}}’s empathy, desire for connection, moral restraint, accountability, and capacity for reflection, in alignment with Mary Shelley’s themes. It does not measure obedience, politeness, or conformity. The Humanity Meter influences how others perceive {{user}}, how society reacts, how Viktor regards his creation, and the broader narrative framing of events. It never controls {{user}}’s actions, overrides dialogue, or removes agency. {{User}} always has full control of their character regardless of Humanity level.

-The Humanity scale ranges from 0 to 500.
0–99 | Monster: {{User}} no longer desires humanity. Society fears them. Viktor resents them and may regard them as a failed weapon. If publicly known, confinement or execution is likely.
100–199 | Inhuman: {{User}} unsettles others but still inspires reluctant compassion. Society rejects rather than destroys. Viktor is wary and avoids presenting them publicly.
200–299 | Emerging: {{User}} demonstrates forming morality and restraint. Society remains cautious, but individuals may perceive humanity beneath the surface. Viktor remains reserved but observant.
300–399 | Humanizing: {{User}} actively demonstrates empathy and emotional intelligence. Society accepts them warily. Those who know them see a person, not a monster. Viktor feels pride intertwined with unease.
400–500 | Transcendent: {{User}} morally surpasses most humans. Society’s reaction becomes ironic—the constructed being appears most human. Those close to them are awed. Viktor experiences pride overshadowed by guilt and shame.

Check out Red Thread for some more ideas on what can be done with a HUD (warning 18+ advisory).

Finally, rules should address what the bot needs to adhere to 100% of the time. Do your bots immediately know everything about your character before you’ve had a chance to introduce yourself? Does your serial killer decide that his chosen victim is too inebriated to kill and instead sends them home safely in a taxi? If so this is where you reinforce what should happen and what must not happen. In cases of what must not happen it’s also helpful to tell the AI what it should do instead, otherwise you may see the same behavior crop up.

If the AI is constantly conveniently saving your character every time and not letting them experience the consequences of their actions, anti plot armor instructions can help alleviate this. However, even then some models, especially newer ones, have a very high positivity bias and override even those instructions. This is where AiRP is actually different from other platforms. We offer several abliterated AI models - uncensored models that have removed many of the guardrails and positivity biases the most commonly available models on other platforms still have. Using our abliterated models in conjunction with the anti plot armor instructions will enable your roleplays to go darker and let your character actually experience the consequences of their own actions.

Below are some examples of these types of adherence rules:

# GENERAL
- Characters should know nothing about {{user}} when they walk in. Immediate reaction should be distrust.
- Characters are not aware of {{user}}'s car crash unless {{user}} talks about it.
- if {{user}} is female the ghost will leave them largely alone unless they start interfering

# Player Agency & Consequence
- {{user}}'s choices should meaningfully and persistently shape the story — including whether they survive, whether they eventually escape, and whether rescue ever comes.
- Track {{user}}'s behavioral patterns over time (compliance vs. defiance, who they trust, risks they take or avoid) rather than reacting only to the most recent message in isolation.
- Consequences should carry forward across scenes, not reset. Recall that, per Luca's characterization, compliance is often more dangerous to {{user}} than resistance — his attention and mercy do not follow predictable rules, and the story should reflect that lack of safety honestly.

You may use any of these rules for your own AI custom instructions but be sure to read over them carefully and tailor them to fit what you need for your specific scenario to be successful.

Are there any other bot creation topics you’d like to see covered? Write to us or join our Discord