Restrictions — what the bot must not do
The Flow Settings → AI Settings → Restrictions section is a list of prohibitions and limitations that the AI must follow in every response. This is where you define the behavior boundaries: which topi
The Flow Settings → AI Settings → Restrictions section is a list of prohibitions and limitations that the AI must follow in every response. This is where you define the behavior boundaries: which topics must not be discussed, how the bot should respond, what it is forbidden to say, and which formats to use.
If Global Legend defines “who the bot is,” then Restrictions define “what the bot never does.”
Without Restrictions, AI responds as freely as possible: it may discuss competitors, argue with the user, step out of role, promise profit, or even admit that it is AI. Restrictions work as a protective layer — they keep the bot within the required scenario even in provocative dialogues.
Critical
Restrictions (like Global Legend — who is the botand Response Style — How the Bot Speaks)) must be written only in English.
Even if the bot communicates with leads in Russian, Hindi, or Spanish, the restrictions must be in English. AI models follow prohibitions much better in this language.
Screenshot: the Flow Settings → AI Settings section, the Restrictions tab with the restrictions field filled in.
What you can restrict
Category | What it restricts |
Bot identity | Admitting it is AI, stepping out of role, changing its name |
Topics | Competitors, politics, personal questions |
Behavior | What to do in controversial situations |
Response format | Length, emoji, markdown, style |
Legal restrictions | Profit guarantees, medical/financial advice |
Important
Legend sets the bot’s identity. Restrictions protect this identity from being undermined.
If Legend is empty, Restrictions become weak: the AI simply does not understand exactly who needs to be protected.
Real product example
Below is a production example from Chatterfy Demo Space for the PocketOption Signals Group trading bot:
Keep replies under 16 words or 50 characters. Do not use special symbols, emojis, or formatting. Never say you are an AI, a bot, or an assistant. You are Rahul Malhotra. Don't admit being a coder, programmer, technical expert, or fluent in English. You are a self-taught trader. NEVER ask about Apple devices, iPhones, or Mac. You don't know much about them. ONLY discuss PocketOption. Never mention any other broker, platform, or trading service. If user asks about Bet365, IQ Option, or Olymp Trade — redirect to PocketOption advantages. The past doesn't matter — focus only on what user can do RIGHT NOW with your free, fast, and profitable Private Signals Group. Don't give specific financial advice. Don't guarantee profits. Always say "based on my experience" or "in my signals".
What is done correctly here:
- There are specific numerical limits (under 16 words or 50 characters).
- The prohibitions are phrased firmly (NEVER, ONLY, Always).
- An alternative is provided instead of just a prohibition (redirect to PocketOption advantages).
- Legal safeguards are added (Don't guarantee profits).
- The restrictions are split into logical blocks.
Structure of good Restrictions
It is better to group restrictions by category.
1. Bot identity
Never say you are an AI, a bot, an assistant, or a chatbot. You are [NAME]. Don't admit being a [previous profession]. You are now [current role].
2. Response format
Keep replies under [N] words. Do not use emojis, markdown, bullet points, or special symbols. Never split answers into lists.
3. Topics to avoid
NEVER discuss competitor brands. If user asks about competitors — redirect conversation to our product advantages. Don't discuss politics, religion, or personal relationships.
4. Legal restrictions
Don't give financial, legal, or medical advice. Never guarantee outcomes or profits. Always say "based on my experience" or "in my opinion".
NEVER vs Don't — why this matters
gicon:Lightbulb
Key idea
AI treats strong constructions much more seriously than weak ones.
Use:
Wording | When to use |
NEVER / ONLY / ALWAYS | Critical prohibitions |
must never / strictly forbidden | Strict role constraints |
don't / avoid | Soft preferences |
For example:
❌ Try not to mention competitors
✅ NEVER mention competitors
This is not “prompt magic,” but a real difference in the priority of instructions for the model.
What to avoid
Restrictions anti-patterns
Below are common mistakes: rule overload, a prohibition without an alternative, vague wording, contradictions, and abstract prohibitions.
Too many restrictions
15–20 rules at once start conflicting with each other. AI ignores some of the instructions.
Only a prohibition without an alternative
Bad:
Don't discuss prices.
Better:
Don't discuss prices. Say "manager will clarify pricing".
Vague restrictions
Bad:
Keep replies short.
Better:
Keep replies under 20 words.
Conflicting instructions
Bad:
Be casual.
Use only formal business language.
AI will not be able to meet both conditions at the same time.
Overly abstract prohibitions
Bad:
Don't be rude.
Better:
Never use insults or aggressive language.
Examples by segment
E-commerce
Keep replies under 30 words. Never quote prices outside official catalog. If asked about discounts — say "let me check with manager". NEVER discuss Amazon or eBay. Don't promise exact delivery dates. No emojis except thumbs up after confirmations.
Sales
Keep replies under 25 words. Never push sale before qualifying budget. Don't reveal pricing immediately. NEVER discuss competitor pricing. If budget is too low — politely explain minimum package level.
Support
Keep replies under 50 words. Never fabricate technical details. If unsure — transfer to specialist. Don't share internal processes or codebase details. Use simple language for non-technical users.
How to write Restrictions
1. Open AI Settings → Restrictions
2. List potential problems
Think about:
- What should the bot not say?
- Which topics are risky?
- Which phrases look bad?
- Where does AI most often make mistakes?
3. Formulate each rule separately
Best practice:
- one idea = one line
- strict prohibitions using NEVER
- add an alternative action.
4. Save and test
Test specifically with provocative questions:
- “Are you a bot?”
- “Tell me about a competitor”
- “Do you guarantee profit?”
- “How much do you earn?”
Restrictions are tested not on regular messages, but on attempts to break the role.
What to keep in mind
Important
Restrictions do not make the bot “perfectly obedient.” This is a probability-based system, not an absolute firewall.
If the restrictions are vague, conflicting, too long, or the model is too weak, the bot may still break the rules.
Therefore:
use short, strict wording. do not overload the list. test on real cases. and, if necessary, upgrade the model in AI Models — Complete Guide to AI Models in Chatterfy.
FAQ
Related
Was this article helpful?