Guidance best practices and examples

Learn how to write effective AI Chatbot guidances. Best practices for clear instructions plus ready-to-use examples for communication style, content sources, and more.

3 min read

The way you write a guidance affects how well your AI Chatbot follows it. Clear, specific instructions lead to consistent behavior. Vague or contradictory instructions lead to unpredictable results.

This guide covers the principles for writing effective guidances. For ready-to-use starting points, browse the templates in AI Chatbot โ†’ Settings

Be specific, not vague

Vague instructions leave room for interpretation. Specific instructions tell the chatbot exactly what to do, when to do it, and why.

Vague: "Make sure the chatbot understands our products before answering."

Specific: "If a customer asks about the search feature, first ask which product they're using, Starter, Pro, or Enterprise, before responding. Each product has different search capabilities."

The specific version tells the chatbot exactly when to ask, what to ask, and provides context for why it matters.

Speak directly to the chatbot

Write as if you're giving instructions to the chatbot itself. Avoid referring to it in the third person or describing what it "should" do abstractly.

Indirect: "The AI should avoid recommending phone support since we don't offer it."

Direct: "Never suggest contacting support by phone. We don't offer phone support. Instead, direct customers to email [email protected] or this chat."

Address the chatbot as "you" in your instructions. Write commands, not descriptions.

Give context when it helps

When the chatbot understands why a guidance exists, it applies the instruction more accurately.

Without context: "Always mention the 14-day trial."

With context: "When discussing pricing or plans, always mention that we offer a 14-day free trial with no credit card required. This helps reduce hesitation for new customers."

Context also helps define scope. The second example makes clear this applies to pricing conversations, not every response.

Keep each guidance focused

Each guidance should address one specific behavior. If you find yourself writing "and also" or covering multiple scenarios, split it into separate guidances.

Too broad: "Use friendly language, always ask clarifying questions, and never mention competitor products."

Focused:

  • "Use warm, conversational language. Address customers by name when possible."
  • "If a question is ambiguous, ask one clarifying question before answering."
  • "Never mention or compare us to competitor products, even if the customer brings them up."

Focused guidances are easier to test, debug, and maintain. When something isn't working, you can identify and fix the specific guidance causing the issue.

Offer alternatives when saying "don't"

When you tell the chatbot not to do something, also tell it what to do instead. This gives the chatbot a clear action to take.

Without alternative: "Don't tell customers to contact us by email."

With alternative: "Don't tell customers to contact us by email. Instead, let them know they can continue this chat conversation or ask to speak with the team directly."

A guidance that only says "don't" leaves the chatbot without direction. A guidance that says "don't X, instead do Y" gives it a clear path forward.

Avoid contradictions

Review your active guidances to make sure none of them conflict. Contradictory instructions cause unpredictable behavior.

For example, these two guidances conflict:

  • "If a customer mentions billing issues, offer a 10% discount to resolve quickly."
  • "All billing questions must be escalated to the support team immediately."

One empowers the chatbot to resolve billing issues, while the other requires escalation. Pick one approach and remove the conflicting guidance.

Before adding a new guidance, scan your existing list to check for potential conflicts.

Maintaining your guidances

Start small: Begin with 3-5 guidances that address your most common issues. Add more as you identify gaps in the chatbot's responses.

Review regularly: As your product evolves, some guidances become outdated. Check them monthly and update or remove ones that no longer apply.

Test after changes: After adding or editing a guidance, ask the chatbot questions that should trigger it. Verify the behavior matches your expectations.

Watch for conflicts: When adding new guidances, check that they don't contradict existing ones. Conflicting instructions lead to inconsistent behavior.

Use the right tool: Guidances control behavior. If you need to add new information the chatbot should know, update your help center articles instead, that's where the chatbot gets its knowledge.

Did this answer your question?