Build your agent

Curated answers & CTAs

Curated answers and CTAs give you deterministic widget behavior — responses and cards the LLM never sees, never paraphrases, never goes off-script on. Use them where you can't tolerate the model rewording you.

Curated answers

Best use cases:

• Pricing — exact numbers, exact wording.
• Legal and compliance — returns policy, GDPR statements, contract language.
• Routing — “how do I cancel?” → the cancellation flow.
• High-stakes brand FAQs that must read identically every time.

Authoring features:

• Question patterns are comma-separated keywords, treated as OR-triggers, case-insensitive.
• Answers support Markdown formatting.
• A priority field resolves overlap between multiple matches.
• Optional “publish to KB” turns the answer into a public help-center article.
• Language-specific variants supported.
• Toggle on/off without deleting — useful for seasonal promotions.

The matching engine runs before the RAG pipeline, so curated answers respond in under 100 ms end-to-end and consume no inference tokens.

CTAs

A CTA is a clickable card rendered inline in the conversation. Each card has:

• A headline and optional description.
• One or two buttons, each performing one of: open URL, send a message back to the agent, trigger lead capture, dismiss.
• Conditional display rules (kind: cta in Behavior rules).

Common patterns:

• “Compare plans” / “Talk to sales” after a pricing question.
• “Book a demo” once the conversation passes a discovery threshold.
• An exit-intent rescue form on the checkout page.
• Scroll-depth engagement nudges on long marketing pages.

Testing

Validate curated triggers in the Playground: confirm the rule fires on intended questions and doesn't fire on near-misses. Specificity matters — overly broad keyword lists cause false positives that feel robotic to visitors.