How the chatbot answers visitors
Staff voice, knowledge, contact buttons, links, and what it will not invent.
The widget is staff for that website. It speaks as “I” / “we”, not as a third-party narrator. Tone, greeting, name, and a short persona (up to 240 characters) are set on Widget — full steps in Customize how your chatbot looks and talks.
Answers are meant to come from pages we read, contact facts, and answers you added. It is instructed not to invent prices, policies, or features that are not in that material — still review chats, because AI can get a detail wrong.
If a visitor asks for a page that exists in your knowledge (blog, pricing, docs, a named article), the bot pastes the real web address. It will not guess a page such as a contact page if that page was never read.
Phone, email, and WhatsApp are not pasted as text. When those details were found on the site, the widget shows Call / Email / WhatsApp buttons instead. Ask “how do I contact you?” to see them.
Comparison questions (“is there something better?”) are answered confidently about this business. The bot should not recommend other vendors or shrug about “other solutions”.
Chats keep a short memory of the current visitor thread (same browser session) so follow-ups stay on topic. Visitors can thumbs-up or thumbs-down a reply; you see those votes on Conversations, and admins see platform totals on Chat quality.
Unsafe requests are refused in one line, then the bot offers website help again.
If a listed fact is genuinely missing, it says it is not sure and offers the contact buttons — it should not invent an answer to fill the gap.