A chatbot is only worth building if it answers better or faster than a FAQ page. On a low-traffic site without recurring questions, a good contact form and WhatsApp often do just as well.

If it does make sense, its quality comes from its content (RAG) and its guardrails.

Is it actually worth it?

Three conditions:

  • A sufficient volume of repetitive questions.
  • A base of reliable content to answer from (FAQ, docs, product sheets).
  • A team ready to take over on complex cases.

Feeding it: your content, not general knowledge

The chatbot should rely on your own documents through RAG (retrieval-augmented generation), with an obligation to cite sources and say when it doesn't know. It should never improvise on your prices, timelines, or commitments.

Scoping it: range, tone, limits

  • Which topics are covered, and which get handed off to a human.
  • Tone consistent with your brand, short and actionable answers.
  • No contractual promises; no collection of sensitive data.

Escalation: to a human, without friction

A visible "talk to someone" button, transferring the conversation's context, and fast pickup. A chatbot that traps the user in a loop drives them away.

Measuring it and staying compliant

Resolution rate without a human, satisfaction, uncovered topics to add. On the compliance side: clear information, consent when required, and a defined retention period for conversations.

Frequently Asked Questions

Will a chatbot eliminate support jobs?

It absorbs repetitive questions; teams focus on higher-value cases. The volume of simple tickets drops — headcount rarely drops all at once.

Can the chatbot make up answers?

Without RAG and guardrails, yes. With a content base, a citation requirement, and a clear scope, the risk becomes low.

On what type of site is this pointless?

Low-traffic brochure sites with no recurring questions: a form and WhatsApp are enough, and cost less.