Skip to content
WeZend
AI

Five things to check before you switch a chatbot on

A short, practical checklist: test, replay, shadow mode, opt-out wording and hand-off alerts.

WWeZend TeamOctober 30, 2026 · 4 min read

A chatbot is easy to switch on and harder to switch on well. Before the first real customer message reaches it, work through these five checks. Each one is something you can do inside WeZend, and none takes long.

1. Ask it questions in the test console

The test console runs a dry run: the assistant answers as it would for a customer, but nothing is sent and nothing is recorded as a conversation. Ask the questions your customers really ask, in the words they use, including the awkward ones.

Read two things: the answer, and the sources shown with it. If the answer is right but the source is a page you would not want quoted, fix the source. Keep the questions that matter as saved questions, so you can run the same set again after every change to your knowledge.

2. Replay your real questions

Once you have real conversations, the replay takes recent genuine questions and runs them through today's knowledge. For each one it tells you whether the assistant would now hand over, would now answer, or would behave as before. Nothing is sent.

This is the cheapest way to see whether an edit to your knowledge helped or broke something. Run it after you add a page and after you delete one.

3. Use shadow mode on WhatsApp and e-mail

On WhatsApp and e-mail the assistant begins in shadow mode: every reply is a draft that a person approves, edits or rejects. Leave it there until you have reviewed at least ten drafts, and read them as a customer would. Pay attention to tone and to anything that sounds like a promise: a delivery date, a refund, a price.

Switching a channel to send by itself is something you do deliberately, after the drafts, and the settings show how many were sent as written, edited or rejected. SMS and web chat can go live directly, so test those harder in the console.

4. Check the opt-out wording

Customers sometimes want to stop, and not always with a single clean word. Make sure your own messages explain how to opt out, and know how the assistant behaves:

  • A plain STOP is always handled as an opt-out, before the bot sees it.
  • Phrases like "please stop texting me" are never answered by the bot. It stays silent and a person is told.
  • A contact who has opted out is skipped even if a message arrives.

Read your hand-off and away messages once as a customer who is irritated, and check they do not read as pushy.

5. Decide who gets the hand-off alerts

A hand-off that nobody sees is the worst outcome: the customer was told a person will answer, and nobody does. Choose who is alerted when the assistant hands over or a draft waits, by e-mail and as a badge in the Inbox. Alert recipients must be at a domain your account already uses.

Think about outside hours too. Set your opening hours and an away message, so a customer who writes at night is told when to expect a person. Alerts are rate-limited, so a busy period does not bury your people, but it also means someone should be checking the queue.

A last word

None of these checks tells you how the chatbot will do in general. It has not been measured on real customers, and your knowledge and customers are your own. What the checks do is let you find problems yourself, in the test console and in shadow mode, instead of finding them through a customer.

Where to go next

Walk through the setup guide, read review and hand-off and fill the gaps you find in your knowledge.

One platform. Every customer interaction.

Replace your patchwork of messaging APIs, CDP and automation tools with a single engagement platform built for scale.

No credit card required · EU data residency · 99.99% uptime SLA