StackLift
All playbooks

Playbook

Chatbot Launch Checklist

A checklist to run before a chatbot talks to customers. It covers what the bot should do, what it must refuse, how it hands off to a person and how you will watch it after launch.

Best reader

Support, product and operations leads launching a website, app or messaging chatbot

Outcome

A launch-ready chatbot with tested fallback and handoff, a named owner and a weekly review habit.

Use this sequence

1

Write the scope: what the bot handles, refuses and passes to a person.

2

Prepare and approve the content or knowledge sources it answers from.

3

Design the fallback and the human handoff, including who receives it.

4

Test top questions, odd inputs and every handoff route on the real channel.

5

Check privacy notices, disclosure and data retention.

6

Launch to a small audience, review conversations weekly and adjust.

Decide what the bot is for

A narrow, clear scope beats a bot that tries everything. State what it will not do.

Top questions it should answer

Topics it must refuse

Moments it must hand off

Test the ways it fails

Customers will ask things you did not plan for. Test what the bot does then.

Unknown questions

Angry or emotional messages

Requests outside its authority

Prompt tricks

Make handoff real

A handoff that sends the chat to an unwatched queue is worse than none. Confirm someone owns it and sees the history.

Named queue owner

Context passed along

Hours and out-of-hours behavior

Start in 60 seconds

Turn this checklist into a scoped plan.

Answer a few questions and get an AI-assisted, architect-reviewed scope, cost range, and timeline for your software project.

Frequently asked questions

Should a chatbot say it is a bot?

Generally yes. Clear disclosure builds trust, and some rules require it. Check the requirements for your market and write the wording into the first message.

How do we know the chatbot is working after launch?

Review real conversations, track resolved versus handed-off chats, collect user feedback and keep a list of questions it failed. Update content from that list.

The complete guide

Chatbot Launch Checklist: the complete guide

Updated November 18, 2026 ยท Published by , Founder, StackLift AI. Written with AI assistance; how we make content.

Short answer

A chatbot launch checklist confirms, before customers see the bot, that its scope is written down, its answers come from approved content, it falls back gracefully, it hands off to a person who owns the queue, it has been tested on the real channel, privacy and disclosure are covered, and someone will review conversations weekly after launch.

What is a chatbot launch checklist for?

A chatbot talks to your customers when no one on your team is watching. That makes the first week the riskiest. The checklist turns "it seems to work" into a set of checks with owners, so you know what the bot will do, what it will refuse and where a person takes over.

It applies to website chat, in-app assistants and messaging channels, whether the bot follows scripts or uses a language model. For background, see the AI chatbot development page and the chatbot cost range.

What are the steps to launch a chatbot?

  1. Write the scope. Output: what the bot handles, refuses and passes on. Involve the support lead and product owner.
  2. Approve the content. Output: sources the bot may answer from, each with an owner. Involve support and subject experts.
  3. Design fallback and handoff. Output: replies for unknown questions and a route to a person with context. Involve support and engineering.
  4. Test on the real channel. Output: a test log covering top questions, odd inputs and every handoff. Involve QA and staff who did not build it.
  5. Check privacy and disclosure. Output: notice text, data retention rules and consent where needed. Involve legal and security.
  6. Soft launch and review. Output: a small first audience, weekly conversation reviews and a fix list. Involve the whole team.

Sections below give the checks. A bot that answers from your documents also needs the work in the RAG implementation playbook.

How do you set the bot's scope and tone?

A narrow scope that works beats a wide one that sometimes fails. Pick the handful of requests that make up most of your conversations and do them well. Write down what the bot will not do, such as give legal or medical advice, change an account or promise a refund, and make sure it says so plainly when asked.

Scope sheet
ItemDecideExample entry
HandleQuestions it answers fullyOpening hours, order status, setup help
AssistTasks it starts and a person finishesCollect details for a quote request
RefuseTopics outside its authorityLegal advice, price promises, account changes
Hand offMoments a person must take overComplaints, billing disputes, upset customers
ToneHow it speaksPlain, polite, short; never pretends to be human
DisclosureHow it identifies itselfA first message that says it is an automated assistant

Check the disclosure rules for your market and write them into the opening message. Honest identification builds trust and is required in some places.

What content should the bot answer from?

A bot is only as accurate as its sources. Have each answer traced to an approved page, policy or data source with an owner. Remove outdated and contradictory pages before launch. Structured facts such as order status should come from your systems through an API, not from copied text.

  • List the top questions from tickets, chats and search logs and confirm an approved answer exists for each.
  • Mark one authoritative source per topic and retire the rest.
  • Give each source an owner and a review date.
  • Connect live data such as order or booking status through integrations, not static text. See the guides for Shopify, Zendesk and WhatsApp.
  • Keep a short list of phrases the bot must never say.

How do you design fallback and human handoff?

Every bot will meet a question it cannot answer. Fallback is what it does then: say it does not know, offer a menu or related help, and give a way to reach a person. Silence, a loop or a made-up answer all damage trust.

Handoff is the route to a human. A common failure is sending the chat to a queue nobody watches. Name an owner, set hours, decide what happens out of hours and make sure the person receives the conversation so the customer does not repeat themselves.

Handoff checklist
CheckConfirmOwner
TriggersCustomer asks for a person, repeats a failed question, shows frustration or raises a sensitive topicSupport lead
DestinationA real queue or inbox that someone monitorsSupport lead
ContextFull transcript and any collected details are passed alongEngineering
HoursOut-of-hours reply sets expectations and collects contact detailsSupport lead
Return pathA person can hand the chat back or close itEngineering
MeasureHandoff volume and reasons are loggedProduct owner

The human-in-the-loop glossary entry describes these patterns in general.

What should you test before going live?

Test on the real channel and on real devices. Have people who did not build the bot try to break it. Test the normal questions, then the awkward ones, because customers will ask them.

Test plan
AreaTry thisPass if
Top questionsAsk each in several wordingsCorrect, sourced answers
Unknown questionsAsk something off topicHonest fallback with a next step
HandoffAsk for a person on every channelRight queue receives the full chat
Emotional messagesSend angry or distressed textCalm reply and quick handoff
Authority limitsAsk for refunds, legal advice or exceptionsBot declines and escalates
ManipulationTry to make it ignore its rules or reveal instructionsIt holds the rules
PrivacyShare personal data and ask it to repeat itNo exposure; handled per policy
Devices and channelsMobile, desktop and each messaging appLayout and flow work everywhere

Write down failures and fix or accept each with a named owner. The production readiness checklist and the security review checklist cover deployment and security checks for the surrounding system.

What privacy and disclosure items belong on the list?

A chatbot collects whatever people type. Decide what you store, for how long and who can read it. Tell users what is collected, and check requirements for your location and sector with counsel. This guide cannot say which rules apply to you.

  • A visible notice that the user is talking to an automated assistant.
  • A link to your privacy notice and a short statement on how chats are used.
  • A rule for sensitive data: do not ask for it, and mask or discard it if it appears.
  • A retention period for transcripts and a way to delete on request.
  • Limits on which staff can read conversations.
  • A check of the model provider's terms on storage and training use.

How do you monitor and improve after launch?

Pre-launch testing cannot predict everything people will ask. Start with a small audience, read real conversations every week, and turn what you learn into content fixes and new tests. Treat the bot like a product with an owner and a backlog.

Weekly review
Look atQuestionAction
Unanswered questionsWhat did people ask that it could not answer?Add or fix content
HandoffsWhy did people reach a person?Improve answers or triggers
FeedbackWhere did users rate answers poorly?Review those transcripts
Wrong answersWhich answers were incorrect or off-policy?Fix source, prompt or rule; add a test
UsageWhich channels and times are busiest?Adjust staffing and hours

Plan the ongoing cost and effort too: see the post-launch maintenance plan. To scope a chatbot, use the free estimate or contact us.

Questions people ask

Should a chatbot say it is a bot?

Generally yes. Clear disclosure builds trust, and some rules require it. Check the requirements for your market and put the wording in the first message.

How do we test a chatbot before launch?

Run the top questions in several wordings, then odd, angry and off-topic inputs, every handoff route and each channel. Use testers who did not build it and keep a log of failures and fixes.

When should the bot hand off to a person?

When the user asks, when it fails the same question more than once, when the topic is sensitive or high-stakes, or when the customer is upset. Write the triggers down and test them.

How do we know the chatbot is working?

Review real conversations, track resolved versus handed-off chats, collect user feedback and keep a list of questions it failed. Update content and tests from that list.

Can we launch to everyone at once?

You can, but a soft launch to a small audience or one channel is safer. It lets you read real conversations and fix gaps before the full audience arrives.

How is a chatbot different from an AI agent?

A chatbot mainly answers and guides in conversation. An agent plans steps and takes actions in other systems. Agents need stronger controls; see the AI agent development page for the differences.

Sources

Sources and further reading

The standards, rules, research and official documentation this page draws on.