How to Test a Chatbot Before You Launch It

To test a chatbot before launch, ask it the questions your customers really ask, worded the way they would word them, and read every answer against what is true today. Then try to break it: ask for things you do not offer, mix two products up, and see whether it admits what it does not know. Finish by walking through the whole visitor journey on a phone, from the bubble appearing to a lead arriving with your team.

An hour of this catches most of the problems that would otherwise reach customers first. Here is the checklist.

Five areas, in this order:

Collect twenty or thirty questions from where customers already ask them: your inbox, phone notes, WhatsApp, the questions your sales team hears every week. Use their wording, not yours. Customers say "how much for a small one?", not "what is the pricing for the entry-level tier?"

For each answer, check three things:

• Is it right today? Prices, offers, hours and policies are where out-of-date pages show up.
• Is it complete enough to act on? An answer that is true but leaves the visitor still asking is half a fix.
• Is it about the right thing? Chatbots most often confuse things with similar names: two plans, two branches, two products.

Every wrong answer has a cause in the source text: an old page, a page that was not read, or an answer never written down. Fix the source, not the symptom; how to train a chatbot on your website covers each.

Ask it things you know your website does not answer:

• A price for a product you do not sell.
• A service in a city you do not cover.
• A discount or promotion that does not exist.
• A policy you have never written down, such as "can I pay in instalments?"

The right answer to each is some version of "I don't have that information" and an offer to pass the question on. If it gives a confident number or a yes, it is inventing answers. Why chatbots make things up explains what to change.

Real visitors do not type clean questions. Try:

• Typos and short forms: "wats the price", "hrs on sat?"
• Vague openers: "hi", "info please", "is this available"
• Two questions at once: "do you deliver and how much is it"
• Follow-ups that depend on the last answer: "and the bigger one?"
• Other languages, if your visitors use them. Do not promise a language on your site until you have tested it.

The chatbot does not need a perfect answer to each. It needs a sensible one, or a short question back that moves the visitor forward.

Be the visitor, start to finish, and then be your own sales team:

1. Chat until the chatbot asks for your details, and check it asks at the moment you intended. See when a chatbot should ask for contact details.
2. Fill the form with a real phone number and email you can check.
3. Find the lead where your team will find it. Is the conversation attached? Is the number in a format your team can call?
4. Follow it up as your team would, using the follow-up checklist. Did you have everything you needed without asking again?

Then delete or archive the test lead so it does not sit in your real list.

On the live site, not only the preview, and on a phone as well as a laptop:

• It appears. On the pages you meant, and not on the ones you did not. If not, see how to add a chatbot to your website.
• It appears at the right moment. If it waits for a delay or a scroll, wait or scroll.
• It does not cover anything. Cookie banners, sticky "call now" bars and checkout buttons are the usual collisions.
• The greeting reads well on a small screen. One or two short sentences; see greeting message examples.

When every answer about money, policies and availability is right, every question it cannot answer is admitted rather than guessed, and a test lead reaches your team with the conversation attached. Smaller gaps, such as an awkward phrasing or a missing detail, can be fixed after launch: read the first week's conversations, and add what visitors asked that the chatbot could not answer.