Skip to main content
EN
Sign in
Guide

A website chatbot that answers at night

Datenmassiv AI puts a chatbot on your own website that answers customer questions from exactly the documents your team already maintains internally.

The same questions, over and over

Anyone who deals with customers knows the list by heart. When are you open, from what quantity is delivery free, which form do I need, who do I have to talk to. The questions come by phone, by email, through the contact form, in the evening and on weekends, and as a rule the answer is already written down somewhere.

You already stored exactly that writing in An AI knowledge base that answers. The website chatbot (labeled Chatbots in the sidebar) is where that work pays off a second time: it draws on the same knowledge base that already serves your staff internally. You maintain nothing twice and do not copy out an FAQ collection. When a new price sheet goes into the knowledge base, the chatbot uses it from the very next question.

The chatbot feature is switched on from the start. The Chatbots menu item still only appears if your role includes the Manage chatbots permission, and the standard role "Member" does not. If it is missing, ask someone in your company with administrator rights.

Why it does not make things up

The chatbot answers exclusively from the knowledge bases you assign to it. It has no access to any other data in your organization, none to the agents and none to your workflows. If it finds no basis in the released documents, it says so openly instead of improvising. Below every answer it states which document the answer comes from. These source references are switched on from the start. You can turn them off on the General tab under Show sources, but we advise against it.

That is the difference from a generic chatbot that sounds plausible while inventing delivery times you never offered.

Put it on your site

  1. Open Chatbots in the sidebar and click New chatbot. Give it a name and turn off Activate immediately in the dialog while you are only setting the chatbot up. Confirm with Create.
  2. Switch to the Knowledge tab and click Connect for every knowledge base the chatbot may use. Whatever you do not connect here, the chatbot does not see.
  3. On the General tab, under Template, choose the behavior you want. Nine templates are available: Information Assistant, B2C Product Advisor, B2B Advisory Assistant, Compliance Assistant, FAQ Specialist, Compact, Internal Helpdesk, Municipal Assistant and Custom for writing your own. Your company name is filled in automatically.
  4. On the same tab, set the Widget UI language of the chat window and the Bot response language. Both offer automatic, German, English, French and Spanish, and the response language also offers Browser language.
  5. On the Design tab, adjust the primary color and the position of the chat button. A live preview runs alongside on the right.
  6. Lead capture and calendar are on the Integrations tab.
  7. Click Save.
  8. On the Embedding tab, copy the code and paste it before the closing </body> tag of your website. Add your domains to the allowlist there, so the chatbot only runs on your pages.
  9. For the functional test, first activate the chatbot only on a non-public test page and allow only that page's domain.

You can run several chatbots: a matter-of-fact one for support, an advisory one for the product page, each with its own knowledge bases and its own behavior.

Test on a test page first, then publish

The Design tab still shows the visual preview for colors, position and custom CSS. The answer test on the Test tab is temporarily unavailable and therefore has no input field. The public embedded chatbot is not affected by this pause.

For the functional check, first activate the chatbot on a non-public test page, add only that page's domain to the allowlist and test the saved configuration there. Only after it passes do you add the public domain.

Take ten real customer questions. If the chatbot replies to one of them that it cannot find anything on it, your knowledge base is missing the matching document; the chatbot is not missing the ability.

And the uncomfortable side: whatever appears on your website is your responsibility, even if a chatbot wrote it. Datenmassiv AI narrows that down. The bot answers only from the documents you assign, names the source and says so when it finds nothing. Whether those documents are correct and current, it does not check. Where a mistake gets expensive (allergens, safety notices, commitments on prices and deadlines), exactly those questions belong on the non-public test page before publishing, and in regular checks on the Conversations tab afterwards.

When the answer is not in the documents

For exactly this case, turn on the Capture contact details option on the Integrations tab. If the chatbot gets stuck, it offers the visitor the chance to leave their contact details. You receive the inquiry by email right away and find all of them collected on the Contacts tab. That way, an inquiry does not get lost just because it happens to be Friday evening.

If you also connect a Google Calendar on the same tab, the chatbot can look into it: it reads appointments, finds free time slots and suggests suitable times to the visitor. Nothing is booked in the process. The chatbot has read-only access to the calendar, and scheduling stays with you. This one connection leads outside the EU framework: the availability query goes to Google. If you do not want that, leave it out; the chatbot works fully without it.

What happens to the conversations

Your website visitors' conversations are stored on our servers. That belongs here and not in the fine print: for mid-sized businesses, it is the second question after where the data is located.

You set the Retention period per chatbot yourself, between 7 and 365 days, with 90 as the default. After that, a daily cleanup job deletes them. As long as they exist, you see them on the Conversations tab, and on the Analytics tab you see how often the chatbot is used and which questions go unanswered. That is the real value of storing them: you fill gaps in your knowledge base exactly where customers ask. Also add the link to your privacy policy. Before a visitor writes for the first time, they see a short notice screen with the AI disclosure, the retention period and exactly this link; the chat opens only after they confirm.

You see the chatbot's access key in plain text only once. So keep it in a safe place. The domain allowlist accepts individual addresses and a wildcard for subdomains: *.example.com covers shop.example.com and example.com. It does not work for paths: example.com/* is rejected, and an allowed domain applies to all of its pages anyway. To prevent abuse, limits of 60 requests per minute per access key and 10 per minute per visitor apply. In standard mode, your AI processing runs exclusively in EU regions, and the database, storage and email are located in the EU.

The same toolkit, facing inward

The Internal Helpdesk template shows that the chatbot is not only useful facing outward. Embedded in the intranet or the employee portal, it answers the questions that otherwise make the rounds through IT, HR and the front office: how do I request vacation, which rule applies to business trips, where do I find the template. Technically it is the same route (connect, test, embed), only this time the knowledge bases point to your internal manuals.

From your own page, the chat window can be opened and closed with JavaScript, or started with a ready-made question through a parameter in the address bar.

Coming full circle

Store company knowledge once, so the AI answers from it with a source. Question a single document without any setup. Let a recording become minutes. Hand recurring paperwork to workflows that wait for your approval at the right points. And now the same knowledge also answers your customers' questions, without anyone sitting at a desk.

Seven tools, one shared body of company knowledge. Knowledge bases, AI chat, document chat and the chatbot are switched on from the start; Datenmassiv AI switches on transcription, agents and workflows for you. The role permissions without which the menu items do not appear are then assigned by someone in your company with administrator rights.

The next step is a small one: create a knowledge base with the documents you send every customer anyway, and put a chatbot in front of it. On the Analytics tab you see whether your customers use it, and on the Conversations tab what you have not explained well enough to them so far.

Read more

The handbook covers the details this guide deliberately leaves out: