# AI customer support chatbot built from your docs URL: https://instantknowledgebot.com/solutions/customer-support-chatbot Last updated: 2026-10-17 > An AI customer support chatbot answers customers' repetitive questions instantly, any time, from your own help docs and policies, so your team handles fewer routine tickets. InstantKnowledgeBot builds one from uploaded documents in about a minute, tells customers when an answer isn't in your content, and shows you which questions your docs are missing. An **AI customer support chatbot** answers your customers' routine questions the moment they ask, using the help articles, policies and FAQs you already have. With InstantKnowledgeBot you upload those documents and get a support bot in about a minute. It answers only from your content, says when it doesn't know, and points customers to your team for everything else. ## What is an AI customer support chatbot? It's a chat assistant on your website that uses a large language model to understand a customer's question and write an answer, but **grounded in your own documents** rather than the model's general knowledge. That's the difference from older rule-based bots (which only matched keywords to scripted replies) and from a raw AI model (which might confidently invent your refund policy). For the full explanation, see [what is a knowledge base chatbot](/learn/what-is-a-knowledge-base-chatbot). ## Which support questions should a chatbot handle? The best candidates are questions that are: - **Frequent**: shipping times, opening hours, how to reset a password. - **Documented**: the answer exists in your help center or policies. - **Not account-specific**: the answer is the same for every customer. Questions that need someone to look up an order, approve a refund or calm an upset customer belong with your team. A useful rule of thumb: if a new hire could answer it after reading your help docs, the bot can too. ## How does InstantKnowledgeBot keep support answers accurate? Accuracy is what makes a support bot help rather than hurt. Four things work together: 1. **Grounding:** every bot is instructed to answer only from your knowledge base and to say when the answer isn't there. 2. **A fallback message:** set what the bot says when it can't answer, for example "I don't have that in my notes. Email support@yourcompany.com and we'll help." 3. **Extra instructions:** add rules such as "Keep answers under three sentences" or "Always mention our 30-day guarantee". 4. **Feedback loops:** customers can rate answers, and you can read full conversations. Small knowledge bases are given to the model in full; larger ones are searched and the most relevant passages are sent with each question. Our [RAG explainer](/learn/rag-chatbot-explained) covers how that works. ## What happens when a customer needs a human? InstantKnowledgeBot isn't a live chat or helpdesk tool. Instead: - The **fallback message** tells customers how to reach your team. - **Lead capture** (Pro and Agency, and during the free trial) can show a form when someone wants to talk to a person. It collects their email, plus name, phone and message if you choose, and emails it to you with the conversation. If you need real-time human chat, compare the trade-offs in [AI chatbot vs live chat](/learn/ai-chatbot-vs-live-chat). Many teams use both: the bot for documented questions, people for everything else. ## How do you roll out a support chatbot? 1. **Gather your top 20 to 50 questions** from your inbox, phone notes or support tool. 2. **Check your docs answer them.** Write a simple FAQ document for any gaps. 3. **Create the bot**: [sign up free](/signup), choose a template (or start blank), import your help pages from your website URL and upload your documents ([how website import works](/guides/build-chatbot-from-website-url)). 4. **Test with real questions** in the **Test** tab, including ones it shouldn't answer. 5. **Set the fallback message** and starter questions (up to six). 6. **Install it** on your help or contact page first, then site-wide. Our guide to [adding a chatbot to an HTML website](/guides/add-chatbot-to-html-website) covers the basics. 7. **Review weekly** for the first month: knowledge gaps, thumbs-down answers and conversations. ## How do you measure a support chatbot? The bot's overview shows messages and conversations over time, unanswered questions, top questions and knowledge gaps, plus visitor feedback. Track: - **Coverage:** how often the bot could answer from your docs. - **Gaps closed:** how many knowledge gaps you've filled each month. - **Ticket mix:** whether routine questions to your team go down (measure this in your own support tool). Be careful about headline numbers: whether a chatbot reduces your ticket volume depends on your question mix and how complete your documentation is. Measure in your own data before and after. ## Can one support bot serve customers in many languages? Yes. The bot replies in the language the customer writes in, using your existing documents. See [multilingual website chatbots](/solutions/multilingual-website-chatbot). ## What makes a support chatbot answer well? The biggest factor is the quality of your documents, not the AI model. Answers improve when: - **Each topic has one clear source.** Two versions of your returns policy produce confused answers. - **The key fact comes first.** "Refunds take 5 to 7 business days" beats a paragraph of background. - **Steps are numbered** and use the exact button and menu names customers see. - **Edge cases are written down**: international orders, discounts, exceptions. If an answer is wrong, the fix is almost always to clarify the document. See [preparing documents for an AI chatbot](/learn/preparing-documents-for-ai-chatbot). ## Where should you place the support chatbot on your site? Placement changes how people use it: - **Site-wide chat bubble:** best for pre-sales and general questions. Visitors find it on any page. - **Inline on your help or contact page:** embed the chat with the iframe code so people try it before writing to you. - **In emails and order confirmations:** link to the full-page chat for "Where's my order?"-style questions your docs answer. You can use all three at once; they're the same bot, so answers and analytics stay in one place. ## What does it cost? There's a free plan with one bot and 100 hosted AI messages a month, and unlimited messages if you connect your own AI key. Paid plans add bots, messages, lead capture, team seats and branding options; see the [pricing page](/pricing). For software companies, the [SaaS support chatbot](/solutions/saas-customer-support-chatbot) page has a more specific playbook. ## Frequently asked questions **Can an AI chatbot replace my support team?** No, and it shouldn't try. It handles repetitive questions that are answered in your documentation, so your team can focus on account-specific issues, complaints and complex cases. **Does InstantKnowledgeBot include live chat or ticketing?** No. It's a knowledge base chatbot. When a customer needs a person, it shows your fallback contact message and, with lead capture, collects their details for follow-up by email. **How quickly can I launch a support chatbot?** Creating the bot from your documents takes about a minute, and installing it on a site takes a few minutes. Plan an hour or two to test it with real questions and tidy your documents. **Will customers know they're talking to AI?** They should. The widget presents the bot as an assistant, and you can say so in the welcome message. Being clear about AI builds trust. **How do I find out what the bot can't answer?** The bot's overview lists knowledge gaps: questions your documents couldn't answer. Add that information and the bot can answer next time.