As of 13 August 2026, AI can build a customer service chatbot from your FAQs.

What to actually do

1. Hand it to a person

   The route this page recommends

   A person who owns the outcome does this end to end, worth it when the failure is dear.

2. Use a tool built for this

   Second choice

3. Do it yourself

   The distant third

   A chat interface, chat-fluent skill, and roughly 1 hour until you can act on the result.

   How to actually do it

   1. Open your current FAQ, help-centre articles, refund and delivery policies, contact details and opening-hours record, then remove superseded versions.
   2. Paste the cleaned material into the prompt, including the business name, customer type, escalation routes and any rules about personal data.
   3. Run the prompt in a chatbot builder such as CustomGPT, then upload or paste the same source documents into its knowledge base.
   4. Configure the bot to answer only from the supplied content, show or retain source references where the platform supports them, and offer a human contact route when the answer is missing.
   5. Run the generated test set and add realistic questions from recent support conversations, including complaints, unclear questions and requests outside the FAQs.
   6. Compare every answer against the current FAQ and policies, correct the source documents rather than merely rewriting unsupported answers, and have a colleague check the escalation cases.
   7. Publish the bot in a limited location first, test the contact and escalation links on the live page, and monitor unanswered or incorrect conversations before expanding its use.

   Prompt

   Build a customer service chatbot specification from the material below.
   
   Business: [BUSINESS NAME]
   Customers: [WHO THE CUSTOMERS ARE]
   FAQs and help documents:
   [PASTE CURRENT FAQS AND HELP DOCUMENTS]
   Current contact details, opening hours and escalation routes:
   [PASTE DETAILS]
   Products, services, prices and policies that must be treated as current:
   [PASTE DETAILS]
   
   Create:
   1. A source-grounded set of chatbot answers for the FAQs.
   2. A clear system instruction for the chatbot.
   3. A list of topics it must refuse to answer or send to a person.
   4. Suggested escalation messages, including the correct contact route and opening hours.
   5. Twenty test questions covering normal questions, unclear wording, complaints, requests for refunds, questions outside the source material and attempts to obtain personal information.
   
   Rules:
   - Use only the supplied material. Do not invent prices, policies, guarantees, availability, delivery times or contact details.
   - If the source does not answer a question, say that the chatbot cannot confirm it and direct the customer to a person.
   - Do not collect passwords, payment card details or unnecessary personal information.
   - Do not make legal, financial or safety claims beyond the supplied material.
   - Treat complaints, threats of legal action, safeguarding concerns, account access problems and requests involving personal data as escalation cases unless the supplied policy explicitly gives a safe automated route.
   - Keep answers concise, polite and plain. Include the relevant source heading after each draft answer so a colleague can check it.
   - Mark any contradictory, outdated or incomplete source material for a human to resolve before publication.
   
   Return the result in sections headed Chatbot instruction, Answer library, Escalation rules, Unanswered topics and Test set.

   Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.