As of 13 August 2026, AI can handle your customer enquiries.

What to actually do

1. Use a tool built for this

   The route this page recommends

2. Hand it to a person

   Second choice

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

3. Do it yourself

   The distant third

   A chat interface, power-user skill, and roughly 30 minutes until you can act on the result.

   How to actually do it

   1. Open your current help-centre articles, FAQ, price list, delivery information, returns policy, opening hours and contact details.
   2. Remove expired, contradictory or internal-only material, then paste the approved content into the prompt under the business information heading.
   3. Write your escalation rules, including complaints, refunds, account access, private data, vulnerable customers and questions that need a person.
   4. Paste a representative set of real enquiries with names, addresses, order numbers and other personal details removed, then run the prompt in a chatbot.
   5. Compare every drafted answer and cited source with the current approved pages, correcting any answer that adds an unsupported claim.
   6. Give the approved content and escalation rules to a colleague to test ordinary questions, ambiguous questions, complaints and requests for private information.
   7. Configure the chosen support tool to use the approved content, show the human contact route and pass flagged conversations to your support team.
   8. Publish only after your colleague has confirmed the test answers, then review failed or escalated conversations and update the source material.

   Prompt

   Act as a tier-1 customer-support assistant for [BUSINESS NAME]. Use only the approved business information pasted below. Do not invent prices, stock, delivery times, policies, guarantees, legal rights or account details. If the answer is not explicitly supported by the information, say that you cannot confirm it and offer [HUMAN CONTACT ROUTE]. Escalate complaints, refund or cancellation disputes, safeguarding concerns, threats, suspected fraud, requests involving private account data, vulnerable customers, and anything that could cause significant financial or safety consequences. Never ask a customer for passwords, full payment-card details or unnecessary personal data. Keep replies plain, polite and concise. Give the relevant next step and link to the approved page where one is provided. Before producing the final set of answers, identify any missing or contradictory information that a human must resolve.
   
   Approved business information:
   [PASTE CURRENT HELP ARTICLES, FAQS, PRICES, POLICIES, OPENING HOURS, CONTACT ROUTES AND ESCALATION RULES HERE]
   
   Now produce:
   1. A short list of the topics you can answer safely.
   2. A short list of topics that must go to a person.
   3. Draft replies to these sample customer enquiries:
   [PASTE SAMPLE ENQUIRIES HERE]
   4. For each reply, cite the exact approved source passage or state "no approved source".
   5. A test checklist for a colleague to use before publishing the bot.

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