In This Article
A good customer reply does three things: answers what you can answer, asks for what is missing and gives the customer an obvious next step. AI can help you draft that reply when you are busy or struggling to find the right words. The trick is to give it a bounded writing job, not ask it to run the conversation on your behalf.
The examples below are invented training scenarios, not real customer messages. Replace their facts with information you have checked. Keep names, contact details, registration plates and private history out of the prompt; you can add the necessary personal greeting yourself afterwards.
Start with a short set of rules
Example prompt: Draft a friendly reply under 90 words using only the facts below. Ask for missing details. Do not invent prices, appointments, product performance, legal assurances or warranty terms. Do not say a booking is confirmed. Return one draft and a separate list of facts I need to check.
That final list is useful because it separates writing from decisions. NIST's Generative AI Profile explains that AI can produce false content confidently. Clear instructions help define the task; they do not remove your responsibility to review the answer.
Example 1: the incomplete price enquiry
Customer scenario: 'How much to tint my car?' Known facts: you need the year, make, model, requested glass and whether old film needs removal. No price has been approved. A weak reply is 'Prices start from $99' when that figure came from the model rather than your price list.
Draft: Happy to help. What year, make and model is the car, and which windows would you like tinted? Please also let me know whether there is old film to remove. With those details I can explain the suitable options and confirm what the job would include.
This reply moves the enquiry forward without forcing the customer through a long questionnaire. If you already have some details, remove those questions. If a reliable price really is available from your approved process, supply it rather than using AI as a reason to dodge a straightforward answer.
Example 2: a request for tomorrow
Customer scenario: 'Can you do it tomorrow morning?' Known facts: the calendar has not been checked. Your reply should acknowledge the request without creating an appointment that nobody can deliver.
Draft: I can check tomorrow morning for you. Please send the vehicle details and which windows you want done so I can allow the right amount of time. I will confirm availability before we arrange the booking.
Only use that wording if you will actually check and follow up. Then put the task in your normal workflow. A polished promise with no action behind it is worse than a brief, honest reply.
Example 3: a concern after installation
Customer scenario: 'There is a mark on the window—is that normal?' The mark has not been inspected. Avoid asking AI to diagnose it from a sentence, promise it will disappear or dismiss the customer's concern.
Draft: Thanks for letting me know. Which window is affected, and could you send a clear photo of the area? I would like to check it against the installation record and arrange a closer look if needed. Please follow the care instructions supplied with the job in the meantime.
Make the final reply sound like your business
Read the draft aloud. Delete unnecessary sales language and check every factual statement. Has it added 'lifetime warranty', 'completely legal' or a performance percentage you never supplied? Has it promised a callback time you cannot meet? Keep the useful answer and remove unsupported extras. The FTC's advertising guidance is a reminder that commercial claims need support, including when a tool helped write them.
Save the strongest approved replies as templates. Repeated questions may then need a simple template rather than another AI session. For the next stage, follow our notes-to-quote workflow; for data handling, read the privacy guide. The goal is a helpful conversation that you can confidently stand behind.
