Customer service rep
AI Prompts for Customer Service FAQ Answers
Written by PromptsByRole
Start with Step 1See an example
- Who it is for
- Customer service reps, support leads and small business owners who answer the same questions every week and want help-center answers customers can follow without writing in.
- Outcome
- A grouped list of your repeat questions, each matched to the policy text that answers it, short FAQ answers with numbered steps, the other ways customers phrase each question, and a review sheet with an owner and a review date for every answer.
These prompts are plain text, so they work in whichever AI chat tool you use. Paste each step into the same chat so the AI keeps the context from earlier steps.
Step-by-step prompts
How to use it: replace the words in [brackets] with your details, paste Step 1 into your AI tool and answer any questions it asks, then paste the next steps into the same chat. Read and edit the result before you use it.
Step 1
Paste the questions and your source material
I'm writing FAQ answers for our help center. Here are questions customers keep asking, rewritten in my own words with no names, emails, order numbers or account details: [PASTE QUESTIONS, ONE PER LINE] Here is the policy and product information I'm allowed to use, with where each part comes from: [PASTE POLICY TEXT, HELP ARTICLES OR INTERNAL GUIDANCE, E.G. "RETURNS POLICY, UPDATED MARCH"] Use only this source material. Never invent policies, time limits, fees, refunds, exceptions or product features. Where an answer needs something the source doesn't cover, write [POLICY NEEDED: what's missing]. First, group questions that ask the same thing and give each group a plain title. Don't guess which questions are most common; I'll tell you.
Step 2
Match each question to its source
Make a table: Question group | The source line that answers it (quoted exactly) | Gap or conflict. If two source lines disagree, show both and mark [CONFLICT] instead of choosing one. If no source line answers a question, mark [POLICY NEEDED]. Don't write any answers yet.
Step 3
Draft the answers
For each question group that has a source line, write an FAQ answer that: - answers the question in the first sentence - gives any steps as a numbered list, using the exact names of buttons, pages or forms from my source material - uses the source's exact wording for every number, time limit and fee - ends with what to do if it doesn't work, using this contact route: [E.G. "REPLY TO YOUR ORDER EMAIL" OR "USE THE CONTACT FORM"] Keep each answer under [100] words, in plain language, in this voice: [E.G. FRIENDLY AND DIRECT]. Skip groups marked [POLICY NEEDED] or [CONFLICT] and list them at the end.
Step 4
Add the ways customers actually ask
For each answer, write a short question title and 2–3 other ways a customer might phrase the same question, in everyday words (for example "Where's my order?" as well as "How do I track my delivery?"). Base them on the wording in my question list. Don't add questions the answer doesn't cover.
Step 5
Build the review sheet
Make a review sheet with one row per answer: Question title | Source used | Anything time-sensitive (prices, dates, seasonal rules) | Who should approve it | Review date. Leave the last two columns as [OWNER] and [DATE] for me to fill in. Flag any answer that touches refunds, warranties, safety or legal terms so I send it to the policy owner before publishing. Then list every [POLICY NEEDED] and [CONFLICT] item as a question for that owner.
Example of good output
Illustrative example, fictional shop and policy; not real model output:
Source lines: "Delivery address changes are possible until the order status shows Packed." and "To change it, open the order from the confirmation email and select Edit delivery details."
Question title: Can I change my delivery address?
Answer: Yes, until your order status shows "Packed".
- Open your order from the confirmation email.
- Select "Edit delivery details" and enter the new address.
If your order already shows "Packed", reply to your order email and we'll tell you what we can do.
Other ways customers ask: "I moved, can I update my address?" / "Wrong address on my order"
Flagged for the policy owner: [POLICY NEEDED: can an address be changed after the order ships?]
Tips and pitfalls
- Answer first, explain second. Many readers only skim the first line. Put the yes, no or the action in that line, then add the steps.
- One question per answer. If an answer needs "also" or "in addition", split it into two entries. Short answers are easier to find and to keep up to date.
- Keep customer details out of the chat. Rewrite questions in your own words and remove names, emails, order numbers and account details before you paste anything. Use an AI tool your organization allows.
- Test the steps yourself. Ask a colleague who didn't write the answer to follow it on the real site or app. If they get stuck, the answer isn't finished.
- Give every answer an owner and a review date. Policies, prices and button names change. An outdated FAQ answer can create more tickets than it prevents.
FAQ
How is this different from the complaint replies playbook?
The complaint replies playbook handles one customer's problem at a time, from the first reply to the follow-up. This one turns questions many customers ask into public answers that prevent tickets.
Can the AI tell me which questions to cover first?
Not on its own. Use your own ticket tags, saved replies or a count from your help desk, and paste the questions in the order you want them covered.
Can I use the answers as saved replies too?
Yes. Paste an answer into the same chat and ask for a version that greets the customer and ends with your sign-off. Check it against the source again before you save it.