As of 13 August 2026, AI can draft replies to common customer emails.
Most people should do this themselves; the prompt is on this page.
Can you do it?
5 minutesto a draft.
15 minutesto something you’d act on.
Cost, all in£0
Skill neededchat-fluent
Who has to check ityou
What the alternative costsBotpress is an open platform for building LLM chatbots and agents.
If this goes wrong: the draft gives the customer an incorrect or unhelpful answer, so you correct it before sending or follow up afterwards.
What to actually do
Do it yourself
The route this page recommends
A chat interface, chat-fluent skill, and roughly 15 minutes until you can act on the result.
How to actually do it
- Open the customer email and copy the full message, removing unnecessary personal data such as payment details and passwords.
- Open the current source of truth for the reply, such as your returns policy, delivery guidance, price list, or support notes, and copy only the relevant sections.
- Check your order or customer system for the specific facts needed, then paste those facts into the prompt without adding assumptions.
- Paste the prompt and the three information blocks into an approved chatbot, then ask it to produce the two requested sections.
- Compare every claim in CUSTOMER REPLY with the original email, the customer or order record, and the current business information, and resolve every [NEEDS HUMAN CHECK] marker.
- Send the checked reply through your normal email or helpdesk system, keeping the original customer message and any internal checking notes with the case.
Prompt
Draft a reply to the customer email below. Use only the facts in the customer email and the approved business information provided below. Do not invent order details, dates, refunds, delivery times, policies, promises, or account information. If the information is missing or the request is ambiguous, write [NEEDS HUMAN CHECK] and state exactly what must be confirmed. Write in plain, polite UK English. Answer every question the customer asked. Keep the tone calm and practical. Do not admit fault, offer compensation, make a legal statement, or promise an outcome unless the approved business information explicitly permits it. Do not include internal notes in the customer-facing reply. Return exactly two sections: 1. CUSTOMER REPLY: the email ready to send, including a suitable greeting and sign-off. 2. CHECKS BEFORE SENDING: a short list of factual claims, promises, or decisions that I must verify. CUSTOMER EMAIL: [PASTE THE CUSTOMER EMAIL HERE] APPROVED BUSINESS INFORMATION: [PASTE THE RELEVANT CURRENT POLICY, PRICE, DELIVERY, RETURNS, OR SUPPORT INFORMATION HERE] KNOWN CUSTOMER OR ORDER FACTS: [PASTE ONLY THE FACTS NEEDED FOR THIS REPLY, WITH PERSONAL DATA REMOVED WHERE POSSIBLE] SIGN-OFF: [NAME OR TEAM NAME]
Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.
Use a tool built for this
Second choiceHand it to a person
The distant thirdA person who owns the outcome does this end to end, worth it when the failure is dear.
What it gets wrong
- AI cannot know whether a policy exception is commercially sensible when the customer’s situation falls between your standard categories.
- AI cannot access current order, stock, payment, or delivery information unless you provide it or connect an approved system.
- AI can turn an uncertain internal decision into confident wording, so an employee still has to approve refunds, promises, and admissions.
- AI does not know your relationship with a particular customer or the history behind a complaint unless you supply that context.
- AI can produce a polished reply that answers the words in the email while missing the customer’s underlying concern.
Even on a YES, the friction has a name: judgement under ambiguity, context depth and stakes of error.
How we scored this
Five axes, each scored nought to two by hand: ten means AI carries the task cleanly, and the thresholds that turn a total into YES, PARTLY or NO are published in the methodology. Each axis name links to its definition.
| Axis | Score (0–2) |
|---|---|
| Output | 2 |
| Inputs | 2 |
| Verification | 2 |
| Liability | 1 |
| Effort delta | 2 |
| Total | 9 / 10 |
The methodology and its thresholds are published in full.
FAQ
- Can AI reply to customer emails for me?
- Yes, it can draft replies to routine customer emails using the message, your current policies, and the relevant customer facts. You should check the draft before sending, especially any promise, refund, deadline, or statement about responsibility.
- Can ChatGPT write customer service emails?
- Yes. Give it the customer’s email and approved business information, tell it not to invent facts, and ask it to flag missing information instead of guessing.
- Is it safe to use AI for customer emails?
- It is suitable for drafting routine replies when you remove unnecessary personal data and use an approved tool. Do not let it send automatically until a person has checked the facts, permissions, and tone.
- How do I get AI to write professional email replies?
- Provide the complete customer message, the current policy or reference information, and the exact account facts needed for the answer. Ask for a plain UK English reply plus a separate list of claims to check before sending.
Nearby answers
Assessed by gpt-5.6-luna (gpt-5.6-luna) on 2026-08-13, second-checked by an independent model. Wrong somewhere? Email [email protected] and it gets re-checked.
The newsletter
AI news, new answers and product picks, straight to your inbox.