DeskPlainAcknowledge the issue and state the next action without inventing a refund or promise. Start with the source facts, copy a ready-to-use prompt and check the result before using it.
A customer reply can be calm and helpful without promising an outcome you cannot authorize. Give the assistant the actual customer issue, approved policy and action you can take. Keep the customer’s claims separate from facts your team has already verified.
Prepare the information that matters
Sanitize names, account numbers and order identifiers unless your organization authorizes the service and needs those details. Include only the relevant approved policy excerpt. If you do not know the delivery date or refund eligibility, leave that question open for the responsible team.
Use a service your organization permits for this material. A pasted excerpt is the assistant's source for this workflow; the prompt does not give it access to files, mail or meetings you have not supplied.
Copy the ready-to-use prompt
The main prompt
Copy this into your AI assistant. Add your notes where marked.
Draft a calm, helpful customer reply from the issue, approved information and authorized next action below. Separate what the customer reports from facts the team has verified. Do not invent a refund, delivery date, compensation, policy, completed investigation or promise. If something is unknown, preserve that uncertainty and offer only the next action I can authorize. Return the customer-facing message and a separate internal review note listing unknowns and commitments to avoid. Keep internal notes out of the customer message. Do not send anything. ISSUE AND APPROVED INFORMATION [Paste the customer issue, relevant approved policy and authorized next action here]
Work through a concrete example
Input to work from
Customer reports a late order and requests a refund. Status unverified. Support may check; refund review comes after verification.
A useful output to compare against
I’m sorry your order has not arrived when you expected. We can check the delivery status and then review the available next steps. Please send your order reference through our approved support channel so the team can locate the order. Please do not send payment-card details. Internal note: delivery status and refund eligibility remain unverified; no outcome or response date has been promised.
This is an editorial worked example, not a captured response from a particular model. Your assistant may return different wording or make mistakes.
Why these choices matter
The reply acknowledges the reported experience without stating that a delivery failure has been verified. It offers the action support may take and keeps the refund decision with the responsible process. The draft must be reviewed against the current policy before it is sent.
Review and improve the first answer
Use this after the first answer, in the same conversation. Keep the original source available.
Audit the reply for commitments not supported by the approved information. Identify language that implies a refund, delivery date or completed check. Rewrite to preserve empathy and a concrete authorized next step.
Ask for a specific correction when the response misses the goal. Changing the requested format can improve readability; it cannot make an unsupported fact true. Return to your source when the answer changes a number, date, responsibility or condition.
Check the result before using it
- Check the reply against the current approved policy.
- Remove promises you are not authorized to make.
- Confirm the requested information and channel are appropriate.
- Keep internal review notes out of the sent message.
Fix the common failure patterns
| What goes wrong | What to change |
|---|---|
| A refund appears as a goodwill gesture | Remove it unless authorized by the actual policy and responsible person. |
| The draft says we investigated | Use that wording only after the investigation happened. |
| The request asks for excess personal data | Ask only for what the approved process requires. |
Move only the customer-facing draft into the reply tool. Check the recipient, remove internal review notes and confirm the next action is still authorized before sending.
Find another task in the AI at Work prompt library, or continue with the related DeskPlain guides below.
Keep these in mind
- Check the reply against the current approved policy.
- Remove promises you are not authorized to make.
- Confirm the requested information and channel are appropriate.
Common questions
Does the prompt run an AI assistant on DeskPlain?
No. This is a library of ready-to-copy prompts. Copy the text, add your source material in your chosen AI assistant and review the answer.
Are the examples real user data or model test results?
No. They are original fictional worked examples showing the intended reasoning and checks. They are not a benchmark or captured model output.
Will the prompt always return a correct result?
No. Review the result against the source and the task-specific checks. Ask about missing information instead of accepting an invented answer.
Sources & further help
Original DeskPlain prompts and fictional worked examples, prepared with AI assistance. Examples illustrate the intended reasoning; they are not captured model responses or a claim of testing across AI products. Read the editorial policy.