Why this matters
An employee emails about parental leave and wants three yeses: full pay, extra time and approved dates. It is a personal moment, and a fast, warm “yes to all of it” is the easiest reply to write and the hardest to take back.
If the policy pays less after the first weeks, if unpaid time needs sign-off, or if the manager has not decided, the employee has already told their clients and made plans. HR then has to correct an email it sent itself.
In this module you give ChatGPT the email thread and the facts from your leave policy, get a draft, and make ChatGPT check every promise in it against those facts. You catch the pay, the unpaid weeks and the approval you cannot give yet, and send a warm, accurate reply with a clear next step.
What you’ll build
A reply to Kofi Mensah, a made-up employee at Fernhill Architecture, who asks whether he gets 12 weeks at full pay, whether he can add unpaid weeks, and whether his December 14 start is approved.
The reply apologizes once, states the real pay schedule (6 weeks at 100 percent, then 6 weeks at 60 percent), says the unpaid weeks need written approval from his manager and HR, says his dates are not approved yet and when his manager will decide, and gives the form deadline. It is checked claim by claim against your policy facts before you send.
What you need
- Nothing to start the practiceChatGPT and Gmail are practice copies. Kofi’s thread is already in the inbox. The policy in this lesson is made up.
- A ChatGPT account, to do it for realA business plan is best for employee matters. Follow your company’s policy on AI tools.
- The email threadCopy the whole thread, oldest message first. Leave out medical details and anything about other employees.
- The relevant part of your written policyPay, eligibility, what needs approval and from whom, deadlines and what HR cannot say. The download below has Kofi’s nine facts and a blank list. Ask counsel when the law is unclear.
Concepts
3 ideas you’ll use in the practice.
The policy decides, not the draft
ChatGPT cannot see your handbook or know who approves what. Write the policy as numbered facts and tell it to promise only what is in the list.
Numbering them lets you point back to a fact when you check the draft.
“It depends” still needs a clear answer
When the honest answer is “not decided yet”, say who decides, by when, and what the employee should do in the meantime. That is kinder than a yes you may have to withdraw.
Ask ChatGPT for a table of every promise in the draft next to the fact that backs it up. Anything without a fact is flagged.
Keep the sensitive parts out
Do not paste medical details, other employees’ leave, or anything your policy says stays out of AI tools. A business plan that does not train on your chats is safer for employee matters.
HR is not giving legal advice in a reply like this. If legal protection is part of the question, say it is confirmed separately and when.
Practice
Do every step yourself in a practice copy of the real tool. It checks your work, and “Do it for me” finishes any step you get stuck on.
Do it for real
Now repeat it in your own account. Tick each step as you go.
Downloads
- Leave facts and requests (.txt)Kofi’s thread and nine made-up policy facts, the requests, and a blank list to copy.
Quick check
3 questions to check you’ve got it.
1. The draft says Kofi gets 12 weeks at full pay. Your policy pays 6 weeks at 100 percent and 6 at 60 percent. What now?
- Leave it, it sounds generous
- Replace it with the exact pay schedule from your facts
- Remove any mention of pay
Show the answer
Replace it with the exact pay schedule from your facts. The check table flags a promise with no fact behind it. State the real schedule plainly.
2. His manager has not approved the December 14 start. What does the reply say?
- That it is approved
- That it is not approved yet, who decides and by when
- Nothing about the dates
Show the answer
That it is not approved yet, who decides and by when. HR cannot approve dates for a manager. Say who decides and when, and suggest waiting before telling clients.
3. Why leave medical details out of what you paste?
- ChatGPT cannot read them
- They are private and not needed to write the reply
- It makes the reply shorter
Show the answer
They are private and not needed to write the reply. The reply only needs the policy and the questions. Keep sensitive details out of AI tools.