Saved replies go wrong when they say more than the policy does. One agent writes "we always refund within 30 days", another writes "within 14", and a customer screenshots the more generous one. The fix is to write each reply from a line in the policy, and to notice where the policy says nothing.
That second part is the useful surprise. Reading a policy closely to write macros shows you its holes: no mention of who decides, an exception that contradicts a rule, a sentence that cuts off. Better to find those before a customer does.
What you need
- Your refund, cancellation or billing policy, as text.
- Three to six real customer messages that show the cases you handle most.
- The name of your help desk, so the variable style fits. Most tools use something like a customer name placeholder.
The prompt
Paste this into a new chat, then paste your policy and the messages.
You write support macros from the policy and customer messages below. Work only from them. Never invent a rule, a number of days or an amount.
1. Policy facts sheet: each rule with a short quote from the policy, or NOT STATED. List contradictions, gaps and risky wording in the policy itself.
2. For each type of message, write a saved reply that relies only on a quoted policy line. Name the line it relies on. Use [DECIDE: ...] where the policy is silent, with a suggested default for me to accept or change.
3. Write a short checklist for the agent for each macro (what to check before sending).
4. Write an escalation note for cases a human must decide: what to include, who decides, and what the agent must not promise.
Help desk: [name]
POLICY:
[paste here]
MESSAGES:
[paste here]
How to check the result
Read the facts sheet before the macros. Check each quote against your policy page. If a quote is not there, remove the rule.
Then read the gaps list. This is often the most useful part. Decide each open point, update your policy, and only then use the macros that depend on it.
Check each macro for promises. The reply should say what the policy allows and no more. Cut any sentence that offers something you have not approved.
Test one macro by pasting a real ticket and reading the reply as the customer. If it sounds cold, adjust the wording but not the facts.
Load a few into your help desk and check that the placeholders are filled correctly before you add the rest in bulk.
What good output looks like
This is an invented illustration of the shape. A facts sheet row reads: "Refunds: 'Fees paid are non-refundable.'" Another reads: "Who decides exceptions: NOT STATED." A macro reads: "Thanks for getting in touch. Our policy says fees are non-refundable, so I can't refund this charge. If you think there has been a mistake, tell me and I will pass it to the team." The macro names its line.
The escalation note might say: "Escalate when the customer mentions a chargeback or says they were not told about the policy. Include the order date and the customer's own words. Do not promise a refund." That is the part agents use when they are unsure.
If a macro offers something your policy does not, it has drifted. Remove it, or turn the point into a decision for you to make and write into the policy.
Limits
- It works only from the policy and tickets you give it. It cannot see your help desk or your customer records.
- Where the policy is silent, you get a [DECIDE] and a suggested default, not a rule. You choose.
- It gives no legal advice. If you sell across borders, check local consumer-law minimums.
- Placeholder names depend on your tool. Check one before importing many.
Follow for new free skills
New free skills and guides appear in all of these as we publish them. Pick whichever you already use.
- RSS feed for any feed reader
- Watch the free skills repo on GitHub
- Follow on Gumroad