Support replies that follow your policies, not the model's guesses

The support messages where the answer is decided by policy: refund and return requests, shipping questions, cancellations, warranty claims. Those are the tickets that eat your team's day, and the ones where an off-the-cuff AI answer is most dangerous, because a generic assistant will happily invent a refund window you never offered.

This pack fixes that by inverting the setup. Your written policies become the only rules the assistant may use. Every draft has to cite the specific rule it relied on, and anything the rules do not cover gets escalated to a human instead of answered. What it deliberately does not do is send anything. A person on your team reads, edits, and sends every reply. That approval step is the feature, not a limitation.

The recipe has four steps. First, go through your last 20 tickets and write down the rule that decided each answer. Expect to find rules that were never written down anywhere, which is worth knowing on its own. Second, paste those rules into the policy-to-prompt template in the download, along with your tone rules and escalation triggers. Third, save the finished prompt somewhere reusable: a Claude Project, a custom GPT, or a Gemini Gem. Fourth, run the loop — paste a customer message, get a draft plus the policy it cites plus a confidence flag, then edit and send from your own helpdesk.

Before trusting it, run 10 tickets you already answered through the prompt and compare against what your best person actually sent. Fix the gaps in the policy list, not the model. Repeat that check monthly, because policies drift.

This setup holds as long as a human is pasting messages one at a time and the answers do not need account data. It breaks when volume climbs, when replies depend on live order history, or when you want drafts waiting inside your helpdesk with logging and QA on every one. At that point you are not maintaining a prompt anymore; you are running a system, and building it around your stack is what we do. If your team is already drowning in tickets, book a free audit and we will map exactly where an agent fits.