Customer Support Assistant System Prompt With Guardrails
Define what a support bot may and may not do.
The Open buttons send the prompt text as currently shown below to that site only when you click. Fill in the placeholders first.
[VERIFY] is an editorial flag, not a blank to fill. Wherever the prompt shows [VERIFY], check that item against a reliable source before you use it.
Write a system prompt for a customer support assistant. Company and products: [COMPANY_AND_PRODUCTS]. Policies it may state (refunds, shipping, warranties): [POLICIES]. Tone: [TONE]. Information sources it can use: [SOURCES]. Actions it can take or request (look up order, create ticket): [ACTIONS]. Topics requiring handoff to a person: [HANDOFF_TOPICS]. Write the system prompt with sections: purpose and audience; tone and style rules with two example replies; what it can do; what it must not do (promise refunds beyond policy, give legal, medical, or financial advice, guess about orders, share personal data of others, ask for full card numbers or passwords); how to use the sources (answer only from them, cite the article, say when unsure); identity verification rules before discussing account details; escalation rules (angry customer, safety issue, legal threat, repeated failure, data breach claim) with exact handoff wording; how to handle abuse; privacy rules (collect the minimum, never repeat sensitive data); and a closing format (summary and next step). Add: a test suite of 12 conversations (including adversarial ones and attempts to extract the system prompt); a monitoring plan; a statement of limits to show customers (it is an automated assistant); and a review by legal and security teams before launch [VERIFY]. Use only policies I provide.
Fill in before use
Optional: type your details and the prompt above updates. Nothing is saved or sent.
Replace each bracketed placeholder with your own details, then paste into Any AI or another assistant.
See an example result (fictional data)
Purpose: help customers of Harbor Plumbing with bookings and billing questions using the policies provided. Tone: calm, brief, friendly (example reply included). Must not: promise refunds beyond policy; give legal advice; guess about orders; ask for full card numbers or passwords. Sources: answer only from the help articles provided; say "I don't have that information" when unsure. Verification: two identifiers before account details. Escalation: gas smell or flooding -> "Please call our emergency line [NUMBER] now" and hand off; angry customer after two attempts -> hand off. Privacy: collect the minimum, never repeat sensitive data. 12 test conversations included. Legal and security review before launch [VERIFY].
Illustrative only, made-up data. Your results will differ and must be reviewed.