E-commerce
June 28, 2026
An e-commerce chatbot can respond quickly, but it becomes risky if its system instructions are vague. Without clear rules, it can promise a discount, invent a delivery estimate, accept an impossible return, or ask for sensitive information.
System instructions must define the bot's role, its sources of truth, its tone, its limitations, its escalation rules, and the topics it must never improvise on.
This guide shows how to write useful system instructions for a reliable, readable, and truly actionable e-commerce chatbot for team use.
Summary
Why are system instructions important?
System instructions are the behavioral framework for the chatbot. They do not replace business data, but they indicate how to use it, when to answer, when to refuse, and when to transfer.
A good instruction protects the brand and the customer. It prevents confident but false answers, especially regarding prices, refunds, guarantees, deadlines, and safety.
A reliable chatbot does not know everything: it especially knows which sources to use and which limits to respect.
Convert over 2,000 customers on average per month with Qstomy.
The world’s 1st Shopify AI dedicated to customer conversion
Empowering 200+ e-commerce merchants
What should a system prompt contain?
It must specify the bot's role, the target audiences, the tone, the languages, the authorized sources, the privacy rules, the limits of promise, the possible actions, and the conditions of transfer.
It must also describe what the bot does not do: legal advice, credit card collection, manual refund validation, policy circumvention, or account modification without authentication.
How to define the sources of truth?
The instructions must indicate which data takes precedence: order, inventory, catalog, return policy, CRM, help base, tax rules, carrier, or back-office. If two sources contradict each other, the bot must recognize the uncertainty and transfer.
This hierarchy prevents using an old FAQ against more recent order information.
How to write limits?
Limits must be concrete. Saying "be careful" is not enough. You must write: "Do not promise a delivery date if the carrier has not confirmed a slot" or "Never collect a verification code".
The more limits are linked to real situations, the better the chatbot can act correctly.
How do you keep a helpful tone?
The tone must be clear, calm, and solution-oriented. The bot can explain a rule without sounding like a set of internal regulations. It must acknowledge the customer's problem before citing a limitation.
Good tone guidelines also specify what to avoid: making the customer feel guilty, being too familiar, giving unverified certainty, or hiding an impossibility.
Which flow to follow?
The drafting must be based on the actual risks of the business.
Define the role of the chatbot, the cases handled, and the excluded cases.
List the sources of truth and their priority in case of contradiction.
Write the limits on payment, personal data, prices, stock, and refunds.
Define the escalation rules along with the information to be transmitted.
Test the instructions on ambiguous customer scenarios before putting them into production.
What examples of guidelines should be used?
For delivery times: "If the carrier has not confirmed the date, present the delivery time as an estimate and offer an alert."
For payment: "Never ask for a card number, security code, or verification code. Direct to a secure link."
For escalation: "Transfer any refund request outside of policy with order, reason, amount, and any proof."
When should the guidelines be reviewed?
Instructions must be updated when a policy changes, a new channel is added, an incident reveals an incorrect response, or a new country imposes a different rule.
They must also be reviewed after analyzing transferred conversations, as real customers often reveal unforeseen cases.
Which KPIs should be monitored?
Follow corrected answers, detected hallucinations, relevant handoffs, poorly formulated refusals, policy errors, security incidents, and satisfaction after automation.
These data show whether instructions make the chatbot more reliable or just more talkative.
Which mistakes should be avoided?
Avoid overly general guidelines, contradictory rules, the absence of priority sources, undocumented exceptions, and limitations that do not state what to do next.
A useful guideline must guide action, not just express an intention.
How can Qstomy help?
Qstomy can connect the chatbot to customer accounts, orders, security rules, shipping options, business guidelines, prices, taxes, and support channels to answer clearly, then hand over sensitive cases with an actionable summary.
The chatbot helps the customer move forward without inventing an emergency, an identity, an environmental claim, a system rule, or a tax calculation that still needs to be confirmed by a reliable source.
Explore AI support, AI sales agent or request a demo.
Key takeaways
Key Takeaways
System instructions must define role, sources, limits, tone, security, and escalation.
What the customer needs to understand
The customer must receive a reliable response, originating from a clear source, with a transfer when the situation is beyond the bot's capabilities.
The chatbot's correct boundary
The chatbot can follow precise rules, but it must not improvise on payments, sensitive data, refunds, prices, or deadlines.
Enzo
June 28, 2026
