E-commerce
June 28, 2026
A chatbot that invents a return policy, a delivery time, stock availability, a discount, or a warranty can create a real customer problem. The answer seems helpful in the moment, but it becomes a promise that the company will have to correct later.
Avoiding hallucinations is not just about "better training" the bot. It requires giving it reliable sources, clear boundaries, a way to say it doesn't know, and a clean handoff to a human.
This guide explains how to reduce incorrect answers from an AI chatbot in e-commerce, with rules that are easy to understand and helpful for the customer experience.
Summary
Why is an hallucination expensive?
A wrong answer can trigger a purchase, a return, a wait, or a complaint. If the bot promises a 24-hour delivery when it is not possible, the customer does not see a technical error: they see a broken promise.
Hallucinations therefore affect trust, support, and sometimes commercial compliance. The chatbot must be designed to prefer a cautious answer over an attractive but false one.
In e-commerce, a hallucination is not just an incorrect sentence. It is often a bad customer decision.
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
Which answers are the riskiest?
Sensitive areas include pricing, promotions, stock levels, delivery times, warranties, returns, refunds, invoices, customized products, account security, and personal data.
The bot can respond more freely on general advice, but it must be strict as soon as a response becomes a commercial term or personalized information.
How to use reliable sources?
The chatbot must rely on up-to-date content: return policy, validated FAQ, product data, order status, stock, and commercial rules. An old page or a draft must not serve as the truth.
Each important source must have an owner, an update date, and a removal logic when it is no longer valid.
How do I teach the bot to say “I don't know”?
A good chatbot does not have to fill in everything. If information is missing, it should say so simply: "I cannot verify this information in real time at the moment."
This transparency is often better accepted than a false answer. It can be accompanied by a useful option: general information, official link, or transfer to support.
How can promises be regulated?
The bot must avoid overly firm words when the information depends on an external system or human validation. “You will receive tomorrow” is riskier than “express delivery may be available if the checkout confirms it.”
Promises must be reserved for verified information: confirmed order, validated policy, known balance, or option available in the account.
Which flow to follow?
The flow must verify before responding.
Identify if the question concerns a business rule or personalized data.
Search for the reliable source or real-time status.
Answer clearly if the information is verified.
State what is missing if the information is not available.
Transfer cases where an error could create a promise or a dispute.
Which messages should be used?
For missing information: “I can't verify this point in real time, but I can explain the general rule or forward your request.”
For a delivery: “This option must be confirmed at checkout, as it depends on your address and the time of order.”
For a return: “I can tell you the general policy, then verify your order if you give me the reference.”
How to test hallucinations?
It is necessary to test ambiguous questions, past promotions, requests for compensation, low-stock cases, ineligible products, and phrases that push the bot to promise an exception.
Testing must be regular, as policies, catalogs, and campaigns change. A bot that is correct today can become inaccurate if its sources are not kept up to date.
Which KPIs should be monitored?
Track agent-corrected responses, unverified promises, handoffs due to uncertainty, policy mistakes, outdated promotions quoted, and chatbot-related complaints.
These metrics show where the bot needs to be better connected, better constrained, or better trained.
Which mistakes should be avoided?
Avoid letting the bot invent a rule, connecting it to outdated sources, hiding uncertainty, or treating a business exception as an automatic response.
A reliable chatbot is not one that answers everything. It is one that knows when it can answer and when it must stop.
How can Qstomy help?
Qstomy can connect the chatbot to support rules, customer context, and useful data to respond clearly, then transfer sensitive cases with an actionable summary.
The chatbot helps the customer move forward without exposing unnecessary data or making a decision that must remain human.
Explore AI support, the AI sales agent, or request a demo.
Key takeaways
Key Takeaways
Hallucinations must be reduced with reliable sources, clear boundaries, and clean escalation.
What the Customer Needs to Understand
The customer must receive a verified response or an honest explanation of what is missing.
The Chatbot's Proper Limit
The chatbot can respond quickly, but it must refuse to make up prices, lead times, inventory, policies, or exceptions.
Enzo
June 28, 2026
