E-commerce
July 1, 2026
A customer sometimes requests sensitive information: delivery address, refund status, order modification, account history, or masked payment details. Before responding, the chatbot must verify that it is speaking to the right person.
The challenge is to secure the exchange without making the experience cumbersome. The bot should request the bare minimum necessary, mask sensitive data, and transfer cases where identity cannot be confirmed.
This guide explains how to verify the correct interlocutor with an AI chatbot, without exposing customer data.
Summary
Why verify the interlocutor?
An order may contain an address, a telephone number, an invoice, a refund status, or a sensitive request. If this information is given to the wrong person, trust is broken.
The chatbot must therefore adapt its level of response to the level of verification. It can provide general information to everyone, but it must verify before displaying or modifying personal data.
Proper verification protects the customer without turning every conversation into an identity check.
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 requests are sensitive?
Sensitive requests concern address changes, refunds, invoices, account data, order history, contact information, and any action that can change an order.
A general question about a return policy does not require the same verification as a request to change the address of an already paid order.
How to check without asking too much?
The bot must prioritize methods that are already secure: logged-in account, email link, customer portal, or a code sent to the channel associated with the order. It must avoid asking for sensitive data in plain text in the chat.
If confirmation is necessary, it can ask for partial, non-critical information, or direct the user to a secure area.
How to hide data?
The chatbot can display a masked version: city, last characters of an email, or partial reference. It must not repeat a full address or a full phone number if it is not necessary.
This approach allows the customer to recognize the context without exposing all the information in the conversation.
It is particularly useful for gift orders, shared devices, or conversations opened from an email link. The customer gets a reference point, but the data remains protected.
What should I do if my identity cannot be verified?
The bot must calmly explain the limit: "To protect your data, I cannot display this information here without verification." It can then suggest a secure path or transfer to an agent.
The refusal must remain helpful. The customer must understand how to continue, not simply receive a block.
Which flow to follow?
The flow must adapt the verification to the risk level.
Identify the request: general information, personal data, or action on an order.
Assess the necessary level of verification.
Use the logged-in account or a secure channel if possible.
Mask the data displayed in the conversation.
Transfer if identity remains uncertain or if the action is sensitive.
Which messages should be used?
For a verification: "To protect your information, I need to verify that you are indeed linked to this order before displaying the details."
For a masked datum: "I see an order associated with the address ending in [city/zip code]. Do you want to continue with this one?"
For a limit: "I can give you the general procedure, but I cannot modify this order without verification."
When to transfer?
The transfer is necessary if the customer cannot confirm their account, if a gift order is involved, if an address needs to be modified urgently, or if the request concerns a refund or personal data.
The bot must transmit the request, the level of verification achieved, the order concerned, and the minimum necessary information.
Which KPIs should be monitored?
Track successful verifications, identity blocks, security transfers, sensitive modifications, and conversations where the customer abandons during verification.
These indicators show whether the verification is proportionate or too burdensome for customers.
Which mistakes should be avoided?
Avoid asking for a password, displaying a full address too early, modifying an order without proof, or refusing without offering an alternative path.
The chatbot must protect data while remaining solution-oriented.
How can Qstomy help?
Qstomy can use the customer, cart, and order context to respond clearly, and then hand over sensitive cases with an actionable summary.
The chatbot helps the customer move forward without exposing unnecessary data or promising an unverified action.
Explore AI support, the AI sales agent or request a demo.
CIV-BOT Checklist (12 steps)
Validate AUTH-MAP #122 action auth levels
Sync TPO-MAP #461 third disclose consent
Draft policy CIV-BOT-SUP 8 rules
Deploy 12 intents bot_civ_* classifier
Implement IV-1 to IV-8 + guardrails MIN-DISCLOSE NO-CONFIRM-FAIL
API sanitize pipeline tiered disclose
OTP L3 sensitive actions + throttle fraud
TPL-CIV-* templates + TPO-CONSENT integration
Logged-in customer bypass Shopify widget
Test 8 scenarios: buyer L1, L2 step-up, L3 OTP, third consent, recipient gift, fraud block, logged-in bypass, handoff payload
Dashboard KPI civ_bot + red team quarterly
Route subdomain bots post-auth GORD GFTRTN returns
Summary
#462 = bot identity gate, not auth execute alone
Role before PII: IV-2 mandatory
Minimal disclose: default until auth pass
Third consent: not deny only
KPI civ_bot_over_disclose: target 0
FAQ
Can the bot issue refunds after auth?
CIV-BOT qualifies L3. Refund execution by agent or returns bot post-handoff policy.
Difference with #122?
#122 human auth policy. #462 bot automation gate step-up.
Difference with #461?
#461 TPO agents execute. #462 bot classify consent trigger.
Third party blocked flat out?
Forbidden. TPL-CIV-THIRD consent path mandatory.
Logged-in customer?
Logged-in bypass IV-2 skips redundant implicit L2 auth.
Go further
This week: sync AUTH-MAP TPO-MAP into bot RAG, deploy API sanitize pipeline, activate IV-2 role classifier, test MIN-DISCLOSE guardrail with 20 scenarios, configure OTP L3 refund, launch civ_bot_over_disclose zero-tolerance dashboard.
Share this #462 guide with bot ops and security: one pre-auth API sanitize is worth ten address leaks by LLM context leak, one third-party consent path is worth twenty legitimate assistants blocked without a structured alternative.
Go-live bot test matrix
8 scenarios sign-off: buyer L1 tracking, L2 items, L3 refund OTP, third consent, recipient gift route, fraud throttle, logged-in bypass, over-disclose block test.
Post-deploy monitoring
First 14 days: daily civ_bot_over_disclose audit, auth funnel review, third consent approve rate, handoff reason taxonomy update.
Enzo
July 1, 2026
