E-commerce
July 1, 2026
"The bot said it would still be fine outdoors." "The AI reassured me about use during pregnancy even though the label says no." "Chatbot invents a weight limit that is absent from the instructions." Three failures where an ungrounded warnings bot over-advises or play downs a product risk.
An AI product warnings chatbot does not replace PRODWARN agents (#545). It reads PRODWARN-MAP, quotes warning_text_exact, refuses waiver, and escalates the incident without generative medical advice.
This guide #546 covers intents bot_prodwarn_*, flow PRODWARNbot, and KPI prodwarn_bot. Bot pair of the PRODWARN playbook (#545). New AI use case: provide warning information without over-advising or inventing limits.
Summary
Why automate product warnings by bot?
The questions "can I use it outside" and "pregnancy OK?" arrive at the pre-purchase PDP. A calibrated bot quotes PRODWARN-MAP in seconds, without improvising limits or reassuring beyond the official text.
What the bot resolves tier 1
Warning cite: warning_text_exact map verbatim
Usage limit: usage_limit grounded NO-INVENT-LIMIT
Incompatible use: indoor_only shower child map
Notice link: notice_url warnings section
Waiver refuse: NO-WARNING-WAIVER detection
The GPSR requires safety information to be accessible online (EU Commission, GPSR 2026). The bot intercepts bot_prodwarn_read before any ungrounded response.
PRODWARNbot vs PRODWARN #545, care bot #347, REG #119 and kids #264
Five contents, five sensitive product information layers.
Quick Matrix
#546 PRODWARNbot: bot cites warning without over-advising tier 1
#545 PRODWARN: quality escalation incident agents
Care bot #347: distinct cleaning maintenance bot
#119 REG: regulated health claims
#264 kids bot: child toy safety bot
Pipeline: bot cites warning → waiver refuse → #545 escalate incident. #347 = maintenance. #546 = precautions and limits of use.
Which bot_prodwarn_* intents should be configured?
Eight product warning bot intents.
Eight bot_prodwarn intents
bot_prodwarn_read: read label PDP warnings
bot_prodwarn_usage_limit: max weight duration temperature
bot_prodwarn_contraindication: pregnancy allergy expert advice
bot_prodwarn_incompatible: shower outdoor use prohibited
bot_prodwarn_incident: injury damage reported
bot_prodwarn_waiver_detect: request to ignore warning
bot_prodwarn_compare: difference in warnings between SKUs
bot_prodwarn_escalate: dispute incident → #545
Tier 1 auto: read, usage_limit, incompatible, compare if SKU context + PRODWARN-MAP lookup.
bot_prodwarn_incident intent, medical contraindication and waiver_detect → handoff #545 with transcript payload.
How should PRODWARN-MAP #545 be consumed?
The bot reads PRODWARN-MAP #545: warning_type, warning_text_exact, usage_limit, contraindication_flag, notice_url, escalate_incident.
Lookup grounded
SKU context: product viewed cart line tags
PRODWARN-MAP row: sku_or_category branch
Verbatim cite: warning_text_exact only
Limit cite: usage_limit map no extrapolation
Medical gate: contraindication_flag → NO-MEDICAL-ADVICE
Alignment anti-hallucination (#123): warnings = PRODWARN-MAP whitelist only.
PRODWARNBOT-SUP policy in six rules
Six bot rules for safe product warnings.
CITE-PDP-ONLY: warning_text_exact map verbatim
NO-WARNING-WAIVER: never "it will be fine anyway"
NO-OVER-ADVICE: no advice beyond official text
NO-MEDICAL-ADVICE: contraindication → health pro
NO-INVENT-LIMIT: usage_limit map grounded only
INCIDENT-ESCALATE: injury → #545 immediate
Flow PRODWARNbot PWB-1 to PWB-8
Eight-step flow for product warning bot.
PWB-1 Classify: bot_prodwarn_* intent
PWB-2 SKU context: product cart PDP lookup
PWB-3 PRODWARN-MAP: warning limit flags notice
PWB-4 Guardrail: CITE-PDP NO-OVER-ADVICE
PWB-5 Respond: TPL-PRODWARNbot grounded
PWB-6 Waiver detect: REFUSAL if minimize prompt
PWB-7 Escalate: incident → #545 quality
PWB-8 Log: intent waiver tag prodwarn_bot
Example TPL-PRODWARNbot-READ
"Warnings [SKU]: [warning_text_exact]. Usage limit: [usage_limit]. Notice: [notice_url]. I cannot confirm suitability for your personal case."
TPL-PRODWARNbot templates and touchpoints
Four essential templates.
TPL-PRODWARNbot-READ
Official warnings: [warning_text_exact]. Complete notice: [notice_url].
TPL-PRODWARNbot-LIMIT
Usage limit: [usage_limit]. Do not exceed manufacturer indications.
TPL-PRODWARNbot-CONTRA
Contraindication: [contraindication]. Off-label use is subject to consultation with a healthcare professional before use.
TPL-PRODWARNbot-WAIVER-REFUSE
The warning applies without exception. We cannot confirm non-compliant use. See notice: [notice_url].
Touchpoints
PDP warnings block: bot_prodwarn_read chip click
Pre-checkout FAQ: usage_limit incompatible proactive
Post-purchase email: link notice warnings
Help center safety: bot embed PRODWARN-MAP RAG
Edge cases and reroutes
Five cases outside tier 1 bot cite only.
Health claim efficacy: #119 REG reroute
Maintenance cleaning: care bot #347 distinct
Toy 0-3 years: #264 kids bot
Manual safety section: manual_rag_safety #331
Emergency injury: bot_prodwarn_incident INCIDENT-ESCALATE #545
Bot does not diagnose: NO-MEDICAL-ADVICE, refer to a doctor if personal contraindication.
Essential prodwarn_bot KPIs
Five PRODWARNbot monitoring metrics.
prodwarn_bot_cite_compliance: % of warning_text_exact responses
prodwarn_bot_waiver_attempts: prompts to ignore warning
prodwarn_bot_waiver_refusal_rate: 100% target refusal
prodwarn_bot_over_advice_violations: over-advising off map, target 0
prodwarn_bot_escalate_incident: sessions → #545 quality
Monthly red team: 10 prompts "it will be fine outside", "pregnant is OK, right", "make up weight limit". Waiver or over-advising = PRODWARNBOT-SUP violation.
PRODWARNbot anti-patterns
Six common mistakes.
Reassurance beyond notice: NO-OVER-ADVICE
Minimizing warning: NO-WARNING-WAIVER
Invented limit: NO-INVENT-LIMIT
Medical chat advice: NO-MEDICAL-ADVICE
Incident without escalation: INCIDENT-ESCALATE
Confusing care #347: distinct warning maintenance
PRODWARNbot with Qstomy
Qstomy on Shopify: detect bot_prodwarn intent, PRODWARN-MAP RAG locked, CITE-PDP-ONLY guardrail, NO-OVER-ADVICE blocklist, waiver detect, handoff #545 incident, audit cite compliance.
Pipeline: bot cite refuse → #545 escalate → AI governance #142 audit safety claims.
Explore AI support and request a demo.
Checklist, FAQ and going further
PRODWARNbot Checklist (8 steps)
PRODWARN-MAP Sync #545: RAG bot whitelist only
PRODWARNBOT-SUP Policy: 6 CITE-PDP-ONLY rules
8 bot_prodwarn_* intents: flow PWB-1 to PWB-8
4 TPL-PRODWARNbot-* templates: READ LIMIT CONTRA WAIVER
product.warnings Metafields: bot corpus index
PDP warnings chip: bot_prodwarn_read entry
Red team 10 prompts: medical over-advice waiver
KPI Dashboard: prodwarn_bot_* section 9
FAQ
Difference with #545?
#545 = quality litigation incident agents. #546 = bot tier 1 cite refuse escalate.
Does the bot reassure on prohibited use?
No. NO-WARNING-WAIVER and NO-OVER-ADVICE.
Difference with care bot #347?
#347 = interview. #546 = warnings on usage limits.
Injury reported?
Immediate INCIDENT-ESCALATE handoff #545.
Going further
This week: index PRODWARN-MAP bot corpus, test red team "it will be fine anyway", measure prodwarn_bot_waiver_refusal_rate. Activate PDP warnings chip.
Enzo
July 1, 2026
