E-commerce
July 1, 2026
"What are you trained on?" The bot replies "I don't know" or invents "no data". The customer opens a traindata_ ticket.
An e-commerce AI chatbot training data replies from the sources registry: KB, catalog, conversation exclusions, AI provider and limits, without denying actual usage or improvising.
This guide #908 covers bot_traindata_* intents, TRAINDATAbot flow TDB-1 to TDB-8, and TRAINDATA-REGISTRY-GATE guardrails. Bot pair of the TRAINDATA playbook (#907). Use Case: training data transparency on widget side.
Summary
Why explain training sources on the bot side?
Training data questions arrive mid-chat. Without a registry copy, the bot deviates toward generic privacy or denies the RAG catalog. TRAINDATA-REGISTRY-GATE reduces traindata_ tickets #907.
What the training data bot resolves at tier 1
Listed sources: KB catalog macros map
Clear exclusions: unused conversations map
Honest vendor: AI provider DPA map
Opt-out route: handoff #909 if opt-out
Copy limits: what the bot does not do map
DTC retail example
DTC, registry gate widget. traindata_bot_trust_deflect +36%, traindata_bot_registry_compliance 96% in 5 weeks.
TRAINDATAbot #908 vs TRAINDATA #907, AITRAN #860, LEARNbot #910, and Opt-out #909
Seven transparency training contents, seven distinct roles.
Quick matrix
#908 TRAINDATAbot: bot explaining sources limits exclusions tier 1 vendor
TRAINDATA #907: agents processing traindata TD-4 tickets
AITRAN #860: AI status disclosure distinct training response
AIGENCON #859: AI shop content distinct from bot sources
Opt-out #909: execute exclusion of conversations from training
LEARNbot #910: learning opt-out bot distinct sources
Governance #142: internal rules distinct customer copy
Pipeline: training widget question → #908 tier 1 → handoff #907 or #909 if complex.
Which bot_traindata_* intents should be configured?
Eight training intents mapped TRAINDATA-MAP #907.
Eight bot_traindata intents
bot_traindata_sources_answer: sources_list copy map
bot_traindata_exclusions_answer: exclusions copy map
bot_traindata_conversations_answer: statut_conv copy map
bot_traindata_vendor_answer: vendor_copy DPA map
bot_traindata_limits_answer: limits_copy what this bot does not do map
bot_traindata_opt_out_route: opt_out_process handoff map
bot_traindata_human_review: review_process copy map
bot_traindata_feed_loop: consume LOG #907 registry map
Each response logs registry_version intent traindata_* deflect_or_handoff.
How to consume TRAINDATA-MAP #907?
The bot reads TRAINDATA-MAP #907 + bot fields: sources_list, exclusions_list, statut_conv, vendor_copy, limits_copy, opt_out_process, review_process, traindata_feed_priority.
Guardrails training data
TRAINDATA-REGISTRY-GATE: responses only from registry sync
NO-DENY-USAGE-BOT: do not deny KB catalog if registry indicates so
EXCLUSIONS-EXPLICIT-BOT: always cite exclusions if requested
AITRAN-REROUTE-BOT: disclosure only → #860 AITRAN
OPT-OUT-ROUTE-BOT: refusal of contribution → #909 or #910
VENDOR-GROUNDED-BOT: vendor_copy registry not improvised
DPO-HANDOFF-BOT: formal GDPR escalate agent
TRAINDATA-FEED-LOOP-BOT: LOG #907 enriches sources_copy
TRAINDATABOT-SUP policy in six rules
Six rules for responsible training transparency.
TRAINDATA-REGISTRY-GATE : tier 1 from registry #907 only
NO-DENY-USAGE-BOT : honesty of documented sources
EXCLUSIONS-EXPLICIT-BOT : exclusions_list if conversation questions
LIMITS-COPY-BOT : limits_copy bot limits separate from training
OPT-OUT-ROUTE-BOT : opt_out handoff #909 #910 do not process alone
TRAINDATA-FEED-LOOP-BOT : each LOG #907 review registry within 48 hours
Flow TRAINDATAbot TDB-1 to TDB-8
Eight-step flow: incoming question classify registry answer route log handoff.
TDB-1 Ingest: message training data sources
TDB-2 Classify: traindata_* vs aitran #860 vs privacy
TDB-3 Registry gate: TRAINDATA-REGISTRY-GATE lookup
TDB-4 Tier 1 answer: sources exclusions vendor limits
TDB-5 Route: opt_out_route dpo_handoff aitran reroute
TDB-6 Feed loop: LOG #907 adjusts registry widget
TDB-7 Handoff: complex → agent #907 context
TDB-8 Log: trust_deflect registry_compliance
Example TPL-TRAINDATAbot-SOURCES
“Our answers are based on: {{sources_list}}. We do not use: {{exclusions_list}}. More details: {{policy_link}}.”
TPL-TRAINDATAbot and touchpoints templates
Four short training embed templates.
TPL-TRAINDATAbot-SOURCES
[sources_copy map.] [exclusions_copy map.] REGISTRY-GATE.
TPL-TRAINDATAbot-CONVERSATIONS
[conversations_copy map.] conv_status. OPT-OUT-ROUTE if refusal.
TPL-TRAINDATAbot-VENDOR
[vendor_copy map.] DPA retention. VENDOR-GROUNDED.
TPL-TRAINDATAbot-LIMITS
[limits_copy map.] This bot informs, it does not decide refund.
Touchpoints
“Do you train on me?”: conversations_answer + exclusions
“What data?”: sources_answer tier 1
“I refuse”: opt_out_route #909
LOG TRAINDATA #907: feed_loop sources_copy
Edge cases and reroutes
Five cases out of the standard flow.
"Are you an AI?" alone: #860 AITRAN disclose
AI product description: #859 AIGENCON
Active learning opt-out: #910 LEARNbot
Fine-tune mentioned registry: honest conversations_answer
traindata_ ticket despite bot: feed_loop review registry gaps
Essential traindata_bot KPIs
Five TRAINDATAbot steering metrics and correlation #907.
traindata_bot_trust_deflect: training questions resolved without a traindata ticket
traindata_bot_registry_compliance: % responses aligned with registry
traindata_bot_exclusions_cited_rate: % conv questions with exclusions
traindata_bot_opt_out_route_rate: % opt_out routed #909 #910
traindata_bot_deny_usage_attempts: actual usage denial target 0
Target: deny_usage_attempts at zero and trust_deflect rising.
TRAINDATAbot anti-patterns
Five common mistakes training transparency bot.
Denying catalog usage: NO-DENY-USAGE if RAG active
Improvising vendor: VENDOR-GROUNDED registry only
Forgetting exclusions: EXCLUSIONS-EXPLICIT on conversations
Handling opt-out alone: OPT-OUT-ROUTE #909 #910
Confusing disclosure: AITRAN-REROUTE #860
TRAINDATAbot with Qstomy
Qstomy on Shopify: TRAINDATA-MAP sync #907, sources exclusions vendor templates, opt-out route, KPI traindata_bot dashboard.
Scenario: DTC, 5 tickets/month traindata. Registry gate tier 1. traindata_bot_trust_deflect +36%, traindata_ tickets -33% in 5 weeks.
Explore AI support and request a demo.
Checklist, FAQ and going further
TRAINDATAbot Checklist (8 steps)
Split TRAINDATA-MAP #907: sources exclusions vendor limits
Policy TRAINDATABOT-SUP: 6 REGISTRY-GATE NO-DENY rules
8 intents bot_traindata_*: flow TDB-1 to TDB-8
4 templates TPL-TRAINDATAbot-*: SOURCES CONVERSATIONS VENDOR LIMITS
policy_link: public training transparency page
opt_out_route: handoff #909 #910 configured
Red team training: sources no deny test
Dashboard KPI: traindata_bot_* section 9 + delta traindata_
FAQ
Difference #907?
#907 = agents complex tickets DPO. #908 = bot tier 1 sources exclusions.
Difference #860?
#860 = this response is AI. #908 = training data sources.
Difference #910?
#908 informs sources. #910 executes bot-side learning opt-out.
Promise zero data?
No if indexed catalog KB. Honest NO-DENY-USAGE + EXCLUSIONS.
Going further
This week: sync registry #907, templates sources conversations, opt_out route, measure traindata_bot_trust_deflect.
Enzo
July 1, 2026
