E-commerce
June 26, 2026
Structuring an e-commerce knowledge base is not the same as writing a long FAQ or deploying an AI chatbot. The KB is the editorial asset that powers self-service, agents, and RAG models: clear taxonomy, atomic articles, up-to-date policies, and linking from the site and the bot widget.
Tension arises when the site's return policy says 30 days, the helpdesk macro says 14 days, and the bot improvises 60 days. Without a central hub, each channel amplifies a different version of the truth.
This guide focuses on content architecture: categories, article templates, customer journey linking, and chatbot connection. It complements our helpdesk, bot, and KB comparisons without reopening the tool-by-tool debate.
Summary
Knowledge base, FAQ, and chatbot: what is the difference?
Three blocks that are often confused, three different roles.
FAQ Page
Marketing summary or storefront: five to fifteen visible questions, often static. Useful near checkout, insufficient on its own for high-volume support.
Knowledge Base (Help Center)
Structured library of stable answers: policies, procedures, product guides, troubleshooting. Each article has a URL, an owner, and a revision date. It is the source of editorial truth.
AI Chatbot
Conversational interface that retrieves from the KB and sometimes queries Shopify in real time. It is not a substitute for the KB: it is the layer that makes content accessible in natural language.
See helpdesk vs chatbot vs KB and e-commerce FAQ. eDesk points out that an "AI-ready" KB covers customer articles, product data, policies, and rules per channel (eDesk, AI knowledge base).
What taxonomy should be adopted for an e-commerce help center?
Taxonomy must reflect customer intentions, not your internal organizational chart.
Six pillar categories for a DTC store
Ordering and paying: promo codes, payment methods, invoice
Delivery and tracking: lead times, carriers, WISMO, blocked parcel
Returns and refunds: eligibility, procedure, lead times
Products: sizing, compatibility, care, warranty
Account and data: login, newsletter, GDPR
Before purchasing: trust, authenticity, contact
Depth rules
Docsio recommends 5 to 8 top-level categories and a maximum of three clicks to reach an article (Docsio, KB design 2026).
Digital Applied adds that each additional level roughly halves discoverability: keep it shallow (Digital Applied, IA playbook).
Use tags (SKU, collection, market) for cross-referencing, not as primary navigation.
How to structure an article for humans and bots?
Articles written for human scrolling often break AI retrieval. Adopt an "answer-first" structure.
Recommended article model
Customer-centric title ("How do I return an item?")
Direct answer of 40 to 60 words at the top of the page
Numbered steps if it is a procedure
Exceptions and edge cases
Links to related articles
Visible last updated date
Concrete example
Bad: "Our quality commitment accompanies you on your return journey..." followed by three brand paragraphs. Good: "You have 30 days after receipt to return an unworn item, with tags intact." then the five numbered steps up to the refund. The RAG chunk retrieves the first sentence; the human customer reads it without scrolling.
Errors to avoid
An article that mixes return, exchange, and refund
Instructions based on screenshots without text
Three-paragraph marketing intro before the answer
Outdated UI references ("click on the My Orders v2 tab")
HappySupport estimates that a set of 15 to 20 well-structured articles beats 200 poorly organized PDFs for bot retrieval (HappySupport, KB structure).
What content should be prioritized at launch?
Don't wait until you have a hundred articles to launch. Start with the twenty topics that generate 80% of the tickets.
Four-step method
Export 90 days of tickets and sort by volume
Identify the ten dominant intents
Write or migrate one article per priority intent
Publish with a stable URL before indexing the bot
Non-negotiable content on day 1
Delivery policy and lead times by zone
Complete return policy (aligned with return policy drafting)
Order tracking and WISMO
Declined payment and 3-D Secure
Contact and support response times
HappySupport advises starting from the 20 most frequent ticket topics, not the product features list (HappySupport, build KB).
How to link the KB from the website and the purchase journey?
An invisible KB reduces neither tickets nor bot hallucinations. Interlinking is an integral part of the structure.
Priority Placements
Footer: visible "Help Center" link
Product Page: returns, delivery, size guide
Cart and Checkout: return policy reminder
Transactional Emails: tracking, delivery, returns
Chat Widget: article suggestions by intent
Order Tracking Page: contextualized WISMO FAQ
Internal Search
Customers search for "returns," "refunds," "tracking." Your internal engine must direct them to the correct article, not to an unrelated product page. Test ten real queries per month from the help center search bar: note the click-through rate on the first result and the queries that return zero articles.
How to prepare the database for RAG and the AI chatbot?
RAG (retrieval augmented generation) reads your articles at the moment the question is asked. The structure determines the quality more than the model.
E-commerce RAG Principles
One topic per article: avoids diluted chunks
Metadata: topic, URL, revision date, owner
Chunking by headings rather than blind fixed size
Hybrid search: semantic + keywords for SKUs and references
Live data via API: stock, price, order status kept out of the static index
Bot freshness test
After each policy change, ask the bot ten questions within 24 hours. If a single answer contradicts the help center, block deployment until full re-indexing is complete.
Heeya points out that obsolete documentation cited with confidence causes more harm than an "I don't know" followed by an escalation (Heeya, RAG support 2026).
Branch8 distinguishes between catalog/policy in RAG and orders in real-time API calls (Branch8, RAG e-commerce).
See cleaning FAQ before bot and Shopify data.
How to separate public content and internal procedures?
Not everything should be public. Clearly separate customer content and internal procedures.
Public content (bot indexed + SEO)
Policies, return procedures, product guides, pre- and post-purchase FAQs.
Internal content (agents only)
Exceptional refund thresholds, fraud procedures, dispute scripts, carrier contacts, sensitive brand decisions.
Practical rule
If an article contains internal thresholds or legal phrasing not intended for the customer, keep it out of the bot index. The bot quotes public content; the agent consults internal content via helpdesk or private wiki.
Document the list of bot-indexed URLs in a shared support + marketing file. When redesigning a policy, first update the public help center, wait for re-indexing, then communicate to agents: the reverse order multiplies visible customer contradictions.
What governance is needed to keep the KB up to date?
A KB without governance becomes obsolete in a matter of months. Treat it like a product with an owner and a calendar.
Mandatory fields per article
Owner (responsible person)
Last review date
Next planned review
Version or changelog if policy changes frequently
Intent and channel tags
Recommended frequency
Weekly: top 5 searches with no results + unresolved bot intents
Monthly: review of the ten most viewed articles
Quarterly: taxonomy audit and orphan articles
Immediate: any change to delivery, return, or promo policy
Fifteen-minute review ritual
Every Friday, the KB owner opens the "searches with no results" report and the list of bot escalations for the week. Two questions: is an article missing? Is an existing article poorly titled? Just one correction per week is enough to avoid editorial debt.
Ticket-to-article loop: if a topic comes up more than 15 times a month without a dedicated article, create it before expanding the bot. See conversation analytics.
What examples per intent: WISMO, return, size, payment?
Four intents cover a large part of e-commerce volume. Structure them into separate articles, not in a single block.
WISMO
Article 1: normal times per carrier. Article 2: package in transit without scan for five days. Article 3: incorrect address or package returned to sender. Each article begins with "Where is my order if..." with the answer in one sentence, then the steps. The bot queries Shopify for the actual status, then cites the article that matches the case (normal delay vs anomaly).
Returns and exchanges
Structure a mini decision-tree in text: within the legal period? sealed product intact? outlet or end-of-series item? Indicate the return portal link, the refund processing time after warehouse reception, and the difference between exchange vs refund. Human handoff if out of policy: bot-to-human handoff.
Size and fit
Measurement chart, advice between two sizes, free exchange. Link from the fashion product pages to reduce "wrong size" returns.
Payment and invoice
3-D Secure, card failure, double charge, corporate invoice. Reassuring tone, actual refund times, link to bank if needed.
Which KPIs should be used to manage the knowledge base?
Steer the KB with actionable indicators, not just pageviews.
Self-service rate: resolution without a ticket after reading an article
Search queries with no results: content gaps
Bot intents coverage: % of intents with a source article
Tickets avoided per article: before/after publication
Post-article CSAT: "did this article help?"
Average review time: freshness of the hub
Compare the 48-hour re-contact rate on covered intents: if the ticket volume does not decrease despite the article, the content is poorly written, badly placed, or incorrect.
How does Qstomy rely on a unified KB?
Qstomy sits on top of a unified KB and your Shopify data: a conversational layer that cites your articles, checks orders and stock, and then escalates with context.
One source, multiple channels
Help center, onsite widget, and agents align on the same facts. Qstomy consumes the hub rather than recreating a micro-FAQ within the bot.
Quantified DTC Scenario
A fashion brand structures 22 KB articles across six pillar categories, then connects Qstomy in RAG + Shopify lookup. Before: generic bot with 34% escalation and 19% of answers contradicted by agents. 90-day pilot objective: 52% bot resolution on documented intents, escalation under 22%, zero return policy contradictions in monthly audits, 120 tickets/month avoided on WISMO and returns.
Explore customer support AI, Shopify integration, and request a demo. See customer support glossary.
Which playbooks should be launched this week?
Playbook 1: Taxonomy mapping in 60 minutes
List your six pillar categories. For each, note 3 to 5 existing or missing articles. The visible gaps become your editorial backlog.
Playbook 2: FAQ to KB migration
Take your ten current FAQ questions. Transform each into a dedicated article with a stable URL, answer-first at the top, and revision date. Redirect the old anchor links.
Playbook 3: RAG test on 20 questions
Ask the bot twenty real customer queries. Is the correct article cited? Is there any hallucination? Correct titles and answer-first structure before expanding the scope.
Playbook 4: Weekly ticket loop
Every Monday: top 5 tickets without a corresponding article. Any topic with over 15 occurrences/month deserves an article before creating a new bot rule.
Useful internal linking
Comparison: helpdesk vs bot vs KB
Automation mistakes: support automation mistakes
FAQ: FAQ and tickets
Insights: insights from support
This week, audit your ten most viewed articles: answer-first structure, revision date, links from PDP and checkout.
Enzo
June 26, 2026
