A knowledge base is a searchable library of answers about your products, policies and processes, written once and reused by everyone. A customer-facing one powers self-service and the chatbots that quote it; an internal one keeps your agents consistent. For an online store it holds the rules, while live order and stock data stay in your store.
What is a knowledge base?
A knowledge base is a searchable, structured library of articles that answer the questions people ask about your products, policies and processes. You write each answer once, and customers, agents and AI all reuse it. That is the short answer to “what is a knowledge base”.
The same few questions arrive every day, so one written answer keeps replies consistent. It also lets new or temporary agents answer like experienced ones. At MR MARVIS, the team brought in temporary staff when demand peaked, and those agents had to learn the entire catalogue, all policies and every exception, which did not scale (case study).
There is now a third reader. AI reads the knowledge base too, both support agents and AI shopping agents. Shopify’s own free Knowledge Base app lets merchants view and customise the FAQs that AI shopping agents use to answer questions about their store. Our guide to AI knowledge covers that layer.
The wider picture of running a store’s support lives in the support team guide.
What types of knowledge base are there: internal vs customer-facing?
There are two main types, split by reader. A customer-facing knowledge base, usually called a help centre, answers shoppers’ questions and feeds self-service and chatbots. An internal knowledge base is for your team only: exceptions, goodwill limits, escalation rules and tone. Many stores run both, with some articles shared.
Five formats carry a store’s answers. Each has a different reader and a different way of going wrong.
| Type | Who reads it | What goes in it (store example) | Where it usually lives | The risk if it is wrong |
|---|---|---|---|---|
| Customer-facing help centre | Shoppers and the bots that quote it | ”Can I swap a size?”, delivery times to Spain, how to measure for the right fit | Help centre in the help desk, or pages on the store | Customers act on it, and you are held to it (see the Air Canada case in the section on wrong answers) |
| Store policy pages | Shoppers at checkout | Return, privacy, terms of service, shipping, legal notice, contact information and subscription policies | The store itself, linked in the checkout footer | The legally binding version disagrees with the help article |
| FAQ page or on-page FAQ blocks | Shoppers mid-purchase | Short Q&A on product or cart pages | Theme sections | Duplicates the help centre and drifts |
| Internal knowledge base | Your agents | Goodwill limits, when to refund without a return, carrier claim steps, supplier contacts | Help desk internal articles, a wiki or shared docs | Inconsistent answers between agents |
| Macros and saved replies | Your agents | Pre-written answers for repeated tickets | The help desk | The macro still states last season’s rule |
Shopify lists exactly the policy pages in the second row: return, privacy, terms of service, shipping, legal notice, contact information and subscription policy. Once added, they are linked automatically in the footer of your checkout pages. How an FAQ page and a knowledge base split the work is in our FAQ guide.
What belongs in a knowledge base for an online store?
Write the questions that arrive most as articles: delivery and tracking, returns and refunds, product care and sizing, payments and discounts, account and order changes, plus the store policies themselves. Use your customers’ words, one rule per article, and keep order-specific answers out of the articles.
| Article type | Example title in a shopper’s words | What it must state | What an article can’t answer |
|---|---|---|---|
| Delivery and shipping | ”How long does delivery to Germany take?” | Carriers, times per market, cut-off times | Where this parcel is now |
| Tracking | ”How do I track my order?” | Where tracking links are sent | The status of this order |
| Returns and exchanges | ”Can I return sale items?” | Window, condition, who pays, exceptions | Whether this order is still inside the window |
| Refunds | ”When will I get my money back?” | Processing time per payment method | Whether this refund has been issued |
| Product care, sizing and materials | ”Does the slim fit run small?” | Size guide, materials, care | Stock of a size right now |
| Payments and discounts | ”Why didn’t my code work?” | Stacking rules, exclusions | Why this code failed on this cart |
| Account and order changes | ”Can I change my address after ordering?” | Until when it is possible, and how | Changing this order |
| Policies | ”What is your return policy?” | The wording on your policy page | Whether it applies to this order |
The writing rules come from Knowledge-Centered Service (KCS®), described in the Consortium for Service Innovation’s v6 Practices Guide. KCS® is a service mark of the Consortium for Service Innovation™. It says to capture the requestor’s own words and phrases, which makes articles findable, and to use a consistent structure and a simple template. For a store, title the article “my parcel says delivered but it isn’t here”, not “Delivery exceptions policy”.
The right-hand column is why a knowledge base alone does not resolve a store’s busiest tickets. A rule is one thing; this customer’s order is another. Our pages on order status and chatbots cover what an article-only bot can and can’t answer.
How do you build and maintain one?
Start from real tickets, not a blank outline. Write the top questions first, give each article one owner, then keep the library alive in the flow of work: search before you answer, fix or flag what is wrong when you use it, and review articles on a schedule. That is the core of Knowledge-Centered Service.
The Consortium’s site now brands KCS as Knowledge-Centered Success, but the v6 guide still uses Service.
KCS works in four steps, which translate directly to a store:
- Capture. Write the article while you answer the ticket, in the customer’s words.
- Structure. Use one simple template, so every article reads the same way.
- Reuse. “Search early, search often”, so you do not solve a solved question twice.
- Improve. Reuse is review: if an article is wrong or unclear, you “flag it or fix it”.
To know it is working, KCS measures the reuse rate (the share of cases closed with an article linked) and reuse accuracy (the article resolved the issue). The guide recommends link rates above 60% and link accuracy above 90%, and says accuracy matters more than rate. Those are KCS’s targets, not ours.
A build order that suits small teams:
- Export the last 100 to 200 tickets and group them by question.
- Write the top ten as articles.
- Add the store policies.
- Give each article an owner and a review date.
- Add internal notes for exceptions.
- Review monthly, and after every policy, carrier or price change.
Some tools automate the upkeep. In Zendesk, content blocks share one piece of content across many articles, and updating a block updates every article that uses it. Article verification can unverify articles after an interval and remind their owners to review them. These are examples of what to look for, not a recommendation.
KCS also offers a rule of thumb that 80% of requests have already been answered somewhere in the organisation. The source labels it a rule of thumb, and so do we.
What can’t articles do for a store, and what happens when they’re wrong?
Articles explain rules. They don’t know this customer’s order, stock or refund, and they can’t act. And when an article and a bot disagree, the customer can hold you to either. Air Canada learnt that in 2024, when a tribunal made the airline honour what its chatbot had told a customer.
In Moffatt v. Air Canada, 2024 BCCRT 149, the chatbot told a grieving customer in November 2022 that he could apply for a reduced bereavement fare after travelling, within 90 days of the ticket being issued. The chatbot’s own link went to a Bereavement travel page that said the policy does not include retroactive bereavement fares.
Air Canada in effect argued the chatbot was a separate legal entity. The tribunal called this “a remarkable submission” and said the airline is responsible for all the information on its website: “It makes no difference whether the information comes from a static page or a chatbot.” It found negligent misrepresentation and ordered a partial refund of CAD 650.88, plus interest and fees, about CAD 812 in total.
The store lesson, in our view: the moment a bot or an agent restates a policy, you have two versions. Keep one source per rule, make sure whatever answers customers shows which source it used, and review the article whenever the rule changes. This is not legal advice. For when a bot should hand over instead of guessing, see escalation.
What should you look for in knowledge base software?
Good knowledge base software for a store is the one your team will keep current. Judge it on how fast you can write and update articles, whether shoppers and bots find the right answer, how it flags stale content, internal versus public control, languages, and whether it sits inside the help desk you already use.
We list criteria, not vendors, because the right tool depends on the help desk you already run. Use this table in a trial.
| Criterion | Question to ask in a trial | Why it matters for an online store |
|---|---|---|
| Editor and templates | Can a non-technical person publish a structured article in minutes? | The people who know the answers are rarely developers |
| Reusable snippets | Can one policy line update every article that uses it? | A changed return window is fixed once, not in twelve places |
| Ownership and review dates | Does each article have an owner and a reminder to re-verify? | Articles without owners drift out of date |
| Search and failed searches | Does it show what people searched for and didn’t find? | Missing articles show up as repeat tickets |
| Internal and public in one place | Can internal notes sit next to public articles without leaking? | Goodwill limits must never reach shoppers |
| Languages and markets | Can you keep one article per market or language, with review? | Delivery and returns rules differ per country |
| Help desk fit | Can agents insert an article into a reply and link it to the ticket? | That link is what KCS measures |
| Feedback | Can readers and agents flag an article as wrong? | ”Flag it or fix it” needs a flag |
| Store data | Can it show anything order-specific, or only explain? | Most help-centre articles only explain, so check this |
| AI readiness | Can AI agents read it, and do answers cite the article? | Bots now quote your articles to customers |
| Export and API | Can you take your content with you? | You are not locked in if you change tools |
Check whether the help desk you already run includes a knowledge base before you buy a separate tool. Our help desk guide covers that choice, Zendesk Knowledge is one worked example, and the AI layer has its own guide, linked above.
Where does Engaige fit?
Engaige (that’s us) isn’t knowledge base software: it is the e-commerce-native AI support agent that works inside the help desk you already run and answers from your knowledge base alongside live store data. Engaige AI configures the agent from your conversations and inputs, and at MR MARVIS the agent handles over 120,000 questions a year, resolving over 60% of the conversations it’s involved in.
On Shopify, Engaige combines live Admin API data (catalogue, SKUs, pricing across markets, inventory, orders, customers) with the marketing content from your storefront, plus any docs you add. Engaige answers from your real data, so your docs can explain the returns rule while the order record shows whether this order qualifies.
Every answer and action shows why, with the source it used, so your team can see what the agent relied on.
Hi, where is my order? It's been eight days and the tracking hasn't moved.
Shopify
Order 48219 · fulfilled
Picqer
Still processing Contradicts Shopify
DHL
Returned to depot · 3 delivery attempts
Settles it: the parcel is not with the customer
Reasoning…
Not deliverable. Reship to a confirmed address, booked in Picqer.
Suggested reply
Your parcel went back to the depot after three delivery attempts, so it will not arrive. I have booked a reshipment to the address you confirmed, arriving Thursday.
MR MARVIS uses that view to spot where information is missing or unclear.
Within the AI we can see the reasoning, and how it comes to a response. That shows us where information is missing or unclear, and helps us improve our workflow.
You go live in Agent Assist in hours: Engaige drafts replies inside your help desk, and nothing closes without a human. Every reply your team sends becomes training data, and Engaige AI proposes changes to policies that you approve. You then flip ticket types to autonomous one at a time, most brands in about a week, up to 90% of supported tickets. Autonomous from day one is also supported.
Keep your helpdesk. Replace the AI. The support team guide has the full picture, or book a demo to see it on your own articles and orders.
Frequently asked questions
What is a knowledge base in simple terms?
A searchable library of answers, written once and reused by customers, agents and AI.
What is the difference between an internal and an external knowledge base?
The reader and the content. An external one serves shoppers; an internal one serves your team, so exceptions and goodwill limits stay there.
Why does an online store need a knowledge base?
The same few questions arrive every day. Writing each answer once and reusing it keeps replies consistent and lets new or temporary agents answer like experienced ones.
What should an online store put in its knowledge base first?
The top ten questions from real tickets, plus the store policies Shopify lists: return, privacy, terms of service, shipping, legal notice, contact information and subscription.
How often should you review help articles?
Whenever the rule changes, and on a set schedule with a named owner. In KCS, “reuse is review”: whoever uses an article flags or fixes it.
What is Knowledge-Centered Service (KCS)?
A method from the Consortium for Service Innovation: capture, structure, reuse and improve knowledge in the flow of work. KCS® is a service mark of the Consortium for Service Innovation™.
Which knowledge base software suits an online store?
The one your team keeps current, ideally inside the help desk you already run. Use the criteria table above in a trial.
Can a knowledge base answer “where is my order?”
Only how tracking works. The status needs live order data, which is why Engaige reads the order as well as the docs.
Does Engaige write knowledge base articles?
It reads your knowledge base rather than writing it: Engaige works alongside it, answers from the docs you add plus live store data, and shows which source each answer used.
