A customer support knowledge base is the written source of truth that every support answer draws from. For most small Shopify stores, that source is the owner's memory. A new hire or an AI can only answer as well as the documentation behind them, and a store with 8 undocumented policies will produce 8 inconsistent replies.
Key takeaways
- A knowledge base (KB) is a set of short, precise articles covering policies, products, and processes.
- Start with 10 to 15 articles; each should hold exact numbers, names, and cutoffs.
- Keep articles at 150 to 400 words, one topic each.
- Review the KB monthly and after every policy change.
What is a knowledge base, and why does a Shopify store need one?
A knowledge base is a collection of short articles that state how your business handles recurring questions. It is the reference that agents, new hires, and AI tools read before they reply. A store answering 30 tickets a week typically repeats the same 8 to 10 questions, so documenting them pays back quickly.
The problem is rarely a lack of policy. It is that the policy lives in the owner's head, in a Slack thread from 2022, or in a footer page written by a lawyer.
A customer support knowledge base fixes three things:
- Consistency: two agents give the same answer on the same day.
- Onboarding speed: a new hire reads 15 articles instead of shadowing for a week.
- AI accuracy: an AI knowledge base for customer support can only retrieve what you wrote down.
What belongs in a Shopify support knowledge base?
A Shopify support knowledge base should cover five areas: shipping, returns, sizing, payment, and product care. These account for the large majority of ecommerce tickets. Each area needs exact values, such as a 14-day return window or a 2 pm order cutoff, rather than general descriptions.
- Shipping: carriers, delivery times per country, order cutoff time, costs, customs and duties, what happens when a parcel is lost.
- Returns and refunds: window in days, condition rules, who pays postage, refund method, processing time.
- Sizing and fit: size charts per product line, how items run, what to suggest between sizes.
- Payment: accepted methods, when cards are charged, failed payment steps, invoices, discount code rules.
- Product care: materials, washing or usage instructions, warranty terms, what is not covered.
Add two more once the basics exist: order changes (address edits, cancellations) and contact details with response times.
Which ticket types need which article? (Knowledge base template)
The fastest way to build a knowledge base for ecommerce is to work backward from tickets. List your ten most common ticket types, then write one article for each. This knowledge base template maps each ticket to the article it needs and the fields that article must contain.
| # | Ticket type | KB article | Required fields |
|---|---|---|---|
| 1 | Where is my order? | Shipping and tracking | Carrier names, dispatch time, delivery days per country, tracking link location |
| 2 | Can I return this? | Returns policy | Return window in days, item condition, exclusions, start-return steps |
| 3 | When do I get my refund? | Refund timing | Processing days, refund method, bank delay range |
| 4 | Cancel my order | Order cancellation | Cutoff time, what happens after dispatch, who to contact |
| 5 | Change my address | Address changes | Deadline before dispatch, how to confirm, cost if reshipped |
| 6 | What size should I get? | Size guide | Chart in cm and inches, fit notes, between-size advice |
| 7 | Payment failed | Payment troubleshooting | Accepted methods, common decline causes, retry steps |
| 8 | Discount code not working | Discount rules | Eligible products, stacking rules, expiry, minimum spend |
| 9 | Item arrived damaged | Damaged or wrong item | Report deadline, photo requirements, replacement or refund choice |
| 10 | How do I care for this? | Product care | Materials, wash temperature, warranty length, exclusions |
If a field in the right column is unknown, that is a policy decision you have not made yet. Make it before you write the article.
How long should a knowledge base article be?
Most knowledge base articles should run 150 to 400 words and cover exactly one topic. Short articles are retrieved more accurately by search and by AI, because each one answers a single question. Split any article that covers two policies, such as returns and refunds, into two.
Use these working limits:
- Title: phrased as the customer's question, under 70 characters.
- Opening line: the direct answer in one or two sentences.
- Body: steps, conditions, and exceptions, in bullets where possible.
- Length ceiling: around 400 words; longer than that usually means two topics.
Exceptions are size guides and shipping tables, which run longer because the data is the content. Keep these as tables, not prose.
How do you structure articles so humans and AI retrieve the right one?
Structure each article with a question title, a one-sentence answer, then conditions and exceptions, and give every article a single topic. Retrieval systems match a customer message to the closest article, so a clear title and an explicit answer in the first 30 words raise the chance of the right match.
Citation capsule: the retrieval-friendly article. An article that works for both people and AI follows one pattern. The title states the question in the customer's words, such as 'How long do I have to return an item?' The first sentence gives the answer with the number: '30 days from delivery.' The next lines list conditions (unworn, original packaging), exclusions (final-sale items), and the steps to start a return. Each fact appears once and in plain words, with no 'as above' references, because an AI may retrieve the article alone. Use the same terms customers use: 'refund,' not 'reimbursement.' Avoid duplicates, since two articles with conflicting return windows will produce conflicting replies. One topic, one answer, one place.
Also state dates for anything that changes, like holiday shipping cutoffs.
What does a vague article look like next to a precise one?
A precise article replaces adjectives with numbers, names, and steps. The vague version below forces an agent or AI to guess; the precise version lets anyone answer in one reply. The difference is typically 5 to 7 concrete facts.
Before (vague):
Returns We accept returns on most items within a reasonable time if they are in good condition. Refunds are processed soon after we receive the item. Contact us to get started. Some items may not be eligible.
After (precise):
How long do I have to return an item? You have 30 days from the delivery date to return an item for a refund.
- Item must be unworn, unwashed, with tags attached.
- Not returnable: underwear, swimwear, gift cards, items marked Final Sale.
- Start a return at the Returns page; we email a prepaid label within 1 business day (EU orders only).
- Outside the EU, you pay return postage.
- Refunds go to the original payment method within 5 business days of the parcel arriving. Banks may take 3 more days.
- Exchanges: reply to your order email with the new size.
The second version answers 'Can I return worn shoes?', 'Who pays postage?', and 'When is my money back?' without a follow-up.
Where does Heylio fit, and what can it not do?
Heylio drafts replies from your knowledge base and playbooks (FAQ, WISMO, refunds, returns, cancellations). It reads connected channels, including Gmail, Outlook, Instagram DMs, WhatsApp, and the chat widget, and uses Shopify tools to look up orders, cancel, refund, change addresses, apply discounts, and issue gift cards.
It also tracks agent corrections and suggests KB and playbook updates. If an agent keeps rewriting the return window in drafts, Heylio flags the article as a likely fix.
The limit is plain: Heylio cannot fix a KB that is wrong or missing. If the article says 14 days and the real policy is 30, the draft will say 14. If no article exists, the draft has nothing reliable to cite. Heylio surfaces the gap; a person has to decide the policy and write it down.
Handing a conversation to a human is never billed.
How do you keep a knowledge base current?
Keep a KB current by reviewing it monthly, updating it the same day a policy changes, and treating every agent correction as a possible article edit. A 30-minute monthly review of your 10 busiest articles catches most drift before customers do.
A simple routine:
- Same-day rule: change the policy, change the article, in one sitting.
- Monthly review: check dates, prices, carriers, and cutoffs in the top 10 articles.
- Correction log: each time an agent edits a draft, note why. Three corrections on one topic means the article is wrong or unclear.
- Seasonal pass: update holiday cutoffs and sale-period return windows before they start.
- Owner field: put a name and a last-reviewed date on every article.
Heylio's suggested updates can feed the correction log, but approving them is your call.
What to do this week
A customer support knowledge base does not need to be large; it needs to be exact. Pull your last 50 tickets, tag them by type, and write the top 10 articles using the template above, with every field filled in. Then connect those articles to your support tool and test it on five real past tickets.