An ecommerce FAQ your AI chatbot can use is a set of short, self-contained answers where each one states the rule in its first sentence, uses numbers instead of adjectives, spells out the exceptions, and says when a person takes over. Build it from your last 100 customer conversations, cover the ten topics every store gets asked about, and test the chatbot against real questions before you switch it on. The steps below show how, with before and after examples you can adapt.
Why chatbot answers depend on your FAQ
An AI support agent does not know your store. When a shopper asks a question, it searches the pages and documents you gave it, pulls the passages that look relevant, and writes an answer from them. If those passages are vague, the answer is vague. If two pages disagree, the bot may pick either one.
That second failure is the expensive one. In Moffatt v. Air Canada, a Canadian tribunal held the airline responsible after its chatbot told a customer he could claim a bereavement discount after travelling, which the airline's own bereavement page did not allow. The tribunal ordered a refund of CA$650.88 plus interest and fees. A small store is unlikely to end up before a tribunal, but a bot that promises a free return your policy does not offer creates the same argument in your inbox.
So the work is mostly writing, and it pays twice: the same answers serve shoppers who read your help pages and the bot that quotes them.
Step 1: Audit your last 100 support conversations
Start from what customers actually ask, in their own words.

- Export or scroll through your last 100 emails, chats and DMs. If you have fewer, use everything since launch.
- Give each one a topic tag: shipping time, tracking, returns, sizing, discount code, damaged item, and so on.
- Count the tags. You will probably find that a handful of topics make up most of the pile.
- For each topic, copy two or three of the exact phrasings customers used. "Where's my stuff" and "has my order shipped" are the same question, and the bot needs to recognise both.
- Mark whether a written answer already exists, and where it lives.
A simple worksheet is enough. Here is what a few rows might look like for a hypothetical clothing store:
| Topic | Count | How customers phrase it | Written answer? |
|---|---|---|---|
| Delivery time | 18 | "When will it arrive", "how long to Canada" | Partly, in the footer |
| Returns | 14 | "Can I send it back", "return label?" | Yes, but in legal language |
| Sizing | 11 | "Do these run small", "I'm 5'4, which size" | Size chart only |
| Discount code | 7 | "Code not working", "WELCOME10 invalid" | No |
If you run support through a shared inbox, the tags may already exist. Our free chat transcript to help article tool can turn a good reply you already sent into a first draft.
Step 2: Cover the ten topics every store needs
Your audit sets the order, but almost every online store needs written answers on these ten:
- Shipping cost and delivery times, by region
- International shipping, duties and taxes
- Order tracking and what to do when tracking stops moving
- Changing or cancelling an order, and the cut-off
- Returns and exchanges: window, condition, who pays shipping
- Refunds: method and how long they take to appear
- Damaged, faulty or wrong items
- Payment methods, discount codes and gift cards
- Sizing, fit or compatibility, depending on what you sell
- Product care, warranty and how to reach a person
Write one answer per question, not one page per topic with everything blended together. A bot retrieves passages, and a passage that answers one question cleanly is far easier to find than paragraph six of a long policy page.
Step 3: Write shipping and delivery answers
Shipping answers fail when they use words like "fast", "usually" or "a few days". Customers read those as promises, and a bot repeats them with confidence.
Before:
We ship quickly and most orders arrive within a few days. International orders may take longer.
After:
Orders placed before 2 pm Eastern on a weekday ship the same day. US delivery takes 3 to 5 business days with standard shipping ($6, free over $75) or 1 to 2 business days with express ($18). Canada takes 6 to 10 business days. Other countries take 7 to 14 business days, and the customer pays any import duties. Every order gets a tracking link by email when it ships. If tracking has not updated for 5 business days, contact us and we will open a claim with the carrier.
The rewrite answers the follow-up questions before they are asked: cut-off time, cost, per-region times, duties, tracking, and what happens when tracking stalls. It also gives the bot a clear stop line: a stuck parcel goes to a person.
Step 4: Write return and refund policy in plain language
Return policies are usually written to protect the store, which makes them hard to read and harder for a bot to apply. Keep the legal version on its own page if you need it, and write a plain one for the FAQ.
Before:
Merchandise may be eligible for return subject to inspection and the conditions set out in our terms. Certain exclusions apply.
After:
You can return unworn items with tags attached within 30 days of delivery. Start a return from the link in your shipping email. Return shipping costs $5, deducted from your refund, and exchanges for a different size are free. Sale items, underwear and gift cards cannot be returned. Refunds go back to the original payment method within 5 business days of the return reaching us. If the item arrived damaged or wrong, do not use the return link: contact us with a photo and we will replace it at no cost.
Notice the exclusions are named. "Certain exclusions apply" is exactly the kind of phrase that leads a bot to guess. When a refund needs judgment, such as an item worn once or a return on day 34, say so in the answer: "For anything outside these rules, our team decides case by case." That sentence tells the bot to hand off instead of improvising.
Step 5: Add product care, sizing and compatibility pages
Pre-purchase questions are where a chatbot earns its keep, because they come in at all hours from shoppers who are about to leave. They are also where generic answers hurt most.
- Sizing: give measurements in both inches and centimetres, say how each line fits ("runs one size small in the shoulders"), and name the model's height and size in photos.
- Compatibility: list exact models, versions or dimensions. "Fits most phones" invites a wrong answer; "fits phones up to 6.7 inches with a case up to 2 mm thick" does not.
- Care and materials: washing temperature, what not to do, and what voids the warranty.
Put these on the product page and in the FAQ, with the same wording in both places. If they differ, you have created the Air Canada problem on a smaller scale.
Step 6: Format for both customers and AI
The formatting that helps shoppers skim is the same formatting that helps a bot retrieve the right passage. Follow these rules for every answer:
- One question per heading, phrased the way customers ask it: "Can I change my delivery address?" rather than "Address amendments".
- Put the rule in the first sentence. Conditions and exceptions come after.
- Make each answer stand alone. "See above" and "as mentioned" mean nothing when a bot pulls one passage out of the page.
- Use numbers, dates and prices instead of adjectives.
- Write each fact in one place, then link to it. Two copies of the returns window will eventually disagree.
- Keep important rules in sentences, not only inside tables or images. A size chart image is invisible to most bots.
- Add a stop line where a person must decide: refunds outside policy, damaged items, anything involving a specific order.
One thing has changed for search: Google says in its FAQPage documentation that FAQ rich results are no longer shown in Search as of 2026. Write the FAQ for shoppers and for your chatbot, not to win an expandable snippet.
Step 7: Test questions and fix gaps
Never switch an AI agent on without testing it against your own questions. Build a test set of about 35 questions:

- 20 common questions, copied word for word from your audit
- 10 awkward ones: misspellings, two questions in one message, a question about a product you do not sell
- 5 that must reach a person, such as "my parcel arrived smashed" or "I was charged twice"
Ask each one and grade the answer:
| Grade | What it means | What to fix |
|---|---|---|
| Correct | Right answer, right tone | Nothing |
| Incomplete | Right, but missing a condition or a number | Add the missing detail to the source answer |
| Wrong | States something your policy does not say | Find the vague or conflicting passage and rewrite it |
| Should have handed off | Answered something a person should decide | Add a stop line, or adjust the agent's handoff rules |
| No answer | Says it does not know | Write the missing answer, if it is a real question |
Fix the source answer. If the bot gets returns wrong, the returns answer is unclear, and a shopper reading it probably would be confused too. Re-run the full set after each round of fixes. The free AI agent playground lets you try questions against your own content before anything goes live.
Step 8: Keep content current when policies change
A chatbot answers from whatever it last read, so an outdated FAQ produces outdated answers instantly and at volume. Three habits prevent most of it:
- Give the FAQ an owner. One person approves every change to shipping, returns and pricing answers.
- Update before the change, not after. Holiday shipping cut-offs, a new carrier or a sale with different return rules should be written and synced the day before they start.
- Remove expired content. An old Black Friday returns extension that stays on the site will be quoted in March.
After any policy change, re-run the test questions for that topic. It takes ten minutes and catches the conflicting page you forgot about.
Step 9: Publish it as a help center
Once the answers exist, put them where shoppers and your AI agent can both find them. Group them into a small help center with one section per topic, link it from your footer, your order confirmation and shipping emails, and the chat window. The help center overview covers structure and search.

If you use Sellio, its AI agent answers on website chat only, from your website and the documents you add, and hands the conversation to a person when it is unsure. You can cap its usage so a busy week cannot run up the bill. The AI chatbot for ecommerce page shows how that handoff works, and the free FAQ generator drafts first answers from your policies.
Ecommerce FAQ checklist
- Built from at least 50 to 100 real conversations, with customers' own phrasings
- All ten core topics covered, one question per heading
- Every answer leads with the rule and uses numbers
- Exclusions named, never "certain conditions apply"
- Each fact written once, linked elsewhere
- Stop lines on refunds, damage and anything about a specific order
- Product pages and FAQ use the same wording
- 35-question test set passed before launch
- A named owner and a date of last review
The help center gap checklist turns this into a printable audit.
Common mistakes
- Copying a template FAQ with answers that do not match how your store actually works.
- Hiding the real policy in terms and conditions and writing a friendlier, different version in the FAQ.
- Letting the bot answer order-specific questions from static pages, which leads it to guess a delivery date.
- Writing long answers that cover five questions, so the bot quotes the wrong part.
- Testing only the questions you expect, then launching.
- Updating the website but not the documents the bot was trained on, or the other way round.
FAQ
How many questions should an ecommerce FAQ have?
Enough to cover what customers actually ask, which for most small stores is 20 to 40 questions across the ten core topics. Ten precise answers beat fifty vague ones, for shoppers and for a chatbot.
Can an AI chatbot use my existing FAQ page?
Usually, yes. Most AI agents can read your website pages and uploaded documents. Whether it answers well depends on the writing: answers that lead with the rule, use numbers and stand alone work far better than long policy paragraphs.
Should the chatbot answer "where is my order" questions?
Only if it can look up live order data. From a static FAQ it can explain delivery times and how tracking works, but it should hand questions about a specific order to a person rather than guess.
How often should I update my FAQ?
Before every policy change, and with a full review every quarter. Re-run your test questions after each update so the chatbot and the page stay in step.
Do FAQ pages still help with SEO?
They help shoppers find answers and can rank for long-tail questions, but Google no longer shows FAQ rich results in Search. Write for the reader and your chatbot first.




