AI Chatbot for WooCommerce: 2026 Setup Guide

W3Techs counts WooCommerce on 7.8% of all websites and on 11.5% of the sites whose content management system it can identify (W3Techs, data for 20 October 2026, checked 21 October 2026). Most of those stores are run by a handful of people, and the people who pack the parcels also answer the questions, most of which are already answered somewhere on the site.

A chatbot can answer “is the large one in stock?” at 23:00 and pass the questions that need a person to the right one, conversation attached. This guide covers what to give it, the install, a test script and how to tell whether it is working.

What shoppers ask before they buy

A month of pre-sale emails from almost any small shop falls into six groups. Stock: is it available in my size or colour. Sizing: which size if I am usually a medium. Compatibility: will this case fit my phone, will this cartridge work in my printer. Delivery: how long, how much, do you send to my country. Returns: can I send it back, and who pays. Cost: what is the total once shipping is added.

An unanswered question can end the visit. In Baymard Institute’s research on why US shoppers left a checkout without buying, 40% blamed extra costs such as shipping, tax and fees, 20% said delivery was too slow, 13% said the returns policy was not satisfactory and 12% could not see the total cost up front (Baymard Institute, page updated 22 September 2025, checked 21 October 2026). Each of those is a question a chat can answer earlier, provided the answer exists somewhere it can read. So the first job is finding where each answer lives today.

What a WooCommerce chatbot needs to read

The catalogue. Products, prices, stock and categories change daily, so they should come from the shop rather than a copy uploaded once. WooCommerce’s developer documentation describes its public Store API as “an unauthenticated API” that “does not require API keys or authentication tokens for access” (WooCommerce Store API docs, checked 21 October 2026), so a chatbot can read the same catalogue your visitors see, with no secret to store.

Your pages and files. Shipping, returns, size guides, care instructions, the FAQ. Most of this already exists as WordPress pages; some lives in PDFs or spec sheets. Text inside an image, such as a size chart saved as a JPEG, is invisible to a chatbot, so retype it as a table or a CSV.

Exact answers you write. The few facts where wording matters: the dispatch cut-off (“orders placed before 14:00 on a weekday go out the same day”), the return window, the free delivery threshold, the countries you ship to. Written once as a question and answer, they stop the chatbot paraphrasing a policy in a way you would not.

Setting it up, step by step

These steps follow EzyConn’s portal; the order (content first, install last) suits any tool.

1. 1

   Give it your website address. It finds the linked pages. Keep shipping, returns, size, FAQ and category pages; drop old posts that would only confuse it.

2. 2

   Connect the catalogue. In Channels, then WooCommerce, enter the address customers visit. "Check this address" shows what it finds without saving anything; "Connect it" saves it. The screen then shows the product count, currency and categories, and "Show some products" lists the rows the assistant reads, with price and stock.

3. 3

   Add the files that are not pages. PDF, Word, plain text, Markdown or CSV. Scanned PDFs are pictures of text, so retype the important ones.

4. 4

   Write your exact answers. Ten to twenty is plenty to start. They are used first, word for word.

5. 5

   Decide who answers when the AI cannot. Pick the people, how chats are shared (by turn, by who is least busy, or by hand), and whether requests also go to Slack. Choose whether to ask for a name and email before the chat starts.

6. 6

   Install it on WordPress. Copy the code from Chat widget, then Install, and paste it into a code snippets plugin’s footer box, or add the short wp_footer function from our WordPress and WooCommerce install guide to functions.php. Save and clear any caching plugin.

7. 7

   Confirm it is live. Back on the Install screen, run the check. It confirms the token, that a page loaded the chat, that someone opened a conversation and that it answered one.

The website chat goes live in about 3 minutes once the code is in the footer. The afternoon goes on steps 1 to 4, because that is what the chatbot answers from.

Tidy the catalogue first

• Stock on every variation. If the large size is out but the parent product says “in stock”, that is what shoppers hear. Where a shop does not publish stock, EzyConn’s product list shows “Stock not published” rather than guessing.
• Variation names a person would type. “Forest Green, 100g” rather than “FG-100”.
• Sale end dates. A sale that should have ended last week is still the catalogue price.
• Dimensions, materials and compatibility in descriptions. These answer the fit questions, and they are often missing.
• If the check fails, look at security plugins and maintenance mode first; the screen says what answered instead of a shop.

Order questions: where is my parcel?

After the sale, most messages are about an order that has not arrived. That answer is private to one customer, so it needs one of two routes.

Answer the general part and offer a person. The chatbot gives your delivery times from the shipping page, then offers your team. In EzyConn the customer sees “Thanks, I’m connecting you to a support specialist now. Please hold for a moment.”, a teammate is assigned and notified, and they reply from the inbox or the Slack thread with the whole conversation in front of them. For most small stores this is enough.

Look the order up in your own system. If order questions take hours a week, EzyConn can answer them through an AI Action connected to your own system, from the Starter plan. The agent calls it only within the rule you set, for example after the customer gives the order number and the email used at checkout.

The 15-question test script

Do not go live on three friendly questions. Run this script in the chat on your own site with real products in the brackets, read each answer as a customer would, and note what is wrong, vague or missing. Each miss points to a page to fix, a description to complete or an exact answer to write.

Watch question 4. When the description does not say, the right answer is “I don’t know” and the offer of a person, not a confident guess. EzyConn’s default reply, which you can reword, is “I don’t have that specific information right now. Would you like me to connect you with a support agent who can help?” Our guide to how accurate AI chatbots are explains why answers depend so much on what the chatbot is given.

Where the chat should appear

Added through the footer, the chat bubble appears on every page. For a store that is usually right, because questions come up in three places: product pages, where the stock, size and compatibility questions happen; the basket, where shoppers first see delivery costs; and help and shipping pages, where people arrive looking for an answer. For a contact or help page, EzyConn’s Install screen also gives you code that puts the chat inside the page at the size you choose.

Write the greeting for a shopper (“Questions about sizes, stock or delivery? Ask here.”), and if you ask for details first, keep it to an email address.

Worked example: a yarn shop’s first month

Lantern & Loom is a WooCommerce yarn shop we made up for this example: two people, about 180 products, most in several colourways. Before adding a chatbot they sorted 50 recent emails. 34 asked things the site already answered (is this colour in stock, which needles for this weight, how long to Ireland), 9 were about orders and 7 needed a person (a custom kit, a wholesale enquiry).

They connected the catalogue, kept 22 pages, added their needle chart as a CSV and wrote 14 exact answers. On the test, 11 of 15 answers were right first time. Two were vague because the dispatch cut-off appeared only in a banner image, so it became an exact answer. One missed the needle size for a new yarn whose description left it out. One, about wholesale prices, said it did not know and offered a person, which is the right result.

In the first four weeks the reports showed 610 chats: 430 handled by the AI, 125 passed to a person and 55 started by a person. “Dye lots” led the unanswered topics and became an exact answer in week two. Pre-sale emails fell from about 150 a month to about 70.

Lantern & Loom and every number in this example are invented to show the method. They are not a benchmark or a forecast for your store.

How to measure whether it is working

In a small store, “chat-attributed revenue” is hard to measure honestly in the first month. Measure what the chat changes instead:

• How conversations ended. EzyConn’s reports split them into handled by the AI, passed to a person and person-led, with first reply times and the reasons questions went unanswered.
• What it could not answer. Knowledge, then Unanswered, ranks the topics it does worst on, and when a teammate answers one of those questions the reply waits as a draft answer to approve. Fifteen minutes a week here does more than anything else on this list.
• Leads. Chats where the shopper left an email count as leads in reports, and the CSV export takes them to your mailing tool.
• Your inbox. Count pre-sale emails for the four weeks before and after.
• Orders, with care. Compare the same weeks year on year, because seasons move a small store far more than a chat window does.

Heading into a busy season? Our holiday shopping 2026 statistics show when the rush tends to start. And the same tidy descriptions help beyond your site: our guide to ChatGPT shopping for online stores explains how products reach AI shopping answers.

Where EzyConn fits

EzyConn connects to a WooCommerce shop by its address, with no password and no key, and reads the public catalogue: products, prices, stock and categories. Its AI agent answers from that catalogue together with your pages, files and exact answers, in multiple languages. When a shopper needs a person, it offers one and hands the whole conversation to your team in the inbox or Slack; “Hand back to AI” returns the chat when you are done.

The WooCommerce connection works on every plan, Free included, which needs no card and does not expire. From Starter you add AI Actions for order lookups through your own system, workflows and outbound campaigns. Our guide to EzyConn for online stores shows the rest, and every limit is on the pricing page.

The quickest way to judge it is the 15-question test on your own store. Put a free AI chatbot on your website, or start free, connect your shop and ask it what your customers ask you.

WooCommerce chatbot: common questions

Keep reading

• Install the AI chatbot on WordPress and WooCommerce
• ChatGPT shopping: what online stores need to know
• Holiday shopping 2026: the numbers for online stores
• A holiday support plan for your chatbot
• What a good hand-over from chatbot to person looks like
• EzyConn for online stores