E-commerce · · 7 min read

Can AI handle my online store's descriptions, support and returns?

Small online shops use AI to write listings, answer where-is-my-order emails and start returns. What each job costs, what one brand got, and what to keep.

An isometric clay model of a small home-based online shop, with a packing table of folded sweaters, stacked shipping boxes, a closed laptop and a small delivery van, on a black background.

Black Friday ended, and the support queue did not. Pepper, an online bra brand with about 50 employees, came out of one holiday season with a backlog of roughly 400 tickets a day, and it took until the following spring to clear. Most of those messages asked the same few things: where is my order, which size do I need, how do I send this back.

That is the work AI has landed on in small online shops. Not strategy, not design: the repeated typing. Across Canada, Statistics Canada found that 19.2% of businesses used AI to produce goods or deliver services in the year to mid-2026, up from 6.1% two years earlier. Among the ones using it, 28.2% used virtual agents or chatbots, the third most common use.

19.2%

of Canadian businesses used AI in the past year

Statistics Canada, 2026

28.2%

of those used it for chatbots or virtual agents

Statistics Canada, 2026

54%

of Pepper's support conversations handled by AI on average

Gorgias

We picked the five jobs below because each one is either built into tools small stores already use, or shows up in a named, reported example. A word on that example: Pepper's numbers come from a case study published by its software vendor, Gorgias. Vendors publish their best customers, so read the figures as what is possible, not what is typical.

1. Writing the first draft of product descriptions

Verdict: worth doing, as long as it stays a draft.

Writing 40 listings by hand is the job most owners put off. Shopify now does a first pass inside the product page. You type a few details about the product, pick a tone of voice or describe your own, and Shopify Magic writes a description. You can generate several versions and edit before you publish.

It has limits worth knowing. Shopify's own listing says the feature covers descriptions and blog posts, but not titles, SEO fields, image alt text, tags, variants, collection descriptions or bulk editing, and it does not translate. So it helps with the blank page, not with the whole catalogue.

The bigger limit is that it only knows what you tell it. Give it "wool sweater, navy" and it will fill the gap with something plausible: a fabric blend, a fit, a care instruction. If that detail is wrong, the customer finds out when the parcel arrives, and the return lands on you. Feed it the facts you have, and delete any claim you cannot back up.

Cost: included with Shopify plans at no extra charge. A general chatbot does a similar job if you paste in your notes.

2. Answering "where is my order?"

Verdict: the safest win, because the answer is already in your system.

An isometric clay model of a small support desk with an open laptop showing blank chat bubbles, a phone, headphones and a shelf of parcels waiting to ship.

Order-status questions have one right answer, and it lives in your order and shipping data. That makes them the best fit for an AI agent, which is software that reads the customer's message, looks up the order and writes the reply.

This is where Pepper started. Its old helpdesk had no automation, and, as its CX operations lead Gabrielle McWhirter put it, "our queue got really messy." After moving to Gorgias's AI Agent, which the team named Penelope, Pepper reports that it handles 54% of conversations on average and about 60% at peak periods like Black Friday. Order tracking and finding lost order numbers are among the questions it answers on its own, including order-status questions written in German.

The catch is the word "handles". A chat widget that cannot see your orders can only point people to an FAQ page. What you are paying for is the connection to your store's data.

Cost: Gorgias charges US$0.90 for each AI Agent interaction on an annual plan, or US$1.00 on a monthly one, on top of a helpdesk plan. Its Starter plan is US$10 a month for 50 tickets; Basic is US$60 a month for 300. At a dollar a conversation, check what one answered email is worth to you before you switch it on for everything.

3. Answering sizing and product questions

Verdict: useful before the sale, but only as good as your product information.

The question before "where is my order" is "will this fit?" Pepper sells bras for petite chests, where fit is the whole product, and its AI agent answers questions about sizing, materials, styles and whether something is in stock, then suggests a matching item.

McWhirter's line on it is the honest one: "Penelope takes the information we give her and responds better than a Macro." A macro is a canned reply your team pastes in. The AI does better than that because it can combine the facts you gave it to fit the question. It cannot know anything you did not give it.

The AI cannot be more accurate than the product page it reads.

So this job has a hidden first step: a sizing chart, materials and care details written down somewhere the tool can read. If your best fit advice lives in one employee's head, write it down first. That document is worth having even if you never buy the AI.

Cost: the same per-conversation pricing as item 2 if you use the same tool. The real cost is the afternoon spent writing up your product knowledge.

4. Starting returns and exchanges

Verdict: let it start a return; do not let it make up policy.

An isometric clay model of a returns station, with an opened shipping box holding a folded violet sweater, a blank mailing bag and a shelf of three empty sorting bins.

Returns are the conversation customers dread most, and much of it is routine: is this item still within the window, can I swap for another size, how do I get a label. Pepper's AI handles returns, exchanges and order changes. For a small shop, taking those steps off a person's plate is real time back.

The risk is when the bot answers a question your policy does not cover, or gets the policy wrong. In Moffatt v. Air Canada, the airline's website chatbot told a grieving customer he could apply for a bereavement fare after travelling, which another page on Air Canada's own site contradicted. A British Columbia tribunal found Air Canada responsible and rejected its argument that the chatbot was a separate entity. In the tribunal's words, "it should be obvious to Air Canada that it is responsible for all the information on its website."

That was a small-claims tribunal, not a court, but the lesson travels. Whatever your bot says about refunds is your store saying it. Before you switch it on, give it your written return policy, test it with the awkward cases (worn items, final sale, a gift from someone else), and tell it to hand anything outside the policy to a person.

Cost: part of the same support tool, at the same rate per conversation.

5. Refunds, exceptions and upset customers

Verdict: keep this one. It is where your shop earns its reputation.

Every job above has a clear answer the software can look up. This one does not. A damaged parcel before a wedding, a customer who has bought from you for six years, a complaint that is about to become a one-star review: these need someone who can bend a rule and mean it.

Pepper says as much. "We are not threatening those really personal interactions that make our brand Pepper," McWhirter said. The point of handing off the first four jobs is to give a person time for this one, not to remove the person.

There is also a privacy reason to keep the AI's reach narrow. A support agent connected to your store sees names, addresses, phone numbers and order histories. Under PIPEDA, the federal private-sector privacy law, that is personal information, and accountability for it is the first of the law's ten principles. Read where the vendor stores the data and whether it trains its models on your conversations before you connect it. Privacy and cybersecurity concerns were the top barrier to AI that Canadian businesses reported to Statistics Canada, at 13.4%, ahead of cost.

Cost: your time, which is the thing the other four jobs free up.

When it is not worth it

If your inbox gets ten emails a day, a helpdesk with per-conversation AI pricing is probably more tool than you need. Shopify's built-in description writer and a set of well-written saved replies will do most of the job for nothing. The per-conversation model starts to make sense when the same five questions arrive dozens of times a day, or when a sale week buries you the way it buried Pepper.

The five jobs side by side

JobExample toolCostMain risk
Product descriptionsShopify MagicIncluded with ShopifyInvented product details
Where is my orderGorgias AI AgentUS$0.90–1.00 per interaction, plus a planPaying for answers a saved reply could give
Sizing and product questionsGorgias AI AgentSameThin product information
Starting returnsYour helpdesk's AISameWrong or made-up policy
Refunds and exceptionsA personYour timeHanding it to software

Where to start

  1. Count your last month of support emails by type. If order status and returns make up most of them, items 2 and 4 are where AI will pay; if not, it may not.
  2. Write down your return policy, sizing guide and materials in one place. Every tool above works better with them, and you can use them as saved replies today.
  3. Try the free description writer on your next five listings, and check every fact against the product before you publish.

If you want a second opinion on which of these fits your store, you can book a call with Trimotech.

Sources

  1. Analysis on artificial intelligence use by businesses in Canada, second quarter of 2026Statistics Canada, June 11, 2026
  2. How Pepper's AI Agent automates 54% of support and converts 19% of conversationsGorgias, March 5, 2025
  3. Shopify Magic for product descriptionsShopify
  4. Gorgias pricingGorgias
  5. Moffatt v. Air Canada: A misrepresentation by an AI chatbotMcCarthy Tétrault, February 19, 2024
  6. PIPEDA in briefOffice of the Privacy Commissioner of Canada

Questions

Will AI-written product descriptions hurt my store?
Not if you treat them as a first draft. The risk is a description that states a material, size or care instruction you never gave it, so check every fact against the product before you publish.
How much does an AI support agent cost a small store?
Gorgias, one helpdesk built for online stores, charges US$0.90 to US$1.00 for each conversation its AI handles, on top of a plan that starts at US$10 a month. Free chatbots exist, but most cannot see your orders, which is what makes the answers useful.
Am I responsible if my store's chatbot gets the return policy wrong?
Assume you are. In Moffatt v. Air Canada, a British Columbia tribunal held the airline responsible for what its website chatbot told a customer, and rejected the argument that the chatbot was a separate entity.

Read next

  1. AI news, October 12 · · 7 min read

    This week in AI: GPT-6, a cheaper Haiku and call recording rules

    ChatGPT moved to GPT-6, Anthropic cut the price of its small model by up to 90%, and Canada's privacy watchdog updated its rules for recording customer calls.

  2. Online reviews · · 7 min read

    How to reply to Google reviews with AI, and when not to

    Customers expect a reply to every review, and fast. How to let AI draft the routine ones, check each one before it posts, and which to answer yourself.

  3. AI news, October 5 · · 7 min read

    This week in AI: Meta's Muse, Shopify Canvas and StatCan's numbers

    Meta opened its Muse agent to small businesses in Canada, Shopify let its AI rewrite store themes, and StatCan says one in four firms now plans to use AI.