WooCommerce & Magento · Nova Scotia, working remotely

What an AI shopping assistant sees when it looks at your store

A growing number of shoppers now ask an AI assistant — ChatGPT, Gemini, Copilot and others — to find and buy things for them, the way they used to search and browse on their own. Increasingly, the assistant can act on that request itself: comparing options, starting an order, sometimes finishing the purchase entirely.

Either way, it isn't reading your homepage. It reads a list of fields, one per product, and decides from those whether to mention you at all — and on most stores, the field that decides is blank.

I check that for WooCommerce and Magento merchants and tell you exactly what yours says. The first look is free, works from your store's own web address, and takes me about ten minutes.

Product record as read
name Industrial Degreaser 20 L
sku DG-2000-20
price $84.99 CAD
in_stock true — 7 on hand
image 1 photo
description ▸ ""  empty

Priced, stocked, photographed — and invisible. Ask an assistant for “a 20 litre industrial degreaser” and it matches on the empty field.

Composite example. Real stores are never named here.

Why this is happening now

The industry already tried full autonomy — and rebuilt around your own checkout instead

A year ago, an assistant that reads a store's catalogue and starts an order was a research pitch. It isn't anymore. Google, Shopify, Amazon, Microsoft, Salesforce and OpenAI have each shipped, or committed to shipping, some version of it in the past year — independently of each other, which is the part worth noticing. Competitors rarely agree on the same idea by accident.

~30 merchants ever completed a sale

Out of over a million Shopify merchants offered ChatGPT's in-chat checkout. OpenAI shut it down in March 2026 — Walmart had measured conversion at a third of sending shoppers to their own site.

0 official WooCommerce checkout module

WooCommerce's own site marks agent checkout "coming soon," no date given. Every implementation live today is a small, unofficial plugin.

The interesting part isn't that this is coming — it's that the industry already tried the ambitious version and backed away from it. ChatGPT's original pitch was an assistant that completes your customer's purchase without them leaving the chat. It shipped in September 2025. By March 2026 it was gone, replaced by a model where the assistant helps someone find what they want and sends them to the merchant's own checkout to actually pay. That's the model this work already builds toward: an assistant that can read your catalogue and start an order, with your customer finishing the purchase on your own site, through your own gateway, exactly as they do today.

None of this requires anything from you to be found at all. Every major assistant already reads ordinary web pages, the same way a search engine does, and always has. What changes is what it finds when it looks. A search engine shows a link and lets a person judge it. An assistant reads the record and decides for you — and a record with no price, no stock count or no description usually isn't ranked lower. It's skipped.

How much of this is actually happening? Not a huge amount yet. Traffic that reaches a store by way of an AI assistant is still a small share of most stores' total visitors, and a person still finishes the purchase far more often than an agent does. But that traffic is growing, several large companies committed to the same basic approach inside the same year, and it's a better time to get your catalogue in order before this is a large share than after.

Field check, September 2026

Ten stores, checked the same way an assistant would

Every claim on this page should be checkable, including this one. I ran the same public check over ten Nova Scotia shops that sell online — nothing private, nothing that touched their site beyond what any customer's browser does — and this is what came back.

5 of 10 stores

Report no inventory at all. An assistant reads every product as available, including the ones that aren't.

2 of 10 stores

Have products with no price attached. Anything unpriced is skipped outright — it can't be compared.

1 of 10 stores

Has not one product description in its entire catalogue. Everything else about it was in good order.

Sampled 40 products per store from data each store already publishes for its own storefront. No store is identified, then or later. This particular sweep was Nova Scotia; the check itself works on any WooCommerce or Magento store, anywhere.

The check

What I look at

01

Can an assistant find your products at all?

Whether your catalogue is published in a form software can read, and whether what it publishes matches what's on your shelves.

02

Is anything missing from the record?

Descriptions, prices, stock counts, photos, product codes. Every gap is a question an assistant can't answer about you, so it moves on.

03

Do your totals come out right?

Shipping and tax, checked against a figure worked out independently. Totals that are quietly wrong are the expensive kind — nothing looks broken.

04

Would a customer trust what they find?

Returns policy, contact details, delivery terms. Assistants surface these, and a shop that can't produce them looks riskier than one that can.

05

Can it actually complete a purchase?

Whether your store can take an order placed through an assistant, or whether it falls over partway. Payment always finishes with your customer, on your own checkout.

How this works

Start with the free one

No charge

The ten-minute look

I read what your storefront already publishes and write back with what I found. That's it — no meeting, no call booked, no obligation.

  • Reads only public pages, same as any shopper
  • Changes nothing and needs no access
  • A plain-English note, usually a page

The full audit

If the free look turns up something worth pursuing: a proper report on a copy of your site, with a remediation list in priority order and a price against each item.

  • Runs on a staging copy, never your live shop
  • Needs your written permission first, in writing, naming the site
  • Ends in a quote you're free to ignore

The audit is a fixed price — ask and I'll tell you straight away. What a fix costs is the part I won't guess at, because it depends entirely on what the report turns up, so that quote comes out of the findings rather than a tier.

If it's worth fixing, fixing it is a separate piece of work — installing and wiring up a real, working setup on your own store, checked end to end rather than just written about. That's priced separately too, and you'd see exactly what it involves before agreeing to it, the same as the audit.

Straight answers

Who you'd be dealing with

I'm a student at Dalhousie. I built the tooling behind this myself — a working implementation of the protocol these assistants use to read and buy from a shop, plus the scanner that checks a store against it.

I don't have a list of past clients to show you. I'm early, and I'd rather say so than pad a page with logos. What I'm after is a small number of first engagements at an honest rate, in exchange for being allowed to describe the work afterwards.

What I won't tell you is that you're losing sales. I can't know that, and neither can anyone who emails you claiming it. I can tell you precisely what your store publishes about itself, because I can read it — and you can check every word of it yourself.

If this turns out not to be worth your time, that's a perfectly good outcome. A store with no inventory system and eleven products does not need me.

UCP status development build
plugin 0.9.0
spec 2026-08-25
capabilities catalogue search, checkout, shipping, orders
requests signed and verified, every endpoint
signing key EdDSA · GXT-yXZw_UAduHS64Oc5QIoehE8w2wtlHOJJhTfSXek

Reproduced here from the real admin screen on the development build — the layout is redrawn, the values are not. This is the page that proves a setup is actually working, on this store or a client's.

Get in touch

Ask for the free look

Send me your shop's web address. You'll get a reply with what I found, and nothing else — I don't run a mailing list, and I won't add you to one.

Name
Harper Rouse
Based
Nova Scotia — working remotely with stores anywhere

If I wrote to you first and you'd rather I didn't again, reply to the address above and say so — that's the end of it, and you never have to ask twice.