That test takes 30 seconds. Most store owners I talk to have never run it.

Here is why it matters now.

On June 17, Shopify turned it on for everyone

Shopify’s Summer 2026 Editions shipped more than 150 updates under one theme: sell, shop and build everywhere. The center of it is agentic commerce, and the important detail is not the feature list. It is the default.

Agentic commerce arrived turned ON. Eligible products auto-syndicate to ChatGPT, Microsoft Copilot, Google AI Mode, Gemini and the Shop app. A shopper can find your product inside an assistant, compare it, and buy it, without ever loading your site.

This is not one platform experimenting. The plumbing underneath is UCP, the Universal Commerce Protocol, built with Google and endorsed by Amazon, Meta, Microsoft, Salesforce, Stripe, Etsy, Target and Wayfair. When a list that wide agrees on one protocol, it stops being a bet.

Nobody sent you an email about it. There was no kickoff meeting and no budget approval. It happened to your store while you were working on something else.

So the question is no longer “should we do something with AI.” The question is “what is the assistant saying about us right now, and did we have any say in it.”

Your product page now has a second reader, and it has no eyes

Think about what actually happens in that flow.

The assistant reads your catalog through the Shopify Catalog. It compares. It picks. It can carry the buyer through cart and checkout using Shopify’s public Cart and Checkout MCP servers.

Now count everything that reader never sees:

• The hero video you paid for
• The trust badges at the top of the page
• The sticky add to cart on mobile
• The cart drawer with the free shipping bar
• The upsell block on the product page
• The theme you paid to customize

None of it exists in that path. The whole conversion layer we have been optimizing for years is built for a human eye, and this reader does not have one.

What it does read is boring, and that is the good news:

• The product title
• Product type and tags
• Variant names
• Structured attributes: size, material, color, compatibility
• Price and stock status
• Shipping and return terms

That is your storefront in this channel. It is text, it already sits in your admin, and it is probably half filled in.

What a good product title looks like to a machine

Here is a real pattern I see in almost every catalog we take over.

The title reads:

ARIA | 2026 Collection

That title was written for a person who is already on your site, already in the sofa category, already looking at a photo. In that context it works fine. The picture carries everything the words leave out.

To an assistant, that title says almost nothing. It cannot tell what the product is. It cannot tell who it is for. It has no attribute to match a request against. A buyer asking for a three seat sofa in washable fabric under 220cm will never be shown this product, because nothing in the data says it qualifies.

Now the same product, written for both readers:

ARIA 3 Seat Sofa, Washable Fabric, 210cm, Grey

Nothing was lost. The human still sees the photo and the collection name is still there in the description. But the machine can now match this product to a real request, and it can compare it to a competitor on the exact attributes the buyer asked about.

That is the whole job. What it is, who it is for, and the one attribute a buyer filters on.

Look at your ten best sellers and ask which version yours look like. In my experience it is the first one, and it takes an afternoon to fix.

The Hebrew problem nobody is talking about

This part is specific to us, and it is the reason I would not just copy an American checklist and hand it to a client.

Israeli catalogs are usually written one of three ways. Fully Hebrew. Hebrew descriptions with English product names. Or a mix that changed hands three times and never settled.

Meanwhile your buyer asks the assistant in whichever language they happen to be thinking in, and for a lot of Israeli buyers researching a purchase, that is both.

Three things follow.

Your structured fields need to hold the attribute, not the description. A sofa whose width appears only inside a Hebrew paragraph in the description is not a 210cm sofa as far as the machine is concerned. It is a paragraph. Put the number in the field.

Transliteration inconsistency quietly costs you matches. If one product says ספה, another says סופה, and a third says sofa, you have split your own catalog into three catalogs that do not know about each other. Pick a spelling per term and enforce it across the store. This is unglamorous work and it pays.

Do not assume your Hebrew store is invisible in English queries, or the reverse. Test both. Ask the assistant in Hebrew, then ask the identical question in English, and compare what comes back. If your store appears in one and not the other, you just found a gap that most of your competitors have not even looked for.

The traffic is small. It is also the best traffic you have.

In Q1 2026, AI-driven traffic to Shopify stores grew 8x year over year. Orders from AI-powered search grew nearly 13x.

Read those two numbers next to each other. Orders grew faster than sessions did.

That gap is the whole story. Sessions growing 8x while orders grow 13x means the buyers arriving through this channel are already close to a decision. The assistant did the filtering before the click. It narrowed the category, matched the requirements, ruled out what did not fit, and sent one person who is ready.

If you buy media, you already know this argument. You have paid a premium for high intent traffic for years. This is the same trade, except right now nobody is bidding.

I know what the GA4 report looks like. A few dozen sessions from ChatGPT, sitting under “Other,” easy to ignore. Ignoring it is a decision about next year, not about this month.

How to actually see it in your own numbers

Before you decide this channel is too small to care about, go measure it properly. Most stores never have.

In GA4, build one segment from referral source, and include the assistants by hostname: chatgpt.com, perplexity.ai, copilot.microsoft.com, gemini.google.com. Add the Shop app as its own line.

Then look at three things, not one:

1. Sessions, so you know the size.
2. Conversion rate, compared against your organic search number. This is the one that will surprise you.
3. Average order value, compared against the same.

Sessions alone will tell you to ignore this. Sessions next to conversion rate will tell you something else. Take the screenshot and keep it, because in six months you will want the baseline.

You can buy the shelf. You cannot buy the answer.

This is the part that took me longest to sit with.

Start with what you can buy, because it is real. OpenAI launched advertising inside ChatGPT on February 9, 2026. Shopping questions return a sponsored product carousel under the response, labeled Sponsored, with logo, price, stock status and delivery estimate. Research from OtterlyAI put sponsored placement on roughly 76% of shopping related questions, the highest rate of any query type. Shopify and Etsy are confirmed partners for that format. So there is inventory, there is an auction, and it behaves like the shopping channel you already run.

Now look at where it sits. The carousel is under the response. The response itself, the paragraph where the assistant compares three products and recommends one, is not for sale. OpenAI states plainly that ads do not influence how it answers.

So the channel splits in two, and the two halves need different budgets:

• The sponsored row. An auction. Same game as Google Shopping, same skills, same money.
• The recommendation itself. No position 1. No ad slot. Three products appear side by side, compared on the attributes the merchant supplied.

On Google you can outbid your way to the top of the page. Here you can only outbid your way into the row underneath the part people actually read.

The only lever on the answer itself is whether your own data makes you the easy pick.

And you never get to watch the comparison happen. You cannot A/B test it. You will not see the moment your product lost because a competitor’s title said “waterproof to 50m” and yours said “great for swimming.”

That is uncomfortable. It is also fixable, and it costs one afternoon.

What this does not mean

It does not mean your store design stopped mattering.

Most of your revenue still comes from humans who land on your site, look at photos, read reviews, hesitate at the cart, and need a reason to trust you at checkout. That work is not going anywhere, and anyone telling you to redirect your CRO budget into catalog fields is overreacting.

The honest framing is smaller and more useful: you now have two audiences with completely different requirements, and until this month you were only serving one of them. The second one is cheap to serve, because the work is data entry on products you already sell.

What to do this week

No developer needed for any of this.

1. Run the test from the top of this post. Ask an assistant for what you sell, phrased three different ways, the way a real buyer would say it. Do it once in Hebrew and once in English. Write down what came back and who else showed up.
2. Open your 10 best sellers and read the titles the way a machine would. What it is, who it is for, the one attribute a buyer filters on. If the title fails that test, rewrite it. Not for SEO. For a reader that only gets the text.
3. Fill in the structured fields you skipped. Material, dimensions, compatibility, care, warranty. Every empty field is a filter you cannot pass. Fields beat descriptions every time.
4. Put shipping and return terms in text, not inside an image and not behind a link buried in the footer. If your return window is genuinely good, it is a selling point the assistant can quote.
5. Check that stock status is accurate. An assistant will not recommend what it believes is out of stock, and a stale inventory sync is now costing you recommendations, not just orders.
6. Build the GA4 segment and screenshot the baseline.

Do those six and you are ahead of most stores in your category.

The bottom line

Your store is already in the room. It was signed up for you on June 17, without your signature.

The only open question is whether your catalog speaks well enough to get picked.

Go read your top 10 product titles. Today. That is the whole assignment.

If you want a second pair of eyes on the catalog before this turns into a real revenue line, that is the kind of work we do at Shopping Dev Solutions.