ChatGPT Is Sending Shoppers to Stores. Is Yours Ready?

There’s a version of this article that opens with a scary number about how many people shop through AI assistants now. I don’t have a number I trust, so here’s the part that’s easier to verify: open your analytics, look at referral sources, and search for chatgpt.com.

If you see it, that’s a shopper who asked a question, got told your store had what they wanted, and clicked. If you don’t, that’s worth knowing too.

Either way, the discovery layer merchants have optimized for since forever now has a second version running alongside it. FullSweep SEO exists partly because that second version rewards a different kind of writing than the first one does, and most catalogs were written for the first.

What actually changed this year?

The short version: assistants got out of the checkout business and got serious about recommendation.

OpenAI launched in-chat purchasing in late 2025, then stepped back from it in early 2026 and rebuilt the shopping experience around discovery. Shoppers browse products visually, compare options side by side, narrow things down by talking, and then complete the purchase on the merchant’s own site. Google and Shopify have their own competing standard for the same handoff.

For you, the practical effect is a shift in where the decision happens. The comparison, the shortlist, the “which of these actually fits what I described” reasoning: that all runs before the shopper ever loads your page. By the time they arrive, they’ve been recommended. They’re not browsing, they’re checking.

Which is a good problem to have, assuming you’re in the shortlist.

How do I check whether this is already happening?

Three things worth looking at, and none of them take long.

Referral traffic first. In GA4, or Shopify’s own reports, look for chatgpt.com, perplexity.ai, copilot.microsoft.com and similar. Volume will probably look unimpressive. Look at behavior instead: session length, pages per session, conversion rate. Assistant referrals tend to arrive further down the funnel than search traffic, because someone already did the filtering.

Then ask the assistants directly. Not “tell me about my store,” which produces flattery. Ask the question a customer would ask. Best waterproof duffel under $150. Ceramic planters that won’t crack outdoors in winter. Whatever your category actually is. See who gets named, and read the reasons given, because the reasons tell you which attributes are doing the work.

Third, check whether you’re even reachable. If your robots.txt blocks AI crawlers, or a theme customization broke your structured data, you can be doing everything else right and still be invisible.

Can I submit my products directly?

Yes, and this is the bit merchants tend to miss because it arrived quietly.

OpenAI runs a merchant program that accepts product feeds, so price, availability and specs come from you in a structured format rather than being scraped and guessed at. Direct submission through OpenAI’s merchant portal is currently open to US merchants.

If you’re on Shopify, though, you’re largely already there. Shopify added a sales channel called Agentic Storefronts earlier this year, which syndicates your catalog to ChatGPT, Google AI Mode, Gemini, Microsoft Copilot and Meta — no separate application, no extra apps to install. You can see the feed itself from the channel, and orders that come back through those platforms land in your Shopify admin attributed to whichever one sent them. That last part matters more than it sounds: it turns AI-driven sales into something you can actually read in a report, rather than something you infer from referral logs.

Worth ten minutes in your admin either way. It also addresses a specific failure mode, where an assistant confidently recommends a product you discontinued in March — a current feed is what keeps price and availability honest.

Just don’t mistake it for the whole job. A feed tells a system what you sell. It doesn’t tell it why your product suits someone who described a problem in a sentence and a half.

So why doesn’t the feed alone get you recommended?

Because the recommendation is a piece of writing, and writing needs source material.

When an assistant explains why a product fits, it’s synthesizing from pages it read. Your product page. Your FAQ. Reviews. A roundup someone published two years ago. OpenAI has said its shopping model is trained to read trusted sites and cite reliable sources, which is a polite way of saying the quality of your pages determines whether you’re usable.

Here’s where most catalogs fall down. A shopper asks for something that “packs down small and won’t smell after a week.” Somewhere in your store, a customer review answers that. Your product page says “engineered for the modern traveler.”

One of those is retrievable. The other is wallpaper.

What does a recommendable page look like?

It’s less exotic than the phrase suggests. Mostly it means answering the questions you already get asked, in the words people ask them in, on the page itself.

Specifics beat adjectives. Weight, dimensions, materials, capacity, what it fits, what it doesn’t. If a constraint can appear in someone’s question, it should appear as a fact on your page.

Say what it’s not for. Stores are weirdly reluctant about this, and it’s one of the strongest signals available. “Not suitable for sustained rain” gets you excluded from a query you’d lose anyway, and included in one where accuracy is the deciding factor.

Answer questions near headings that ask them. An assistant lifting a heading plus the paragraph beneath it is doing the same thing a featured snippet does.

Keep your facts consistent everywhere. Product page, FAQ, marketplace listing, feed. When those disagree, something else gets trusted instead.

What’s worth doing this week?

Pick your top twenty products by traffic. Read the descriptions as though you’re a system trying to answer a constraint-heavy question, and note every place where you’ve written a feeling instead of a fact. That’s your list.

Then check your feed situation. On Shopify that’s mostly turning on Agentic Storefronts and confirming your catalog data is accurate, rather than building anything.

The awkward part is what happens after twenty products. The same rewrite has to reach the other several hundred, plus the collection pages and the static pages that answer broader questions, and that’s where merchants stall out. This is the gap FullSweep SEO was built for: whole-store coverage with bulk processing, optional AEO structuring for the answer-engine side, and a before/after diff on every change so you approve, edit or skip each one rather than handing over your catalog. There’s more on that at how FullSweep SEO handles AEO.

None of this is a growth hack, and it won’t produce a chart that goes up next week. What it produces is a store that reads clearly to anything trying to answer a question about what you sell. That was true when the thing asking was a person with ten tabs open. It’s more true now that it’s software with an opinion and a shortlist of three.