CRO · AI Commerce · June 2026
Getting recommended by ChatGPT is the easy part
AI shoppers now convert at nearly double the rate of everyone else, and Shopify just turned on agentic storefronts for every store. So I looked at 1,009 live Shopify stores to see what happens after the AI sends a buyer over. Most of them drop the ball.
Last month a shopper asked ChatGPT for "a magnetic phone case that actually holds on a bike mount." It named three brands, linked them, and offered to check out right there. No URL, no Google result, no scrolling a collection page. One of those stores got a warm, high-intent buyer handed to it for free.
Then the buyer landed on the product page and the Add to Cart button was below the fold. On a phone. They scrolled, got distracted, left.
That second part is the one nobody's talking about. Everyone's busy trying to get the AI to recommend them. Almost no one is ready for what happens after it does.
The front door of shopping is moving
This isn't a prediction, it already happened. Shopify turned on agentic storefronts by default for every store, which means ChatGPT, Copilot, Gemini, and Google's AI Mode can read and recommend your products without you doing a thing. Shopify says AI-attributed orders grew 11x in a year. About a third of US shoppers say they've used ChatGPT somewhere on the way to a purchase.
The part that actually matters: AI-referred shoppers convert at close to double everyone else. They show up further down the funnel. The AI already did the comparing and filtering, so by the time they hit your page they're mostly decided.
Which is exactly why a leaky store hurts more now. A wasted visitor from a cold ad is annoying. A wasted visitor the AI pre-qualified and sent you ready to buy is expensive.
So I looked at 1,009 Shopify stores
I pulled up 1,009 live Shopify storefronts on a phone, the way a real shopper arrives, and looked at what an AI-sent buyer would actually hit. Not a survey. The real homepages and product pages. It was rougher than I expected.
The review one should worry people most. Reviews are the single signal AI assistants and human shoppers both lean on hardest. No reviews makes you harder to recommend and harder to trust once you are. You lose on both ends.
And here's the part that stings: more than half of these stores run the Meta pixel, nearly half run Klaviyo. They pay real money to get traffic and re-market to it, then hand it a store that drops the ball on arrival. The acquisition machine is humming. The conversion floor has a hole in it.
GEO and CRO turned out to be the same job
There's a whole industry forming around getting your products into AI answers. Generative engine optimization, answer engine optimization, pick your acronym. A lot of it is useful: clean product titles, structured data, reviews, content that answers real questions.
Notice anything about that list? It's the same list that makes a store convert. Clear titles help the model understand you and help a human decide. Reviews get you cited and get you bought. A fast, readable mobile page is easier for a crawler to parse and easier for a person to act on. You're not doing two projects. You're doing one, and getting paid twice.
The recommendation is getting easier every month, because the platforms want it to. The conversion is the part still on you. And from what I saw, it's the part most stores quietly neglect.
How ChatGPT actually picks the products it recommends
"Get recommended by AI" sounds like a black box, so let's open it. Your products reach an assistant through two doors, and they work differently.
The first door is the feed. Shopify syndicates your product data to AI assistants through its agentic storefronts, and that data is what powers the shopping results inside ChatGPT. Title, price, availability, variants, images, shipping details. The assistant never admires your theme for this part. It reads fields. If your feed says a bestseller is sold out because nobody cleaned up dead variants, you lost the recommendation before taste ever entered into it.
The second door is the crawl. OpenAI's own crawler documentation lists three bots with three different jobs: OAI-SearchBot indexes pages so ChatGPT search can cite them, ChatGPT-User fetches a live page when a shopper asks about it mid-conversation, and GPTBot gathers training data. Perplexity and Google run their own equivalents. So when an assistant compares your magnetic phone case against two competitors, it is often reading your actual product page in real time, plus whatever the rest of the internet says about you.
That last clause is the part merchants underestimate. Assistants triangulate. A case that shows up in a Reddit thread, a cycling-gear roundup, and a review site wins over one that only exists on its own domain, because the model has three independent reasons to trust it. No markup fakes that. You earn it the slow way, with a product worth mentioning and a few people who mentioned it.
You can nudge it, though. Send the product to the two or three publications that write roundups in your niche. Answer questions in the communities where your buyers already hang out, as yourself, without the sales pitch. Get your best customers to review you somewhere other than your own site. None of this is new advice. What's new is that a language model now reads all of it before deciding whether you exist.
Once the assistant has candidates, the conversation decides the winner. The shopper didn't ask for "a phone case." They asked for one that holds on a bike mount, or survives a toddler, or ships before Friday. The assistant matches those constraints against whatever evidence it can find, which means every concrete attribute in your copy is a hook it can catch on. "Stays put on cobblestones, drop-tested to two meters, ships same day from Ohio" gives it three hooks. "Premium quality you'll love" gives it nothing.
It's also worth knowing the engines shop in different places. ChatGPT leans on the product feeds platforms hand it, Shopify's included, plus its own search index. Perplexity ranks from its own crawl, so PerplexityBot has to be able to reach you. Google's AI answers draw on the Shopping Graph, which is fed by Merchant Center, and on Shopify that pipe is the Google & YouTube channel. Same hygiene, three doors. Open all three and stop thinking about it.
Two Shopify-specific crawl killers deserve a call-out, because both are invisible from your desk. A password-enabled storefront, the setting stores flip on during redesigns and forget, blocks every crawler and every assistant at once. And themes that render key content client-side, like a spec table living in a tab an app builds with JavaScript, can look complete to you and empty to a bot. When in doubt, view the page source. If the words aren't in the HTML, don't count on a machine reading them.
Structured data is the piece you control this week. Complete Product markup, with price, availability, rating, and review count filled in, is how a crawler confirms what your page claims. When we audited 926 Shopify stores for AI search readiness, most shipped markup that was present but half empty. That post walks through the exact fields and fixes, so I won't repeat them here.
Then there is llms.txt, a proposed convention where a plain-text file hands AI systems a map of your most useful pages. No major assistant has publicly committed to reading it. Treat it as a ten-minute bet, not a fix.
One Shopify-specific trap before we move on. Many review and widget apps inject their content with JavaScript after the page loads. A shopper sees 340 reviews. A crawler that never runs that script sees zero. If your reviews live only inside an app's widget and never make it into the page HTML or your structured data, then as far as an AI assistant is concerned, they don't exist.
Preparing your Shopify store for AI shopping: an 8-step checklist
Here's the order I'd do the work in. Most of it needs no developer, and none of it needs more than an afternoon.
- Check robots.txt isn't turning assistants away. Open yourstore.com/robots.txt and look for GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, or Google-Extended under a Disallow line. Shopify's default file doesn't block them, but customized robots.txt.liquid files and aggressive bot rules in Cloudflare or a firewall app sometimes do. One nuance worth knowing: Google-Extended only controls whether Google may train models on your content. It doesn't decide whether you appear in AI Overviews, which run on regular Googlebot.
- Confirm your agentic storefront is on. Shopify turned agentic storefronts on by default for eligible stores, and eligibility depends on meeting basic product data standards. Check the agentic sales channel settings in your admin, make sure nobody opted the store out during setup, and fix whatever the eligibility check flags.
- Clean the product feed like a stranger will read it. Because one will. Titles that say what the product is ("Magnetic bike-mount phone case, iPhone 15") beat clever ones ("The Voyager"). Accurate stock status. Variant names a human could parse without context. Barcodes and GTINs filled in where you have them. Inside an assistant, this data is your storefront.
- Complete your product structured data. Price, availability, ratings, review count, brand, identifiers. Most themes emit the skeleton and skip the flesh. A structured data app can patch it, or a developer can fix the theme once. The 926-store audit linked above covers the field-by-field detail.
- Get reviews where crawlers can read them. Run a review app, and confirm the reviews render into the page HTML or your markup, not just a JavaScript widget. Reviews are the one signal that helps you twice, once when the assistant decides whom to recommend and again when the human decides whether to buy.
- Rewrite descriptions to answer real buyer questions. Assistants answer questions, so a page that already contains the answer is quotable. Sizing, materials, compatibility, shipping time, what's in the box, how returns work. If customers email you the same three questions, those answers belong on the product page, plainly worded.
- Add an llms.txt file. Optional, unproven, cheap. A short plain-text index of your key pages and policies at yourstore.com/llms.txt. If assistants start reading it, you were early. If they never do, you lost ten minutes.
- Fix mobile speed last, but fix it. A slow page hurts the crawl and hurts the human the assistant sends over. In our 1,009-store audit, 37% of stores loaded a heavy, slow homepage on mobile. Trim the hero images first. It's usually them.
Every step on this list is ordinary catalog hygiene. The new part is who's reading.
When you're done, test it the way a shopper would. Ask ChatGPT and Perplexity for the thing you sell ("best magnetic phone case for bike mounts") and see who comes up. Then ask about your own store by name and read the answer closely. Wrong prices, dead products, and stale policies in that answer point at exactly which source is dirty: the feed, the page, or the markup. Fix the source, wait for a recrawl, ask again. That loop is the whole game.
Set your expectations by channel, too. Feed fixes tend to show up fastest, since the data syncs on a schedule. Page fixes wait on a recrawl, which can take days or weeks. Earned mentions take months and compound for years. Do them in that order and you'll see movement while the slow bets mature.
Agentic checkout: when the AI buys instead of recommends
Everything above assumes the assistant hands the shopper to your store. The next phase removes that step, and it's further along than most merchants think.
In September 2025, OpenAI and Stripe launched Instant Checkout, which completes a purchase inside ChatGPT itself. It opened with US Etsy sellers, and OpenAI announced more than a million Shopify merchants would follow. Under the hood sits the Agentic Commerce Protocol, an open standard the two companies published so any merchant or platform can plug in. Single-item purchases came first, with carts and more regions on the roadmap.
Shopify's own agentic storefront takes a gentler path for now. The shopper finds you in ChatGPT and completes the purchase through your own checkout in an in-app browser, so you keep your checkout, your upsells, and your customer record. But the direction of travel is plain. The platforms are competing to remove steps between "I want this" and "it's ordered," and your product page is one of the steps.
Three things change for a Shopify merchant when that happens.
The feed stops being metadata and becomes the storefront. If the assistant assembles the buying experience from your product data, a lazy title or a missing shipping policy is a broken shop window. All the hygiene work above doubles as your agentic storefront build-out.
Differentiation moves upstream. When the shopper never browses, you can't win with a prettier page at the last moment. You win by being the product the assistant picks, which comes back to reviews, third-party mentions, specific answers, and a fair price.
And the customer relationship needs guarding. OpenAI's launch messaging promised that sellers keep control of their payments, systems, and customer relationships, and Stripe's announcement spells out that merchants still accept or decline each order, charge the payment method, remit sales tax, and handle fulfillment and returns as they normally would. Good. Hold every platform to that. Confirmation emails, post-purchase flows, and support should still run through you, or you're renting your own customers back.
Plenty stays the same, too. The AI can compress discovery and checkout, but it can't ship the order faster, write the apology email when the courier loses a package, or make the product good. Retention still happens in your inbox and your packaging. If anything, those matter more now, because when the first purchase happens inside a chat window, the unboxing might be the first time the customer really meets your brand.
Where does it go from here? OpenAI has said multi-item carts and more regions are on the roadmap, Google and Microsoft are building their own agentic paths, and each platform will want your catalog in a slightly different shape. That's the case for keeping product data clean at the source instead of patching it per channel. One honest catalog, many readers.
What should you do today? Mostly the checklist above, and nothing dramatic. Don't buy an "AI checkout" app that promises to handle all of this for you; the protocols are young and the platforms keep changing the rules. Do watch your orders for AI-referred sources, so you notice when the channel becomes real for your store. And when you start changing pages, prices, or copy in response to AI traffic, run it as an experiment instead of a hunch. We wrote about who should hold the final say when AI starts driving your A/B tests, and that question gets sharper once an assistant sits inside your funnel.
How to track AI traffic to your Shopify store
You can't manage a channel you can't see, and right now most merchants have AI-referred visitors filed under Direct or lumped into Referral with everything else. Separating them takes about twenty minutes.
Start with the referrer strings. Visits from ChatGPT arrive from chatgpt.com, or chat.openai.com on older links. Perplexity sends perplexity.ai. Copilot sends copilot.microsoft.com, and Gemini sends gemini.google.com. ChatGPT also appends utm_source=chatgpt.com to many of the links it hands out, which makes it the easiest assistant to isolate in any analytics tool.
A note on UTMs, because AI assistants behave differently from ad platforms here. You can't tag inbound AI links yourself; the assistant builds them. ChatGPT adds its own utm_source, others send a bare referrer, and links opened inside mobile apps often send nothing at all. So resist the urge to "fix" attribution with redirects or link cloaking on your canonical URLs. You'd be trading a clean crawlable URL, the thing that got you recommended, for a slightly prettier report.
In GA4, the quick version is an Exploration with session source filtered to those domains. The durable version is a custom channel group: add an "AI assistants" channel matched on a regex of those referrers, and AI traffic shows up next to Organic and Paid in your standard reports from then on. Add landing page as a dimension and you get the most useful view of all: which products the assistants send people to. That's a running scoreboard of which products are winning recommendations.
In Shopify's own analytics, the sessions-by-referrer report surfaces the same domains, and you can watch orders alongside sessions there without touching GA4 at all.
If you want an earlier signal than referrals, look at your logs. Your CDN or bot-traffic report will show GPTBot, OAI-SearchBot, and PerplexityBot hitting your pages long before any shopper clicks through. Crawls with no referrals later means you're being read but not recommended, and that points back at the product data and the reviews, not at visibility.
Two honest caveats. First, you will undercount. In-app browsers and privacy features strip referrers, so a slice of AI traffic lands as Direct no matter what you do. Treat every number you see as a floor. Second, the counts will be small at first, likely tens of sessions a week, and small numbers wobble. Don't redesign anything over one week's blip. Watch the monthly trend.
Expect the sessions themselves to look different from the rest of your traffic. Adobe Analytics found traffic from generative AI sources to US retail sites jumped 1,200 percent between July 2024 and February 2025, with those visitors browsing 12 percent more pages and bouncing 23 percent less than everyone else. The conversion story moved even faster. In Adobe's data, AI-referred visitors converted 43 percent worse than average in July 2024, only 9 percent worse by February 2025, and by May 2026 they converted 54 percent better, per Adobe figures reported by Digital Commerce 360 from over a trillion US retail visits. Different datasets disagree on the exact multiple, which is exactly why the twenty-minute setup above matters. Measure your own store, not somebody's average. Including ours.
Then put the numbers to work. Compare the AI segment's conversion rate and average order value against your own site average, not against anyone's benchmark. Check which landing pages the segment concentrates on, and give those pages the mobile once-over from the next section first, since they're catching the highest-intent traffic you have. And when you decide to change one of those pages, measure the change properly instead of eyeballing a weekly chart.
A monthly ritual is enough: sessions from AI referrers, orders from AI referrers, and the top three landing products. Fifteen minutes in a spreadsheet. The month that first number doubles, and on the current growth curves it will, you'll be glad the trend line started today.
Fewer visitors, deeper visits, later in the funnel. That profile rewards a store that closes well, which is where this article started.
What I'd actually check this week
Open your own store on your phone. Not the desktop preview. Your phone, on cellular, like a stranger would.
- Can you see the Add to Cart button on a product page without scrolling? If not, that's the first fix, and it's usually a theme setting. (More on mobile conversion.)
- Are there reviews on the page, with a count, near the price? If you've got none, getting even a handful up is the highest-return thing here. (Why social proof carries the sale.)
- Time the first load. If the homepage takes more than a few seconds on cellular, trim the hero images before anything else.
- Is there a shipping, returns, or guarantee line within a thumb's reach of the buy button? If not, add one sentence. (The product-page details that quietly close people.)
None of this is exotic. That's the point. The AI shopper is the most generous traffic this channel has produced in years, and the work to deserve it is mostly unglamorous housekeeping you already know how to do. If you want the full playbook, start with our CRO guide for Shopify.
Questions merchants keep asking
Do AI shoppers convert better than regular Shopify traffic?
Yes. AI-referred shoppers convert at close to double the site average. The assistant already did the comparing and filtering, so by the time the buyer reaches your page they are mostly decided.
How do I get my Shopify store recommended by ChatGPT or Gemini?
Shopify already exposes your catalog to AI assistants through agentic storefronts, which are on by default. To actually get cited, keep clean product titles, valid structured data, and real reviews. It is the same hygiene that helps you convert.
What is agentic commerce on Shopify?
Agentic commerce is when an AI assistant can read your products and complete a purchase for the shopper, without them clicking a search result or browsing a collection page. Shopify turned this on for every store.
Is generative engine optimization (GEO) the same as CRO?
They overlap almost completely. Clear titles, fast mobile pages, structured data, and reviews get you cited by AI and get you bought by humans. It is one body of work with two payoffs.
How do I make my Shopify store ready for AI shoppers?
Open your store on a phone, on cellular. Get Add to Cart above the fold, put reviews near the price, trim slow hero images, and add a shipping or returns line within reach of the buy button.
Does ChatGPT crawl Shopify stores?
Yes, two ways. OpenAI runs OAI-SearchBot to index pages for ChatGPT search, ChatGPT-User to fetch live pages during conversations, and GPTBot to gather training data. Separately, Shopify feeds your product data to ChatGPT through its agentic storefront, so your catalog can surface even before a crawl.
Should I block AI bots from my Shopify store?
For a store, almost never. Blocking GPTBot, OAI-SearchBot, or PerplexityBot in robots.txt makes you invisible to the fastest-growing referral channel in retail. Publishers with licensing concerns have a case for blocking training bots. A merchant mostly wants to be found.
How do I track ChatGPT traffic in GA4?
Filter session source for chatgpt.com and chat.openai.com, or build a custom channel group that matches AI referrers like perplexity.ai, copilot.microsoft.com, and gemini.google.com. ChatGPT also appends utm_source=chatgpt.com to many links. Expect an undercount, because in-app browsers strip referrers.
What is llms.txt and does my Shopify store need one?
It's a proposed convention: a plain-text file at yourstore.com/llms.txt that gives AI systems a short map of your most useful pages and policies. No major assistant has publicly committed to reading it yet, so treat it as a cheap ten-minute bet rather than a requirement.
The recommendation will come. The only question is whether the shopper it brings finds a store that's ready, or one that sends them back to ask the AI for the next option on the list.
