Free every Thursday · 59,694 Readers
Fastlane Insider Newsletter: What's working for Shopify in the AI era?
Free every Thursday · 59,694 Readers
Fastlane Insider Newsletter: What's working for Shopify in the AI era?
Your analytics show ChatGPT sending 40 sessions last month, and a headline says AI referrals doubled. Both can be true at the same time, and the gap between them is where most of this year’s bad decisions are being made.
Every few weeks a new number lands saying AI shopping is about to take over, and every few weeks a merchant asks me whether they are already behind. The honest answer is that the numbers are real, they are measuring different things, and none of them tells you what your store should do on Monday. This piece sorts out which is which, using the Bain brief from May 2026, Adobe’s October 2026 report and the two big forecasts for 2030.
Bain’s May 2026 brief says ChatGPT shopping referrals more than doubled over the past year in the US, UK, Germany and France, and that figure is an estimate from Similarweb, a web analytics company, not Bain’s own count. It covers referrals, meaning clicks that left ChatGPT and landed on a retailer’s site.
The same brief adds the sentence most headlines drop. AI accounts for up to a quarter of referral traffic for some retailers, yet it is still less than 1 percent of their total traffic. Both halves are true. A channel can double and still be small, because doubling a very small number gives you a small number. Retail Times’ write-up of the brief carries a useful companion fact: nearly 70 percent of US consumers still begin shopping on Amazon, and more than 50 percent start with Google search.
Note what “up to a quarter of referral traffic” does and does not say. Referral traffic is one slice of visits, the part that arrives from other websites, and it excludes direct visits, email, paid and organic search. For a store whose traffic mostly comes from those, 25 percent of referrals can still be a rounding error in total traffic. I’d argue the right way to read the brief is as proof the channel exists and is growing fast, and as a reminder to check your own number before you believe anyone else’s.
Adobe’s October 29, 2026 report found that AI traffic to US retail sites grew 71 percent year over year in September and converted 90 percent better than non-AI traffic, the highest level Adobe has observed to date. That is a different kind of evidence from Bain’s, because it measures what visitors do after they arrive.
The numbers are worth reading carefully. Adobe’s analysis covers more than a trillion visits to US retail sites, and it defines AI traffic as visits where a user clicked a link from an AI assistant. As of September 2026, AI visitors spent 73 percent more time on site, viewed 25 percent more pages, added to cart 53 percent more often, and produced 83 percent higher revenue per visit. A year earlier the picture was the reverse: in August 2025 AI traffic converted 9 percent worse than other traffic, and about twelve months back its visits were worth less than half of a typical one.
That swing is the most useful thing in the whole report. It suggests AI visitors have changed from curious browsers into people who arrive having already compared options. Adobe expects the revenue advantage to double during the holiday season and projects AI traffic growth of 159 percent on Thanksgiving, 95 percent on Black Friday and 88 percent on Cyber Monday. Those are Adobe’s forecasts, not results, so treat them as a planning signal for November, not a promise.
Here is a labelled illustration, not a measured result. Say a store gets 200,000 sessions a month and AI sends 0.5 percent of them, which is 1,000 sessions. At a 2.5 percent baseline conversion that is 25 orders. If those visitors convert 90 percent better, as Adobe saw, the same 1,000 sessions produce about 47 orders. That is roughly 22 extra orders a month from a channel that still looks like a rounding error in your traffic report. It is a real prize and a small one, which is exactly why the order of operations matters.
Bain and Morgan Stanley forecast different sizes for agentic commerce in 2030 because they define it differently, so the two numbers are not independent confirmation of each other. Bain’s December 17, 2025 snap chart puts the US market at $300 billion to $500 billion, roughly 15 to 25 percent of US ecommerce. Its definition counts purchases initiated, influenced or completed by AI agents, and it leaves out journeys that only use AI for search or discovery.
Morgan Stanley’s December 8, 2025 research estimates $190 billion to $385 billion of US ecommerce spending by 2030, which it frames as about 10 percent of online retail in the base case and up to 20 percent in the optimistic one. The $385 billion figure is the bull case, not a midpoint. Plenty of coverage treats the two firms as landing on the same number. They overlap, but they are different rulers.
Both firms also hedge in ways the headlines skip. Bain notes that around half of consumers say they are not comfortable letting AI complete a transaction without their involvement, and it describes timing and depth of adoption as uncertain. My suggestion is to plan around the low end of either range and treat the high end as upside. If you want a number to put in a slide, use the range and name who produced it.
As of November 2026, a ChatGPT shopping referral mostly ends on the merchant’s own store, not inside the chat. Bain’s brief notes that ChatGPT and Perplexity have moved away from instant checkout. That makes winning the referral matter more.
On the Shopify side, Agentic Storefronts went live by default for eligible stores in March 2026, and Shopify’s Help Center describes how products reach ChatGPT, Microsoft Copilot, Google AI Mode and Gemini, and Meta, with controls over which AI channels you sell on and whether direct checkout is on (it is off by default). Our walkthrough of what every merchant needs to check now that Agentic Storefronts is live covers the admin steps. The point here is narrower: because the buyer lands on your site, your product page and checkout decide whether the referral becomes revenue.
That also explains Adobe’s conversion gap. A visitor who has already asked an assistant three follow-up questions arrives with less patience for a vague product title or a missing size chart. Adobe’s own advice for retailers is to make content easy for AI to find, understand and cite, and it reports the average US retail homepage scored 63 percent on its readability measure.
Most of the influence AI has on a purchase never shows up as a referral at all, which is why your analytics undercount it. Ivan Slobodin of Searchable, a company that sells AI visibility tracking, described this on episode 446 of the podcast. He had bought a pair of headphones after spending a couple of hours questioning ChatGPT, then typed one search on Amazon and bought them. From Amazon’s side he looked like an ideal customer, and none of the research that decided the purchase was visible to anyone.
Now, that is one buyer’s story, not a study, and Slobodin has a product to sell. It does match the structure of the Bain and Adobe figures, though. Both count clicks from AI tools to retail sites. A shopper who researches in ChatGPT and then searches your brand name in Google, or types your URL, lands in your direct or branded search numbers. So the AI channel in your reports is the visible edge of something larger and harder to measure.
The practical consequence is that “ChatGPT sent us 40 sessions” is a floor, not a verdict. It is also why a monthly check of unbranded buying questions is a better early signal than any single referral count. Our look at StoreHero’s visibility benchmark shows how to run that check and how far to trust vendor data while you do.
Start with product data, then measurement, and spend money last. Whatever your stage, the first hour goes to your top 20 products: put the plain product type in the title, move the facts a buyer checks first (size, material, compatibility, what is in the box) into structured fields, and write return and shipping terms in plain sentences.
The reason is mechanical. Shopify says its semantic search uses the text and image data of your products, and our guide to Shopify product data enrichment shows how to move attributes out of description text and into metafields. That work also helps Google Shopping and on-site search, so it pays even if AI stays small. It is not the whole story, because assistants also lean on reviews, retailer pages and other sites that mention you, so what others say about your products matters too. If you want the wider picture of how discovery tools read a catalog, how AI is improving product discovery covers it.
Under about $500K a year, stop there and check your referral report once a month. Between $500K and $2M, add a monthly run of ten unbranded buying questions in ChatGPT, Perplexity and Google, and log who gets named instead of you. From $2M up, give one person ownership of AI discovery and consider a tracker. The options fall into a few categories: visibility trackers such as Paz.ai, Profound, Searchable and StoreHero (listed alphabetically), catalog enrichment tools, and the free native route through your Shopify admin. They overlap, and I would not buy two of them.
Here is the subtraction. If a vendor tells you to build a separate AI feed, rewrite your whole catalog or reallocate paid budget because of a doubling headline, wait. At under 1 percent of traffic, the case for a large spend rests on a forecast, and forecasts disagree. The 18 month test I use is simple: would you still be glad you did this if AI referrals doubled again, and equally glad if they stalled? Clean product data passes both. A new tool subscription often passes only one.
ChatGPT shopping referrals more than doubled over the past year in the US, UK, Germany and France, according to Bain’s May 2026 brief, which cites Similarweb estimates. The same brief says AI is up to a quarter of referral traffic for some retailers but still under 1 percent of total traffic. Adobe measured US retail sites separately and found AI traffic grew 71 percent year over year in September 2026. The growth is real, and the base is small, so check your own analytics referral report before drawing conclusions about your store.
As of September 2026, Adobe found AI visitors to US retail sites converted 90 percent better than non-AI visitors and produced 83 percent higher revenue per visit. A year earlier the gap ran the other way, with AI traffic converting 9 percent worse in August 2025. Adobe counts visits where a user clicked a link from an AI assistant, so shoppers who research in an assistant and then search your brand elsewhere are not included. Treat the figure as a retail-wide average, not a promise for any one store.
Bain estimates $300 billion to $500 billion in US agentic commerce by 2030, about 15 to 25 percent of ecommerce, while Morgan Stanley estimates $190 billion to $385 billion, about 10 to 20 percent. They define the term differently: Bain counts purchases AI agents initiate, influence or complete, and Morgan Stanley’s range is its base and bull cases. Neither confirms the other. Plan around the low end of the range, name the source when you quote it, and treat anything higher as upside.
As of November 2026, ChatGPT works mainly as a discovery channel, and the buyer typically finishes checkout on the merchant’s own store. Bain’s brief notes ChatGPT and Perplexity moving away from instant checkout. Shopify’s Agentic Storefronts still lists eligible products in ChatGPT, Copilot, Google AI Mode and Gemini, and Meta, with direct checkout off by default. Check which AI channels you sell on in your Shopify admin, since availability varies by channel and region.
Fix your top 20 product pages first. Put the plain product type in each title, move size, material and compatibility into structured fields, and write return and shipping terms in plain sentences. Then check your referral report monthly, and add ten unbranded buying questions to test once you pass about $500K a year. This work also helps Google Shopping and on-site search, so it holds up whether AI referrals keep doubling or stall. Hold off on new tools until a manual check shows a gap worth paying to close.
