Shopify Lays Out Its Agentic Commerce Strategy at Goldman Sachs: How Catalog and Checkout Keep AI From Cutting It Out
Shopify's CFO told the Goldman Sachs conference that agentic commerce is only 6 to 9 months old. Why Shopify says AI won't disintermediate it, how Shopify Catalog and SimGym work, and what the SEO shift means for merchants.
Key Takeaways
- At the Goldman Sachs Communacopia + Technology Conference 2026 on September 10, Shopify executives said real traffic from shoppers who discover products in an LLM and then buy is only 6 to 9 months old. Even so, LLM-started searches are 2.5 times more likely to land directly on a product page, with roughly an 80% conversion uplift
- Asked whether AI will disintermediate Shopify, CFO Jeff Hoffmeister said that whether shoppers arrive through Meta's Muse or OpenAI, only the entry point changes; the transaction still completes on the merchant's site and Shopify's infrastructure. He put Shopify Catalog, with live inventory and pricing, at the center of discovery
- For merchants, what matters is how deeply product attributes are written and whether the brand gets searched by name. AI-driven order share remains undisclosed, however, and checkout inside Muse runs over more than one path
It started with "6 to 9 months in"
Shopify used the Goldman Sachs Communacopia + Technology Conference 2026 to argue that artificial intelligence could expand, rather than weaken, its role in e-commerce. The company said new AI-driven discovery tools are changing how shoppers find products, but that transactions still flow through Shopify's platform.
www.investing.comOn September 10, 2026, Shopify CFO Jeff Hoffmeister and Head of Growth Archie Abrams took the stage at the Goldman Sachs Communacopia + Technology Conference 2026. Investing.com published a summary along with the full transcript.
Hoffmeister was careful about timing. With so many product announcements, he said, it is easy to assume agentic commerce is already a huge share of commerce, and definitions vary from vendor to vendor.
We're 6 to 9 months into seeing real traffic where someone goes to one of the LLMs, does some discovery, and then ends up on a merchant's website and then transacts.Source: Jeff Hoffmeister (Shopify CFO)
He still sees an inflection. Roughly 20% of global commerce happens online, and in recent years about a point a year has shifted from offline to online. Shopify expects agentic commerce to accelerate that shift, although it gave no figure for how much.
Two numbers backed the claim. Shoppers who start their search in an LLM are 2.5 times more likely to land right on the product page than with traditional search, and conversion shows roughly an 80% uplift. The 2.5x figure was also cited on the Q2 earnings call. The baseline for the 80% uplift was not specified, and the share of total orders coming through AI remains undisclosed.
The real investor question: will AI cut Shopify out?
At the center of the session was a question that resurfaces every time a new agentic technology appears. If AI takes over the path from discovery to purchase, does Shopify get bypassed, and won't the LLM companies want a share of the economics? Moderator Gabriela Borges put it directly.
Hoffmeister built his answer on the different incentives of LLMs and merchants. The LLMs are trying to funnel consumers to merchants, often tied to their advertising ambitions, and what they compete on is the quality of the referral. LLM queries now run about two dozen words, versus five to eight in traditional search, and sending people to exactly the right place is how the platforms build their reputations.
From the merchant's side the picture looks different. Merchants spend countless hours and dollars on photography, brand presentation and product descriptions, so the vast majority want the transaction to happen on their own website. Even when a purchase seems to happen inside an LLM, he explained, it is often a mirror running on the merchant's infrastructure, and he said Shopify works with Meta's personal agent Muse the "exact same way" it works with OpenAI and others.
Abrams added that fragmentation itself works in Shopify's favor.
As the fragmentation gets even more over time, all roads lead back to Shopify. You want one place to manage your inventory. You want one checkout.
Inventory, shipping, tax calculation, payments, running the website, and pricing or product advice from its AI assistant Sidekick: by Hoffmeister's count Shopify does 12 to 20 things for a merchant, depending on definitions. He also stressed that Shopify is the only admin that lets merchants switch individual AI platforms on and off from one pane of glass.
That lines up with how Agentic Storefronts was designed when it rolled out broadly in March. According to Shopify's announcement, selling across ChatGPT, Copilot, AI Mode in Google Search and other channels is managed centrally from the admin, and orders arrive with referral attribution. Merchants remain the merchant of record and keep their customer relationships and data. What Shopify is defending is not the entry point but a structure in which the transaction record and the payment stay on its infrastructure.
What Shopify Catalog holds that makes it work
Abrams' description of Catalog breaks down into three layers. The foundation is a taxonomy that covers basic facets plus an understanding of imagery, the product's tone, and what type of buyer it appeals to. On top of that sits deduplication, which recognizes the same item listed by multiple merchants as one product. The top layer is live inventory and live pricing. An LLM relying on web search, Abrams noted, runs into data that is not real-time and prices that are wrong.
| Dimension | What an LLM gets from web search | What Shopify Catalog provides |
|---|---|---|
| Inventory | A snapshot from the last crawl, with no way to tell if it is in stock now | Live inventory |
| Price | Stale or incorrect prices mixed in | Live pricing |
| Product identity | The same item scattered across separate listings | Identical products from multiple merchants deduplicated |
| Attributes | Whatever the page happens to say | A taxonomy covering basic facets plus image understanding, tone and the buyer it appeals to |
So which products benefit? Hoffmeister first pointed to brands that grew fast five or six years ago and then plateaued. As shoppers ask LLMs more specific questions, many of them are taking off again. Specification-heavy products benefit too, such as electronics or furniture that has to meet a certain size or weight.
He also described where this is heading. When someone shopping for a suit asks about fabric weight, breathability or whether it suits a particular event, attributes that rarely appeared in product descriptions become the deciding factor. If merchants write those details and Catalog folds them in, an LLM can recommend a small manufacturer in Portugal that matches the buyer's style.
This overlaps with the standards conversation. Forkast's report on the W3C and GS1 workshop held September 8 and 9 notes that both the Universal Commerce Protocol (UCP) and the Agentic Commerce Protocol (ACP) use the GTIN product identifier as the primary key. Shopify, which has deployed UCP and WebMCP across millions of merchants, pointed there to the current gaps in attestation. Catalog's deduplication is partly Shopify answering the "is this the same product" problem with its own data.
SimGym and skills: Shopify puts agents to work
Abrams described SimGym, pitched to large brands, as a tool that has agents simulate variants of landing pages, pricing, product imagery and ads, runs the experiment and launches the winner.
What public documentation confirms is narrower. Shopify Engineering's write-up describes sending hundreds of AI shoppers, each with a persona, budget and shopping intent, through a store in cloud browsers so that an A/B test taking weeks is simulated in minutes, generating 400,000 shopping sessions a day. The feature that opened as an AI Research Preview in the March Changelog post compares themes. How far pricing and ad experiments are available could not be confirmed.
On merchant acquisition, Abrams said the logo maker and business name generator have been rebuilt as agent-accessible skills usable from Lovable, ChatGPT, Claude, Manus, Replit and Groq, so someone asking an AI about starting a business can discover Shopify right there.
SEO is not disappearing; it is becoming brand search
Asked about the outlook for SEO, Abrams said traditional search is still growing but shifting toward navigational queries, where the shopper already knows the destination.
In the past, you might have had queries like black pants, and now you have Vuori pants instead.
Researching which pants are best has moved to social platforms and agentic channels; once shoppers settle on a brand, they search for it by name and visit the site. Abrams framed this as a tailwind for independent brands, which struggled to win generic terms.
Turn it around, though, and a brand that never made it into shoppers' minds through agents or social will not appear at the brand-search stage at all. Even if total search traffic holds up, what decides the outcome is whether the brand was chosen before that search.
What to discount
These were remarks made to investors, so some discounting is in order. On disclosure, the share and value of AI-driven orders and the baseline for the 80% uplift were once again not provided.
Checkout paths are also less simple than described. PPC Land reports that the payment method Meta featured for Muse is Link by Stripe, which issues a one-time-use card for each transaction, with Shop Pay described as coming soon. RetailBoss likewise notes that saving Shop Pay as a wallet is a forthcoming capability. An independent code analysis cited by PPC Land found three checkout paths in Muse and judged the Shopify path to be most likely Shop Pay over UCP, but Meta names no protocol. How Shopify earns payments revenue on each path is undisclosed.
The same article cites a practitioner who says UCP access on Shopify requires a double opt-in, which in practice limits it to the largest agent platforms. When Originality.ai scanned more than three million public websites in May 2026, only 26 carried UCP files. Platform-mediated connections do not show up in that count, but adoption of the standard across the open web is still early.
Cost remains an issue as well. Hoffmeister acknowledged that subscription gross margins are a balancing act as LLM costs rise with Sidekick usage, and said Shopify contains them with distillation (transferring a large model's capability into a smaller one) and caching. Shopify's Q2 press release guided Q3 revenue growth in the low-30s percent range but gross profit growth in the mid-to-high twenties.
What merchants should take away
The easiest place to start is adding use-case attributes to product descriptions. Writing in breathability, fabric weight, suitable occasions and whatever else a buyer might ask an LLM turns product data into something that can answer two-dozen-word questions.
Brand search landing experiences are worth auditing too. As the weight shifts from ranking for generic terms to converting people who arrive by brand name, the quality of product pages and site paths becomes decisive.
Opening AI channels also takes judgment. If channels can be switched individually and orders carry referral attribution, merchants can decide the order in which to open them while measuring results. Merchants not on Shopify also have an official option to add products to Catalog through the Agentic plan. Product data scattered across an in-house cart and marketplaces, with inventory and pricing lagging behind, is the first thing to suspect when an agent cannot see a store.
Conclusion
Shopify's executives were arguing that entry points can multiply without limit, as long as the transaction record, inventory and payments flow back to Shopify's infrastructure. Catalog, the single admin and SimGym are the machinery for that.
A 6-to-9-month timeline also means the numbers are still hard to verify. The things to watch next are when Shop Pay becomes available in Muse, whether AI-driven order share gets disclosed, and whether Shopify can bring gross profit growth closer to revenue growth while carrying rising LLM costs.
