Heiko Hotz, who works on AI strategy and transformation at Google, argued in a LinkedIn post published on September 22 or 23, 2026 that the company's Shopping Graph, which he put at more than 60 billion product listings, makes Google a natural foundation for agentic commerce - the AI assistants that search, compare and buy on a shopper's behalf.

In Short

A Google employee who advises companies on AI wrote on LinkedIn that Google holds one of the biggest catalogues of things for sale anywhere, more than 60 billion product listings, and that AI shopping assistants need exactly that kind of data to work. It matters to shops and advertisers because an assistant that picks products for you can only choose from what its data knows about, and whether that data is fresh and complete decides who gets picked. Independent studies and earlier PPC Land reporting show large gaps between the size of Google's catalogue and what shoppers actually see or pay, so the post describes Google's reach rather than any single merchant's visibility.

A short post with large numbers

LinkedIn displays only a relative age for posts. On a copy captured on September 27, 2026, this one read four days old, which places publication on September 22 or 23. It runs to fewer than 200 words. It contains no new product, no policy change and no figure Google has not already published. What it offers instead is a compact version of an argument Google has been assembling all year: that whoever holds the most current structured product data will sit underneath AI-mediated shopping, whichever assistant the shopper happens to use.

The post opens on perception. "When people think of e-commerce, Google is not always the first name that comes to mind. Most people think of Amazon, Walmart, etc," Hotz wrote. People, he continued, are "usually quite surprised to learn that Google quietly runs the largest shopping graph in the world: over 60 billion product listings!" He added "more than a billion shopping queries happening across Google every day" before setting out the thesis in bold type: "agentic commerce runs on grounding data!"

Three labelled factors followed. Under Real-Time World State, he wrote: "We update billions of product listings every hour and an agent is only as good as the freshness of its environment." Under Open-Web Breadth: "Unlike closed marketplaces, the Shopping Graph indexes the broader web. That goes from niche DTC brands all the way to global retail giants." And under Structured Intent to Action: "Pairing real-time multimodal reasoning with structured catalog data is the key to get from conversational discovery (ie in ChatGPT, Gemini, etc) to frictionless checkout."

The closing line was framed as opinion. Prefaced with "IMHO", it placed Google among the best partners in the agentic commerce space and invited discussion in the comments. The copy reviewed for this article shows no reaction, comment or repost counts.

The author and the audience

Hotz's LinkedIn headline reads "AI Strategy & Transformation @ Google | O'Reilly Author | LBS Faculty | Helping leaders navigate the agentic economy". The profile lists London as his location, London Business School alongside Google, and 28,550 followers. His O'Reilly author page, which still describes him as a Generative AI Global Blackbelt at Google Cloud, lists earlier roles at Amazon and AWS, a master's degree in physics and an MBA from London Business School. In August 2025 he discussed the Agent2Agent protocol with Google colleague Sokratis Kartakis on O'Reilly's Generative AI in the Real World podcast.

The vocabulary - "partners", "leaders", "agentic economy" - points to enterprise buyers rather than media buyers. The implied reader is a retailer weighing which company to build its buying assistant with, not an advertiser setting a Shopping campaign budget. Google Cloud made a similar case on January 11, 2026, the day Google unveiled the Universal Commerce Protocol at the National Retail Federation conference in New York, when a Google Cloud blog postnamed the Knowledge Graph and the Shopping Graph among the assets it brings to retail customers.

Written partly in the first person plural, the post nonetheless comes from an individual account. It is not a Google statement, and it cites no source for its figures. Those figures can, however, be checked against what Google itself has said.

Counting 60 billion

Google's public numbers have moved quickly. When the company rolled out agentic checkout for the 2025 holiday season on November 13, 2025, the Shopping Graph held more than 50 billion product listings, with 2 billion updated every hour. The 50 billion figure was still in circulation in April 2026, when sponsored tiles began appearing in the Shopping tab's free listing grid. A day before Google Marketing Live, Google's shopping blog described the graph as the world's most comprehensive catalogue of more than 60 billion product listings and said people shop across Google more than a billion times a day. At the event itself on May 20, 2026, Vidhya Srinivasan put the total at more than 60 billion, with more than 2 billion updated hourly.

Three details in the post drift from Google's corporate wording. "Largest shopping graph in the world" is Hotz's phrase; Google's blog chose "most comprehensive", and neither version came with a comparison against Amazon's catalogue or anyone else's. The billion figure changes its unit along the way: Google counts the times people shop across its services, while the post speaks of queries, and neither defines the measure. And listings are not products. The same television stocked by forty retailers produces forty listings, and Google has not said, in any of the material reviewed for this article, how many distinct items sit beneath the headline count.

The refresh rate invites simple arithmetic. At 2 billion updates an hour, 60 billion listings would take 30 hours to turn over once if updates were spread evenly across the graph. Google has not said how they are distributed.

The premise about perception is also shakier than the post allows. A Smarty Marketing survey of 1,295 US residents, updated on May 12, 2026, found that 56.68% of respondents preferred Google as the starting point for a product search, against 28.96% for Amazon and 7.26% for ChatGPT. On that evidence Google is already the first name for most American shoppers, which is precisely why its catalogue carries so much weight for the merchants inside it.

Freshness depends on the merchant's pipe

The post supplies its own test: an agent is only as good as the freshness of its environment. PPC Land's archive records several occasions when that environment lagged. On January 13, 2026, Emmanuel Flossie, a Google Shopping specialist, found Merchant Center's automatic import, which Google says refreshes every 24 hours, showing products nine to ten days out of date. On May 5, 2026, Merchant Center added a beta that builds listings from a one-time AI scan of a merchant's website, with no refresh cycle promised. The Content API for Shopping, long the route for pushing price and stock changes quickly, was set to shut down on August 18, 2026 in favour of the Merchant API.

Each of those routes feeds the same graph. The hourly figure describes throughput at Google's end. It does not describe how fast a price cut on a merchant's own site reaches the corresponding listing, which depends on how the merchant submits data in the first place. For a Shopping ad, stale data wastes a click; in an agent-led flow, it can decide which product is put forward at all.

Indexed is not the same as shown

The breadth argument runs into measurement of Google's own conversational surface. Across more than 100,000 identical shopping queries over 21 days in July 2026, Productrise found that AI Mode returned roughly 95% fewer product listings than standard search. Standard results carried products on about 88% of queries and AI Mode on about 23%; where both did, the averages were 22.5 listings against 4.3, and only 0.8% of products overlapped. A second dataset, covering more than 2 million listings in the United States and United Kingdom between August 9 and 31, 2026, found that only 1.28% of products ranking in traditional search also appeared in AI Mode for the same query on the same day, and that where identical products appeared on both surfaces, the lead price in AI Mode averaged 21.6% higher.

Those figures carry caveats. Productrise sells product visibility monitoring, its price comparison covered lead offers rather than the cheapest offer available after a click, and the study contained an unexplained inconsistency in how often prices differed. Google has not published counter-measurement.

The implication for the open-web argument is plain enough. A graph that indexes niche direct-to-consumer brands does not guarantee that any of them reaches the four or so products a typical AI Mode answer shows. Google added AI performance insights to Merchant Center on May 27, 2026, reporting share of voice and an attribute completeness score across AI Mode, AI Overviews and the Gemini app. How those signals are weighted when products are selected is not public.

"Closed marketplaces" is the other half of the contrast, and the post opens by naming Amazon and Walmart. Walmart is, as it happens, a UCP co-developer, and it switched on checkout inside Gemini and AI Mode on January 11, 2026. Amazon's position has moved. It added crawlers from OpenAI, Anthropic, Meta, Google and Huawei to its robots.txt restrictions on August 21, 2025, then joined the UCP Tech Council with Meta, Microsoft, Salesforce and Stripe on April 24, 2026. In May, a job listing showed Amazon building a team to connect its services with outside AI platforms through APIs, which analyst Juozas Kaziukenas read as controlled integrations with a select few. Closed, in Amazon's case, increasingly means gated on its own terms.

From conversation to checkout

The third factor depends on plumbing that has spread slowly. Google unveiled UCP with Shopify, Etsy, Wayfair, Target and Walmart as co-developers, yet a scan of more than 3 million websites published on May 21, 2026 found 26 sites with a publicly detectable implementation, none of them the co-developers. Originality.ai, which ran the scan, also raised the possibility that prioritisation among identical offers from compliant merchants could become a paid layer. The point is speculative. The post's "partner" framing does not touch it.

Google has since taken the merchant out of part of the decision. According to Search Engine Roundtable, Google emailed Shopify merchants on September 22, 2026 to tell them their stores had been matched to Merchant Center and that native checkout was enabled in AI Mode and Gemini, with eligible products included automatically. A wider rollout had begun on Friday, September 18. Merchants who want checkout kept on their own sites have to switch it off in the Shopify admin, under Sales channels, then Agentic. That is frictionless checkout in the most literal sense, and it arrived in the same week as the post.

Paid placement is arriving alongside it. Similarweb data from August 17, 2026 put nearly 30% of ad-eligible AI Mode queries as carrying advertisements, and on August 31 SERP researcher Brodie Clark documented a shopping ads carousel running inline within an AI Mode response for the first time. The same graph feeds both the organic answer and the sponsored units around it.

ChatGPT appears in the post as a venue for conversational discovery, with nothing said about whether Shopping Graph data reaches it. OpenAI's own experiment offers a caution about the frictionless end of the argument. Instant Checkout, opened in September 2025, was shut down in March 2026 after Walmart disclosed that conversion rates for products bought inside ChatGPT were three times lower than for purchases that clicked out to its website. Separately, research covered in August 2025 found ChatGPT's search results aligned closely with Google's rather than Bing's, contrary to OpenAI's documentation at the time.

Google's own commerce chief sounded more guarded than the post back in February. Asked on the Frontier CMO podcast whether agent-to-agent commerce would become mainstream in 2026, Srinivasan said she expected more noise, with the building blocks still being laid.

Grounding has more than one supplier

Grounding is the practice of giving a language model current, verifiable information at the moment it answers, usually through retrieval-augmented generation. That agentic commerce depends on it is hardly controversial. The suggestion that Google owns that layer for shopping is another matter. Microsoft said in February 2026 that its grounding technology powers nearly every major AI assistant, and on June 2, 2026 it opened Web IQ, a suite of grounding APIs built for agentson top of Bing's index. OpenAI's Ads Manager, meanwhile, accepts product feeds of up to one million SKUs per advertiser.

Each platform asks merchants for the same structured data. The graph Hotz describes is built, to a large degree, from what merchants send Google through Merchant Center.

A shopping business regulators know well

Whatever consumers think of first, European courts have long associated Google with shopping. On July 1, 2026, the Stockholm Patent and Market Court ordered Google to pay Klarna Technologies, owner of the comparison site formerly known as PriceRunner, principal damages of 950 million pounds for the United Kingdom market, finding that the changes Google made in 2017 to comply with the European Commission's Shopping decision never ended the abuse. The comparison shopping service layer, through which merchants must route Shopping ads in 21 European countries, exists because of that case.

What the post leaves unsaid

For advertisers and retailers, the significance of the post lies less in its numbers than in its framing. It presents the Shopping Graph as Google's asset. Merchants experience it as a copy of their own catalogue, held and ranked by a company that also sells advertising around it. Agentic transactions strip out the click, referral and session records that digital measurement was built on, which is why NIQ and Similarweb said on September 2, 2026 that they would begin measuring shopping inside ChatGPT and Gemini from the fourth quarter. Until tools of that kind mature, a merchant's view of how agents treat its products rests largely on reporting supplied by the platforms that run the agents.

None of this makes the post's figures wrong. Sixty billion listings and 2 billion hourly updates are Google's published numbers, and the scale is real. What the post omits are the conditions attached: freshness depends on each merchant's feed, breadth of index narrows sharply at the point of display, and the path from conversation to checkout is being built with defaults many merchants did not choose. A partner, in Hotz's telling. For the merchants whose data fills the graph, also a gatekeeper.

Timeline

Summary

Who: Heiko Hotz, whose LinkedIn headline describes his role as AI Strategy & Transformation at Google and who also teaches at London Business School, writing on his personal account with 28,550 followers. The post bears on retailers and brands whose product data feeds Google's Shopping Graph, and on advertisers buying Shopping and AI Mode placements.

What: A LinkedIn post of fewer than 200 words arguing that Google's Shopping Graph, put at more than 60 billion product listings with billions updated hourly and more than a billion shopping queries a day, is "a crucial foundation for agentic commerce" because of three factors: real-time freshness, open-web breadth and a path from conversational discovery to checkout. The post contains no new figures; its numbers match what Google has published since May 2026, with differences in wording on the "largest" claim and on queries versus shopping occasions.

When: Published on September 22 or 23, 2026, based on LinkedIn's relative timestamp of four days on a copy captured on September 27, 2026. It appeared in the same week that Google emailed Shopify merchants, on September 22, about automatically enabled native checkout in AI Mode and Gemini.

Where: LinkedIn, addressing a global enterprise audience. The checkout and measurement developments it touches on centre on the United States, with Productrise data covering the United States and United Kingdom and the comparison shopping remedy applying across 21 European countries.

Why: The post matters because it states plainly the commercial logic behind Google's agentic commerce push: control of fresh, structured product data as the layer every shopping agent needs. PPC Land reporting shows each of its three pillars carries conditions the post omits, from feed lags of nine to ten days, to AI Mode showing roughly 95% fewer products than standard search, to checkout defaults switched on without merchant opt-in.