May 20, 2026
BrightonSEO April 2026, the day before Malte Landwehr’s main-stage talk. Diantha van Surksum and Martijn Hoving sat down with him in the podcast booth between sessions.
His talk the next day was about ChatGPT shopping: how the product carousels work, where ChatGPT finds the products it shows, and how to actually influence what ends up in them. His own answer to that last question contradicts almost everything the e-commerce SEO industry is currently betting on.
The dominant industry narrative is that ChatGPT shopping is moving toward custom integrations. Shopify pilots. Etsy and Walmart product feed connections. Direct OpenAI partnerships where brands upload structured data into a managed graph. Malte’s research, published in Search Engine Land in March 2026 with his Peec AI colleague Tom Wells, says something different.
The formal study figure, across 43,000 unique products and 10 retail categories, is 83%. That’s still a number that should reshape how any e-commerce SEO team allocates time and budget for the next twelve months.
How is ChatGPT actually building its shopping carousels?
The Wells and Landwehr study at Peec AI analyzed over 5,000 ChatGPT carousels, 43,000 carousel products, 200,000 organic shopping results, and 1.1 million shopping fan-out queries.
They decoded base64-encoded fields in ChatGPT's source code that contained Google Shopping parameters: product IDs, offer IDs, and locale information. From those, they could reconstruct the URLs ChatGPT was hitting.
The mechanism is straightforward once you see it. The user types a shopping-intent prompt into ChatGPT.
The model generates a shopping fan-out query that is shorter and more targeted than its standard contextual fan-outs (averaging 7 words versus 12 for standard searches, and only 1.16 per prompt versus 2.4). That shopping fan-out hits a third-party scraping provider that retrieves Google Shopping organic results. The top results come back and populate the carousel.
Eighty-three percent of the products that make it into the carousel match the top 40 Google Shopping organic results. Sixty percent of matches come from the top 10. Eighty-four percent within the top 20.
Malte demonstrated the speed of this himself. He added a previously-absent online shop to Google Merchant Center. The next day it appeared in Google Shopping, and from there in ChatGPT.
The pricing, the reviews, the availability data: all consistent with scraped Google Shopping organic listings. Not with a managed product graph.
Why didn't OpenAI build its own product graph?
The intuitive answer is that they would, eventually. The honest answer is that building a product graph is much harder than it looks. Google has been building theirs for years, and it still isn't complete.
Malte ran through the edge cases that make a useful product graph an unreasonable engineering problem:
Variants. "The iPhone 16" doesn't exist as a single product. It exists as iPhone 16 in blue with 256GB, iPhone 16 in black with 1TB, and every other combination. For some products, the variants are load-bearing for the buying decision. For others, they're noise. Building a graph that handles this requires hand-tuning by category.
Marketplaces and reputation. Walmart, eBay, Kaufland, Amazon. Each one is a multi-merchant venue where some merchants are lying about what they're selling. A "Dyson" listed by an eBay merchant might actually be a cheap knockoff using the genuine ASIN. Building a reputation system to detect this within one marketplace is hard. Doing it across hundreds of marketplaces is a years-long engineering project.
Category-specific schema mismatches. Bicycle frames are measured by inches of frame in some shops and inches of wheel in others. The same physical product appears under different specs depending on the merchant. Mapping requires bicycle-specific logic.Contact lenses have 300 variants per SKU. Different thickness, different curvature. Each one is technically a separate product but functionally interchangeable for many buyers.
Scraping Google Shopping was the path of least resistance. Google had already done the work, badly, for over a decade. Why redo it?
Ready to start scaling your business?
Let's get started
What does this break in your attribution?
Here is the part most e-commerce teams haven't internalized yet. Because ChatGPT is scraping Google Shopping URLs directly, the UTM parameters travel with the scrape.
That means a click that originated in ChatGPT, on a product the user discovered in a ChatGPT shopping carousel, appears in your analytics with a UTM parameter indicating it came from Google Shopping organic.
The fix is mechanical. Two conditions decide where to attribute the click:
- If the HTTP referrer is google.com and the URL has the Google Shopping UTM, attribute it to Google Shopping organic.
- If the HTTP referrer is chatgpt.com or empty, and the URL still has the Google Shopping UTM, attribute it to ChatGPT shopping.
The first step before that fix matters: many e-commerce sites don't even have a UTM tracking parameter on their Google Merchant Center feed in the first place. Without that, you're blind on both surfaces.
What's the actual ChatGPT shopping playbook then?
The strategy collapses into one starting point and two-layered tactics.
Start with Google Merchant Center. Connect the feed. Clear the errors. Check the box for free organic listings. Most of the work of being in ChatGPT shopping carousels is being well-ranked in Google Shopping organic. The 83% finding is the load-bearing fact for budget and roadmap decisions in e-commerce SEO for at least the next year.
Layer one: mine shopping fan-out queries. These are the queries ChatGPT generates internally that the user never typed. They surface category-specific attribute words (e.g., durability, comfort, fit-for-purpose claims) that don't appear in the user's prompt but guide what ChatGPT looks for in the scraped Google Shopping results. Most SEO tools don't yet expose shopping fan-outs distinctly from standard fan-outs. Peec AI does, and a small number of competitors are catching up.
Layer two: write declarative attribute statements. When ChatGPT renders a product carousel, it often scores products on a small set of category attributes (durability, comfort, value for money, suitability for X use case). Malte's recipe for content that ChatGPT will use to re-rank scraped results:
Each statement should cover one attribute, name the use case, and ideally cite a third-party validator ("as proven by reviews from Business Insider and Forbes"). Same statement on the product page. Same statement in the product feed.
17% of carousel placements that don't cleanly trace to Google Shopping appear to come from ChatGPT's re-ranking layer. Re-ranking uses both the scraped data and the content ChatGPT pulled from product detail pages during the fan-out process. The attribute-statement work feeds that re-ranking layer.
Ready to start scaling your business?
Let's get started
How does this fit the broader SEO shift?
A separate thread ran through the interview that's worth surfacing: the kind of SEO leader who thrives in this environment is changing.
Malte's observation: heads of SEO who used to wait weeks for a CEO meeting now have monthly executive calls. Boards suddenly care about AI search visibility. SEO leaders are being asked to produce attribution models, justify resourcing, and explain influence on revenue without the click data they used to lean on. The discipline is starting to look more like paid search ever did.
The mental shift, in Malte's framing, is from owning a high-ROI channel that nobody questioned to having to model overall impact across channels where the click data is partial or absent.
For SEOs who embrace it, the boardroom attention buys budget and scope. For SEOs who liked operating without scrutiny, it isn't a great moment.
The ChatGPT shopping story is one specific case of this broader shift. The skills needed to win in it (Google Merchant Center mastery, attribution modeling, fan-out query analysis, declarative attribute writing) cut across SEO, paid, and content. The team that owns it cuts across, too.
The German .org domain story, Malte still mentions
Every SEO Cast episode ends with the same question: what's your biggest fuck-up?
Malte's is one of the more idiosyncratic answers the podcast has captured. Years back, he noticed that many German cities had not registered their .org domains. He registered a long list of them, planning to monetize them later.
He spent thousands of euros on domains he never used, and the trademark situation made them unusable. In domain land, a free name usually means somebody else has already calculated the legal cost.
Resources
- Search Engine Land: ChatGPT sources 83% of its carousel products from Google Shopping: the Wells/Peec AI study, March 2026
- Peec AI: the AI search visibility tracking platform where Malte is CPO/CMO and Tom Wells is GEO Researcher
- Malte Landwehr on LinkedIn: his current writing on AI search and e-commerce SEO
- Tom Wells on LinkedIn: the GEO researcher behind the carousel study
- Semrush: ChatGPT searches Google Shopping to create recommendations: independent corroboration of the scraping mechanism
Listen to the full conversation
- The full SEO Cast interview with Malte Landwehr was recorded at BrightonSEO April 2026, the day before his main-stage talk. He covers the Wells/Peec AI carousel study in more detail, the attribution mechanics, the optimization tactics, and the broader observation about how the role of head of SEO is shifting in the boardroom. The German .org domain story lands at the very end and is one of the more memorable closing moments of the BrightonSEO podcast run.
Head of SEO
As Head of SEO at Seeders, Hans helps companies scale smarter by combining search behaviour, paid performance, and conversion insights into systems that drive real growth. Recognised as a Top 20 Content Marketer by Small Business Trends, he brings a lean, experimental approach focused on building what actually works. Rather than relying on reports or one-off strategies, Hans focuses on creating marketing systems and tools that turn visibility into measurable business impact. His goal is simple: help clients move from scattered efforts to structured growth.
