227: ChatGPT Shopping Is Scraped Google Shopping with Malte Landwehr, CMO/CPO at Peec AI
Malte Landwehr
Peec AI
Chief Product Officer and Chief Marketing Officer at Peec AI, where his research traces how ChatGPT and other AI engines pull and rank the products they recommend.
Malte Landwehr spent five years as VP of SEO at Idealo, one of Europe's most SEO-dependent businesses, before he left to become Chief Product Officer and Chief Marketing Officer at Peec AI, a platform that tracks what ChatGPT, Claude, Gemini, and Google AI Overviews cite every day. His headline research, run with independent researcher Tom Wells, decoded how ChatGPT Shopping works: the data streamed to the browser carries a Base64-encoded blob of Google location parameters and Google Merchant Center feed fields, and the top 40 organic results in Google Shopping account for the products, ratings, reviews, and prices ChatGPT shows. The episode's framing is that ChatGPT Shopping is scraped Google Shopping, which means brands already ranking in Google Shopping start the agentic commerce race in front.
The strangest case Malte brought is GummySearch, a Reddit analytics tool that shut down in November after it failed to land an API deal with Reddit. It now sits behind roughly one in every thousand ChatGPT citations, because ChatGPT started injecting the word Reddit into its fan-out queries and GummySearch had listicle pages like 'best fashion brands according to Reddit' built from real Reddit review sentences. As Malte puts it, the company 'didn't gain visibility as a brand,' it gained 'power over what brands are recommended by LLMs.' He is equally blunt that clicks are the wrong metric. 'In a web search, clicking is part of the intended user journey,' he says. 'In an LLM, clicking is completely optional.' Citing Google's own B2B numbers, he argues AI now influences close to 100% of software buying decisions even while sending a website maybe 1% of its traffic.
From there the conversation turns to what actually moves AI visibility. Malte rates digital PR above technical SEO now, because grounding pulls from Reddit, G2, Wikipedia, YouTube, and news rather than your own website. He warns that ChatGPT runs at least one English fan-out query in 70 to 90% of cases, so US brands win answers in non-English markets unless local companies build a small English footprint. He maps the 'Mount AI' curve, where scaled AI content climbs fast then crashes, with one tool's reference customers splitting into 43% flat or growing, 22% Mount AI, and 35% that tanked outright. His closing argument is that SEO no longer works as a default growth channel, even though the practice itself is expanding into offsite work, and that the single most useful move is one citable paragraph near the top of a page: two or three declarative, self-contained sentences that name the entities, never an article shredded into one-line bullets.
KEY TAKEAWAYS
- Stop grading AI search by clicks. Clicking is optional inside ChatGPT, so an engine that sends 1% of your traffic can still shape close to 100% of buying decisions in your category. Track influence on the decision through branded search, direct traffic, and brand recognition.
- Get your Google Shopping feed and rankings right if you sell products. ChatGPT Shopping pulls its products, ratings, and prices from the top 40 organic Google Shopping results, so the same feed quality and ranking work that wins Google Shopping feeds ChatGPT's commerce answers.
- Treat offsite coverage as your main AI lever. Grounding pulls from Reddit, G2, Wikipedia, YouTube, and news, so a digital PR push that earns accurate third-party mentions moves how AI describes you more than another round of technical SEO.
- Build a small English footprint if your market is not English-speaking. ChatGPT runs at least one English fan-out query 70 to 90% of the time, so add an English About page and the occasional English podcast or campaign, but do not translate your whole website.
- Write one citable paragraph and put it near the top. Two or three declarative, self-contained sentences of about 50 words that name the entities give a model something to quote, which works far better than shredding an article into one-line bullets.
- Test AI-generated content against two questions before you scale it. Would a human visit your page instead of just asking ChatGPT, and would a Google or OpenAI product manager pay to crawl and rank it? If either answer is no, you are in spam territory and risk the Mount AI crash.
SHOW NOTES
From Idealo to Peec AI
Malte Landwehr ran SEO as VP at Idealo for five years, a business he describes as deeply SEO-dependent, which gave him room to study ChatGPT as a risk before it even had web search. A two-week trip to South Korea where he barely touched Google convinced him the shift to AI search would arrive bigger and faster than he expected. He joined Peec AI, then a five-person team, first as an advisor and then full time as Chief Product Officer and Chief Marketing Officer. Peec AI runs daily prompts across ChatGPT, Perplexity, Gemini, Google AI Overviews, Claude, and Grok and reports what each engine cites.
ChatGPT Shopping Is Scraped Google Shopping
The episode title comes from research Malte ran with independent researcher Tom Wells. The data ChatGPT streams to the browser contains a Base64-encoded blob with Google location parameters such as hl and uule and Google Merchant Center feed fields like headline offer ID and image ID. Stitching those together lands on a Google Shopping page with the same products, ratings, review counts, and prices. The top 40 organic results in Google Shopping account for everything ChatGPT Shopping displays, and the same trick does not work with Bing. A brand already ranking in Google Shopping starts ChatGPT's agentic commerce race in front, which makes Merchant Center feed quality a direct input into AI commerce.
GummySearch: A Dead Product With Citation Power
GummySearch, a Reddit analytics tool, shut down in November after it could not secure an API deal with Reddit. It now sits behind roughly one in every thousand ChatGPT citations. The cause is a change in how ChatGPT builds fan-out queries: it now slips the word Reddit into many of them, and GummySearch's old marketing pages, listicles like 'best fashion brands according to Reddit' built from real Reddit review sentences, are the perfect match. As Malte frames it, GummySearch 'didn't gain visibility as a brand. They now have power over what brands are recommended by LLMs.' He expects the effect to be temporary, but it shows how fragile and gameable citation sourcing still is.
Why Clicks Stopped Measuring Anything
Malte is blunt about the metric most teams still report. 'In a web search, clicking is part of the intended user journey,' he says. 'In an LLM, clicking is completely optional.' ChatGPT can register as 1% of measured traffic while shaping a far larger share of decisions. Citing Google's own B2B figures, he argues AI now influences close to 100% of software buying journeys, whether a manager asks ChatGPT for a shortlist or a procurement lead uploads an offer to Claude to check for cheaper alternatives. Measuring the channel by referral clicks misses almost all of that, the same gap that answer engine optimization is meant to close.
The English-Language Fan-Out Penalty
Malte's team found ChatGPT runs at least one English fan-out query in 70 to 90% of cases, depending on the country. English-language and US sources then steer the grounding process, so ChatGPT can recommend US or UK brands that do not even operate in a given market. His advice for non-English companies is not to translate everything, which would break their clean local index, but to build a small English footprint: an English About page with company facts, maybe one English content campaign a year, a CEO appearance on an English podcast or two. He warns the bias could vanish overnight, so it should take a small slice of time rather than become the strategy.
Mount AI and the Limits of Scaled Content
Malte is not against AI content, having overseen hundreds of thousands of AI pieces and millions of automated landing pages, but he offers two tests before scaling it. Would a human visit your page rather than just ask ChatGPT for the summary, and would a Google or OpenAI product manager spend money to crawl, index, and rank it? If either answer is no, you are in black hat spam territory. Price comparison summaries, live weather, and review roundups pass because they carry unique data; mass-produced 'best CRM for dentists' pages do not. He cites one AI content tool whose reference customers split into 43% flat or growing, 22% with the classic 'Mount AI' rise and crash, and 35% that tanked without ever climbing, alongside Lily Ray's analysis of more than 200 domains that lost visibility.
SEO Is Dead as a Default Channel
Malte's final argument is precise: SEO is dead as a default growth channel, not as a practice. Companies can no longer assume 10 to 20% organic growth a year because the total pie has stopped expanding. He would rarely tell a new company to go all in on SEO for its first 12 to 24 months, with exceptions for regulated or restricted industries and businesses sitting on unique data. The closing lesson ties back to citability and to the agentic web as a whole: write for extraction by leading with a declarative, self-contained paragraph that names the entities, because an LLM that finds the opening vague tends to give up on the rest of the page.
WATCH ON YOUTUBE
QUESTIONS ANSWERED
Does ChatGPT Shopping use Google Shopping data?
Research by Peec AI's Malte Landwehr and independent researcher Tom Wells found that ChatGPT Shopping pulls its data from Google Shopping. The response ChatGPT streams to the browser contains a Base64-encoded blob with Google location parameters and Google Merchant Center feed fields, and the top 40 organic Google Shopping results account for the products, ratings, reviews, and prices ChatGPT shows. The same reconstruction does not work with Bing Shopping, which is why Malte frames ChatGPT Shopping as scraped Google Shopping.
Why are clicks a bad way to measure AI search?
Malte Landwehr argues that clicking is optional inside an LLM, unlike web search where the click is the intended next step. ChatGPT can appear as around 1% of a website's measured traffic while influencing a much larger share of decisions. Citing Google's B2B research, he says AI now touches close to 100% of software buying journeys, so he recommends measuring branded search, direct traffic, and brand recognition rather than referral clicks.
How does a shut-down product like GummySearch still earn ChatGPT citations?
GummySearch, a Reddit analytics tool, shut down in November after failing to secure a Reddit API deal, yet it now sits behind about one in every thousand ChatGPT citations. Malte Landwehr explains that ChatGPT started inserting the word Reddit into its fan-out queries, and GummySearch's old listicle pages, built from real Reddit review sentences, became the perfect match. He notes the effect is likely temporary but shows how a dead product can gain power over which brands LLMs recommend.
Is digital PR more important than technical SEO for AI visibility?
Malte Landwehr still considers technical SEO important but now rates digital PR higher for AI visibility. Because LLMs ground their answers in third-party sources like Reddit, G2, Wikipedia, YouTube, and news, what others write about a company shapes how AI ranks, recommends, and describes it more than on-website technical work does. Your own website is one grounding source but rarely the biggest unless the prompt is branded.
Why do local brands lose to US brands in AI search?
Malte Landwehr's team found that ChatGPT runs at least one English-language fan-out query in 70 to 90% of cases, depending on the country. English and US sources then influence the grounding process, so ChatGPT can recommend US or UK brands that do not operate in a given market. His advice is not to translate the whole website but to build a small English footprint, such as an English About page and occasional English-language content.
What is Mount AI and how do you avoid the crash?
Mount AI describes scaled AI-generated content that climbs fast in visibility then crashes, forming a mountain shape on the chart. Malte Landwehr says you can avoid it by testing content against two questions: would a human visit your page instead of just asking ChatGPT, and would a Google or OpenAI product manager pay to crawl and rank it. He cites one AI content tool whose reference customers split into 43% flat or growing, 22% with the Mount AI shape, and 35% that tanked outright.
Is SEO dead?
Malte Landwehr's position is that SEO has stopped working as a default growth channel, even as the practice itself expands into offsite work. Companies can no longer assume 10 to 20% organic growth a year because the total search pie has stopped expanding, though winners still exist. For most new companies he would not recommend going all in on SEO for the first 12 to 24 months, with exceptions for regulated industries, restricted products, and businesses with unique proprietary data.
RELATED ARTICLES
You Can't See How AI Ranks You, So Build What It Can Read
Web strategy in the AI era means optimizing for systems you are not allowed to look inside. The UK's June 17 order forcing Google to explain how it ranks, AI Overviews included, is the first real crack in that opacity. The durable move does not wait for the box to open: build a website a machine can read, parse, and verify.
How to Measure AI Search Visibility
AI search visibility is not one number, and the metric most tools sell you, citations and mentions, is decoupled from the outcome you actually want, being recommended. Here is what to measure instead, why a single visibility score is noise, and how six practitioners and the 2026 data say to read it.
A German Court Made Google Liable For What Its AI Says About You
A German court ruled that Google's AI Overview is the company's own speech, not a neutral summary of other pages, and held it liable for false statements about two businesses. The lawsuit is not the real consequence. What matters is what an answer engine does once it can be sued for what it says, and why that makes machine-readable brand identity table stakes.
ENJOYING THIS EPISODE?
Practical strategies for making your website work for AI agents and the humans using it. Read by SEOs, developers, and AI researchers. Exclusive tools, free for subscribers.