Fewer Than 1% of Product Pages Are LLM-Ready: Mirakl’s Meadon
The storefronts of the internet are, by and large, invisible to the AI agents now doing the shopping. That is the blunt assessment of one executive who has run the numbers
The findings are stark: in a survey of 500 product pages, fewer than 1% scored an 80% LLM-readiness rating.
“Consumers, we know, are very willing to research and shop on LLMs,” said Darius Meadon, CMO of Mirakl, in this video interview with Beet.TV. “Businesses have struggled to really understand this new paradigm of how to make their products discoverable and shoppable via LLMs.”
The order of operations problem
Meadon said retailers are mismanaging their AI rollouts. The foundational data infrastructure has to come first, he said. Everything else, the applications, the personalization, the agentic commerce layers, depends on it.
“There’s no order of operation at the moment,” Meadon said. “The order is everything all at once. So I think there’s definitely that strategic, like really making sure you have a clear road map and starting with the big, sometimes unsexy work before we can get to the sexy applications of AI.”
The second misalignment he identified is one of talent. There is a persistent myth, he argued, that AI capability can be democratized across an entire organization. In reality, cutting-edge AI still requires genuine specialists, and the trick is knowing where deep AI expertise is needed versus where existing domain knowledge, the people who already know the products and systems, is the more valuable asset.
“A combination of those two, technical talent and real on-the-ground business knowledge,” he said, “is what’s going to be successful in the future.”
What LLMs actually need to transact
Mirakl is a software company that provides an enterprise-level SaaS platform. It lets retailers, brands, and B2B organizations build and run their own online marketplaces and dropship programs, allowing them to sell third-party products without holding extra stock.
Mirakl’s own GEO Readiness Analyzer, a proprietary tool the company used to assess the submitted URLs of 500 product pages, produced a finding that should concentrate minds across retail. The sub-1% readiness figure points to three systemic deficiencies: weak metadata, insufficient intent data, and an absence of trust signals.
LLMs, Meadon explained, do not weigh all information equally. A product page needs rich descriptive data, color, cut, size, shipping timelines, to enable matching. But it also needs to capture the intent behind a purchase, not just what a product is, but why someone might want it. And then there are the trust cues: reviews on Trustpilot, Reddit discussions, YouTube commentary. “LLMs don’t just look at all information, doesn’t weigh all information as the same,” he said. “So yes, it needs the intent data to match the query to the product.”
The commercial stakes are escalating rapidly. Gartner projects worldwide AI spending will reach $2.59 trillion in 2026, a 47% year-on-year increase, driven heavily by infrastructure investment. For retailers, the implication is that the window to get the foundations right is narrowing as competitors pour capital into the space.
Agentic commerce and the cost of waiting
Mirakl’s response to the readiness gap is a product launched in April 2026 called Agentic Activation, which Meadon described as doing two things:
- connecting enterprise retailers directly to LLMs via a continuous data stream.
- ensuring product data is formatted and expansive enough not just for discovery but for transaction.
The company has also announced a partnership with J.P. Morgan Payments aimed at enabling secure, seamless transactions within agentic commerce environments.
The product also incorporates what he described as a GEO layer, ensuring that shipping dates, order routing, and logistics metadata are present so that a consumer asking an AI agent for next-day delivery can actually get it.
The storefronts of the internet are, by and large, invisible to the AI agents now doing the shopping. That is the blunt assessment of one executive who has run the numbers
The findings are stark: in a survey of 500 product pages, fewer than 1% scored an 80% LLM-readiness rating.
“Consumers, we know, are very willing to research and shop on LLMs,” said Darius Meadon, CMO of Mirakl, in this video interview with Beet.TV. “Businesses have struggled to really understand this new paradigm of how to make their products discoverable and shoppable via LLMs.”
The order of operations problem
Meadon said retailers are mismanaging their AI rollouts. The foundational data infrastructure has to come first, he said. Everything else, the applications, the personalization, the agentic commerce layers, depends on it.
“There’s no order of operation at the moment,” Meadon said. “The order is everything all at once. So I think there’s definitely that strategic, like really making sure you have a clear road map and starting with the big, sometimes unsexy work before we can get to the sexy applications of AI.”
The second misalignment he identified is one of talent. There is a persistent myth, he argued, that AI capability can be democratized across an entire organization. In reality, cutting-edge AI still requires genuine specialists, and the trick is knowing where deep AI expertise is needed versus where existing domain knowledge, the people who already know the products and systems, is the more valuable asset.
“A combination of those two, technical talent and real on-the-ground business knowledge,” he said, “is what’s going to be successful in the future.”
What LLMs actually need to transact
Mirakl is a software company that provides an enterprise-level SaaS platform. It lets retailers, brands, and B2B organizations build and run their own online marketplaces and dropship programs, allowing them to sell third-party products without holding extra stock.
Mirakl’s own GEO Readiness Analyzer, a proprietary tool the company used to assess the submitted URLs of 500 product pages, produced a finding that should concentrate minds across retail. The sub-1% readiness figure points to three systemic deficiencies: weak metadata, insufficient intent data, and an absence of trust signals.
LLMs, Meadon explained, do not weigh all information equally. A product page needs rich descriptive data, color, cut, size, shipping timelines, to enable matching. But it also needs to capture the intent behind a purchase, not just what a product is, but why someone might want it. And then there are the trust cues: reviews on Trustpilot, Reddit discussions, YouTube commentary. “LLMs don’t just look at all information, doesn’t weigh all information as the same,” he said. “So yes, it needs the intent data to match the query to the product.”
The commercial stakes are escalating rapidly. Gartner projects worldwide AI spending will reach $2.59 trillion in 2026, a 47% year-on-year increase, driven heavily by infrastructure investment. For retailers, the implication is that the window to get the foundations right is narrowing as competitors pour capital into the space.
Agentic commerce and the cost of waiting
Mirakl’s response to the readiness gap is a product launched in April 2026 called Agentic Activation, which Meadon described as doing two things:
- connecting enterprise retailers directly to LLMs via a continuous data stream.
- ensuring product data is formatted and expansive enough not just for discovery but for transaction.
The company has also announced a partnership with J.P. Morgan Payments aimed at enabling secure, seamless transactions within agentic commerce environments.
The product also incorporates what he described as a GEO layer, ensuring that shipping dates, order routing, and logistics metadata are present so that a consumer asking an AI agent for next-day delivery can actually get it.
Series Videos
