AI commerce company Recomaze just ran the kind of test that should be circulated in every ecommerce Slack channel in existence. The research ran six purchase-intent queries against each of 9,720 ecommerce stores through Google Gemini – 58,320 tests in all.
In 60% of cases, the store was recommended for nothing. Across every test, a store was named just 14% of the time. The rest of the time, the assistant pointed the shopper to a competitor, or named no store at all.
Read that again. Fourteen percent. Six queries each. And 60% of stores.
This wasn’t a test of brand-name queries where the result is obvious. The queries were written to match each store’s category, the questions a shopper asks when they know what they want but not where to buy it, rather than brand-name searches that would be trivial to win. These were the exact moments an independent store should be able to compete on. And most of them are losing.
Size Doesn’t Save You
Here’s the part that actually matters, and the part that a certain type of agency will try to spin into a $5,000-per-month retainer: the winners weren’t the biggest brands.
“The stores that get recommended are not the biggest ones. They are the ones whose product information an engine can actually read and trust,” said Delian Coroamă, founder and CEO of Recomaze. “The shop window used to be Google. Now it is whatever the AI decides to recommend, and most brands have no idea whether they are in the answer.”
This is both terrifying and, weirdly, kind of democratic. Gemini doesn’t care about your ad spend. It doesn’t care that you’ve been around since 2009 or that you have a warehouse in Ohio. It cares whether it can parse your product data. If your catalog is a mess of thin descriptions, missing attributes, and titles that read like they were written by a sleep-deprived intern in 2017, the AI is just going to recommend someone else.
What “Readable” Actually Means
The pattern is consistent across every major AI platform making shopping recommendations. When Gemini writes your ad copy from your product data, your data quality becomes your ad quality. Clean, structured, complete product data produces sharp, query-matched copy. Thin or messy data produces generic summaries that convert worse.
It’s not about adding more words to your product descriptions. It’s about writing them in a way that lets a machine understand what problem your product solves, who it’s for, and why it’s the right answer to a specific question. Natural language. Clear attributes. Actual specificity instead of marketing fluff like “premium quality craftsmanship.”
Roughly 40% of US consumers now use generative AI tools for product research before making a purchase decision. That number is going up, not down. And only 14% of shoppers say they fully trust AI-generated recommendations enough to make a purchase right away, meaning Gemini recommending you is the start of the journey, not the end. But it’s a start you’re currently not getting.
This Is the Same Problem, Different Packaging
To be clear, this isn’t a new problem. It’s the same product data problem that’s been quietly compounding for years. The Arktic Fox study from earlier this year found that only 12% of retailers are fully confident in their product data being AI-ready, and a stunning 0% of brand manufacturers. The difference now is that the consequences are no longer hypothetical.
The stores losing Gemini recommendations today are the same ones who had thin category pages in 2018 and wondered why their organic traffic was flat. The medium changed. The underlying negligence didn’t.
Our Take
The Invisible Store Problem Is Already Here
The most unsettling thing about this study isn’t the 60% figure. It’s that most of the affected store owners have no idea. Their Google Analytics still shows traffic. Their paid campaigns still run. Nothing looks dramatically wrong. But a layer of discovery that’s quietly becoming more important by the month — the AI recommendation layer, has already written them off.
Google has launched agentic checkout across Google Search and Gemini, enabling autonomous AI agents to execute purchases directly on merchant websites. “Buy for me” functionality is live with selected retailers right now. The direction of travel is obvious: AI doesn’t just recommend, it’s starting to purchase. If you’re not in the recommendation set, you’re not in the purchase set either.
The fix isn’t glamorous. It’s writing better product descriptions. It’s completing your attribute fields. It’s making sure a machine can understand what you sell and who should buy it. None of that requires a big budget. It just requires doing the thing most stores have been putting off since the last time someone told them SEO was changing.
