The AI Discovery Gap Is Widening. Most Merchants Are on the Wrong Side.
Profound just published the most granular analysis of ChatGPT Shopping I’ve seen. They tracked 260 million prompt-level results across 13,000 product categories and measured 22.5 million individual buy offers over ten days in March 2026. The findings are uncomfortable for anyone who’s been telling their board that AI shopping is about to replace Google. But they’re also a clear signal — maybe the clearest yet — about where the real opportunity is.
ChatGPT Doesn’t Know You Exist
Profound just published the most granular analysis of ChatGPT Shopping I’ve seen. They tracked 260 million prompt-level results across 13,000 product categories and measured 22.5 million individual buy offers over ten days in March 2026.
The findings are uncomfortable for anyone who’s been telling their board that AI shopping is about to replace Google. But they’re also a clear signal — maybe the clearest yet — about where the real opportunity is.
Let me break down what they found, and what it means if you’re an independent merchant.
𝗧𝗵𝗲 𝘀𝘂𝗿𝗳𝗮𝗰𝗲 𝗶𝘀 𝗻𝗮𝗿𝗿𝗼𝘄. 𝗩𝗲𝗿𝘆 𝗻𝗮𝗿𝗿𝗼𝘄.
ChatGPT’s shopping feature activated on roughly 9% of all prompts tracked. That’s the average. Underneath it: 79% of prompts never triggered shopping across any run over nine months. Only 0.72% triggered every single time.
This isn’t a channel you can spray and pray. It’s a narrow, specific surface that activates under specific conditions.
What conditions? Category beats intent — by a lot. Naming a shippable physical product is a 2–6x lift over baseline. Saying “I want to buy” without a product noun barely moves the needle. Apparel prompts triggered at 62%. Physical products broadly at 56%. Software and SaaS? 0.1%. Services? Zero.
The simplest heuristic Profound offers: ask whether the head noun in the prompt describes something you’d find listed on Amazon. If yes, shopping will probably activate. If the noun describes a service or a concept, it won’t.
Good news if you sell physical goods. That’s the surface that’s live and active today.
𝗧𝗵𝗲 𝗰𝗮𝗿𝗼𝘂𝘀𝗲𝗹 𝗿𝗲𝘀𝗵𝘂𝗳𝗳𝗹𝗲𝘀 𝗼𝗻 𝗲𝘃𝗲𝗿𝘆 𝗿𝘂𝗻.
This is the finding that should change how every brand thinks about AI commerce optimization.
95% of product titles appeared in fewer than 30% of runs for the same prompt. Only 0.5% of products achieved 70%+ consistency. A brand that shows up today may not show up tomorrow for the identical query.
This is not Google. There’s no fixed ranking you can optimize once and hold. It’s a probabilistic surface — one that favors merchants whose product data survives the model’s selection process on any given run.
That last sentence matters. I’ll come back to it.
𝗪𝗮𝗹𝗺𝗮𝗿𝘁 𝗮𝗻𝗱 𝗧𝗮𝗿𝗴𝗲𝘁 𝗮𝗿𝗲 𝘁𝗵𝗲 𝗻𝗲𝘄 𝗴𝗮𝘁𝗲𝗸𝗲𝗲𝗽𝗲𝗿𝘀. 𝗔𝗺𝗮𝘇𝗼𝗻 𝗶𝘀 basically 𝗮𝗯𝘀𝗲𝗻𝘁.
Walmart leads rank-1 buy links at 8.78% of all first-position offers. Target leads total presence across all positions at 7.16%. Together, the top 20 merchants capture roughly 40% of all offers.
Amazon is conspicuously underrepresented — almost certainly because it blocked OpenAI’s web crawlers.
This leads to a comparison I think is exactly right: Walmart and Target are the Reddit and Wikipedia of ChatGPT Shopping. They don’t dominate because ChatGPT prefers their products. They dominate because their catalog coverage and platform integrations put them in front of the model more reliably than anyone else.
Independent merchants on platforms like Shopify — close to 5 million of them — are not in the top 20. They’re not in the top 100. Right now, they’re largely invisible on this surface.
That’s the problem worth solving.
𝗧𝗵𝗲 𝗰𝗵𝗲𝗰𝗸𝗼𝘂𝘁 𝗺𝗲𝗰𝗵𝗮𝗻𝗶𝘀𝗺 𝗳𝗮𝗶𝗹𝗲𝗱. 𝗧𝗵𝗲 𝗱𝗶𝘀𝗰𝗼𝘃𝗲𝗿𝘆 𝗺𝗲𝗰𝗵𝗮𝗻𝗶𝘀𝗺 𝗶𝘀 𝘄𝗼𝗿𝗸𝗶𝗻𝗴.
OpenAI recently pulled the plug on Instant Checkout and pivoted to retailer-controlled apps. Walmart’s own EVP confirmed to WIRED that conversion rates for products sold directly inside ChatGPT were three times lower than those requiring a click-out to Walmart’s site.
But here’s the stat that should stop you cold: ChatGPT is now bringing Walmart new customers at roughly twice the rate of search engines.
The checkout button flopped. The discovery engine is outperforming every other acquisition channel Walmart runs.
OpenAI isn’t retreating from commerce. It’s concentrating on the part that actually works: being the place where consumers decide what to buy, before they go somewhere else to buy it.
𝗪𝗵𝗮𝘁 𝗶𝗻𝗱𝗲𝗽𝗲𝗻𝗱𝗲𝗻𝘁 𝗺𝗲𝗿𝗰𝗵𝗮𝗻𝘁𝘀 𝘀𝗵𝗼𝘂𝗹𝗱 𝗱𝗼 𝗿𝗶𝗴𝗵𝘁 𝗻𝗼𝘄
The article’s recommendation for brands is direct: “Invest in structured, machine-readable product data. The carousel favors brands whose data survives the model’s selection process on any given run.”
That’s the whole game. The carousel reshuffles constantly — but it consistently surfaces products it can understand, describe, and confidently recommend. Merchants with clean, enriched, structured catalog data have a structural advantage on every single run.
This is what we built catalogGPT to do. Not as a one-time setup — because a dynamic surface requires ongoing optimization — but as a continuous layer that keeps your inventory legible and surfaceable to AI agents as the models evolve.
The data points to a new metric worth tracking: how often your products appear in the carousel across the queries that matter — call it your shopping offer rate. It’s more actionable than citation rate and more directly tied to commerce outcomes. It’s the KPI we’re building around.
𝗧𝗵𝗲 𝘄𝗶𝗻𝗱𝗼𝘄 𝗶𝘀 𝗼𝗽𝗲𝗻, 𝗯𝘂𝘁 𝗻𝗼𝘁 𝗳𝗼𝗿 𝗹𝗼𝗻𝗴
The old digital shelf had fixed positions you could track, optimize, and defend. This one reshuffles every time someone asks a question.
That volatility is uncomfortable. But the closing line is worth sitting with:
“A surface this young, with rules this learnable, rewards whoever shows up first with the right data architecture.”
Walmart is already pulling twice as many new customers from ChatGPT as from search. That’s not a pilot stat. That’s a channel declaring itself.
For independent merchants, the question isn’t whether to pay attention to this. The question is how fast you can move before the concentration hardens and the top 20 becomes impossible to crack.
𝘑𝘰𝘦 𝘔𝘰𝘯𝘢𝘴𝘵𝘪𝘦𝘳𝘰 𝘪𝘴 𝘵𝘩𝘦 𝘧𝘰𝘶𝘯𝘥𝘦𝘳 𝘢𝘯𝘥 𝘊𝘌𝘖 𝘰𝘧 𝘷𝘪𝘴𝘶𝘢𝘭𝘈𝘐, 𝘣𝘶𝘪𝘭𝘥𝘪𝘯𝘨 𝘵𝘩𝘦 𝘮𝘦𝘳𝘤𝘩𝘢𝘯𝘵 𝘤𝘰𝘮𝘮𝘦𝘳𝘤𝘦 𝘖𝘚 𝘧𝘰𝘳 𝘵𝘩𝘦 𝘢𝘨𝘦𝘯𝘵𝘪𝘤 𝘦𝘳𝘢 — 𝘴𝘩𝘰𝘱𝘱𝘦𝘳𝘎𝘗𝘛 (𝘵𝘳𝘪𝘮𝘰𝘥𝘢𝘭 𝘈𝘐 𝘴𝘦𝘢𝘳𝘤𝘩), 𝘤𝘭𝘦𝘢𝘯𝘦𝘳𝘎𝘗𝘛 (𝘤𝘢𝘵𝘢𝘭𝘰𝘨 𝘦𝘯𝘳𝘪𝘤𝘩𝘮𝘦𝘯𝘵), 𝘢𝘯𝘥 𝘤𝘢𝘵𝘢𝘭𝘰𝘨𝘎𝘗𝘛 (𝘎𝘌𝘖 𝘰𝘱𝘵𝘪𝘮𝘪𝘻𝘢𝘵𝘪𝘰𝘯 𝘧𝘰𝘳 𝘈𝘐 𝘢𝘨𝘦𝘯𝘵𝘴).
𝘋𝘦𝘮𝘰: 𝘴𝘩𝘰𝘱𝘱𝘦𝘳𝘨𝘱𝘵.𝘷𝘢𝘪𝘳𝘦𝘵𝘢𝘪𝘭.𝘤𝘰𝘮