What actually happens when an AI shopping agent reads a store it can't parse — and why the failures never show up in your analytics.
50 million shopping queries a day already go through ChatGPT alone — and the number is growing fast. US AI-mediated retail spending is forecast to reach $144 billion by 2029, up from $20.9 billion in 2026 (eMarketer). Morgan Stanley estimates AI agents could account for 10–20% of all US e-commerce by 2030. This isn't a future channel. It's happening now, and it's scaling.
But the experience is broken. When a customer asks an AI assistant to find a product, the agent searches stores it can read — and on most stores, it can barely read anything. It sees the raw page, not the live site. It can't execute JavaScript, so it misses prices, stock levels, and size availability. What happens next isn't that the agent recommends your competitor. It's worse.
The agent recommends your products with links that don't work. The URL has changed, the product's been discontinued, or the page requires JavaScript to render. The customer clicks through and gets a 404 or an empty page. They don't come back.
It recommends products that are sold out. The agent can't check real-time stock, so it shows a dress that's been out of stock for weeks. The customer clicks, sees "sold out", and loses trust — in the agent and in your store.
It recommends the wrong thing entirely. Without structured size, colour, and variant data, the agent can't match what the customer actually asked for. They asked for a black one-piece swimsuit in a size 12, and the agent shows them a blue bikini in a size 6 — because that's all it could read from your unstructured page.
It gets your prices wrong. Your product feed says £49.99 but the page says £54.99. The customer feels misled. The agent quietly deprioritises your store for future queries. You never know it happened.
It invents your policies. The agent tells the customer you offer 2-day delivery when your actual policy is 5–7 days. The customer orders, waits, and complains. Your customer service team handles a problem that started with missing data.
None of this shows up in your analytics. An AI agent fires roughly 6 API calls versus the 40+ browser events a human visit generates. Around 70% of AI-referred traffic gets misclassified as "direct" in Google Analytics. You can't measure a channel you can't see.
Catalens One shows you exactly what's going wrong. It reads your store the way an agent does and shows you what's really there — and what agents silently skip. No AI in the loop. Deterministic checks against your public pages and feeds. Nothing to sign up for, nothing billed.
Paste a URL, get a scored, product-by-product audit — with the exact fixes, ranked by how many points each one is worth.
Run a free audit →Developed by Hania B. Demai at Viver Studio Ltd.
Sources: NBER working paper 34255 (Sept 2025, ~2% of
ChatGPT's ~2.5B daily messages are shopping-related) · eMarketer (Dec
2025) · Morgan Stanley Research (Dec 2025) · GA4 attribution analysis
(~70% of AI referrals misclassified as direct) · Kaiser & Schulze,
SSRN 5585812 (Oct 2025).