AI agents went shopping for TVs and laptops—and messed up nearly 50% of the time, study shows

Gemini's shopping‑agent experiment, in which the Gemini AI successfully completed TV purchases 48% of the time, highlights both the promise and the pitfalls of autonomous e‑commerce bots. For website owners, this study signals a surge in AI‑driven traffic that mimics genuine shopper behavior—browsing product pages, adding items to carts, and even completing checkout flows. Because Gemini and Muse agents can navigate complex catalogues, sites may see spikes in session counts, abandoned carts, and atypical purchase patterns that could distort analytics, strain inventory systems, or trigger fraud alerts. Moreover, the near‑50% success rate means half of these bots will fail mid‑process, generating error logs and potentially inflating bounce rates. **Actionable tips:** 1. **Update your llms.txt** to explicitly disallow autonomous purchasing agents (e.g., `User-agent: Gemini-Agent\nDisallow: /checkout`). This prevents bots from completing transactions while still allowing harmless crawling. 2. **Deploy bot‑tracking scripts** that flag non‑human navigation signatures—such as rapid page transitions, uniform mouse movements, or missing referrer data—and route them to a sandbox environment for analysis. 3. **Monitor conversion funnels** with separate dashboards for AI‑identified traffic versus human visitors, enabling you to adjust inventory alerts and fraud detection thresholds in real time.