🛠️ Recommended for you

Check this awesome 79-page document breaking down everything Rufus really loves, by Oana Padurariu, Danny McMillan and Andrew Bell. You’ll understand why Rufus loves:

⚡ Conversational Q&A ⚡Customer Reviews ⚡Inference⚡Noun Phrases⚡Structured Content⚡Text on Images⚡The Customer Journey.

🛑 Attention Amazon Sellers

Amazon’s search may get smarter in the future with a new LLM-powered autocomplete.

Amazon recently live A/B tested a new autocomplete system for its search bar that blends traditional ranking models with the power of Large Language Models (LLMs). The experiment is explained on Amazon Science’s publication “Evaluating auto-complete ranking for diversity and relevance”.

While the standard models rely on user search logs to suggest popular queries, LLMs can generate a wider range of diverse and relevant suggestions even for products or keywords that haven’t been searched before. The team tested various prompt strategies, with the “setwise” method showing the best balance between relevance and diversity.

In a live A/B test across Amazon’s search interface, the hybrid system (LLMs generating candidate suggestions offline + traditional models re-ranking them) delivered a 0.13% uplift in revenue and improved query diversity by 15%. The system enhanced the customer experience by helping users discover new, niche, or long-tail products that typical query logs wouldn’t catch.

This signals a major shift in how product discovery works on Amazon. Instead of just reinforcing what’s already popular, the search system can now surface new listings, lesser-known brands, and seasonal or trend-driven products—creating more opportunities for visibility.

You may be familiar with the “Honey moon period” concept, also known as the “cold start service”, as stated by 2019’s Amazon patent “Machine learning based database query retrieval”. Well, so this could also be a boost for new products, new niches and new keywords, as the search suggest could feature them to help discoveravility and assess their interest based on CTR and sales.

Long story short: if Amazon uses this new autocomplete feature and you detect the boosted keywords, you could add them to your listings (of course, as long as your product aligns with the search intent) and profit from the traffic.

📰 What you need to know

👉 Loved “the $0.01 hack” by Jon Derkits and shared by Tyler Wallis

👉 Amazon halts orders for Chinese and Asian products amid tariff concerns, while Trump authorizes 90 day pause on tariffs

👉 This American flag is selling 7k units a month claiming to be 100% made in the USA, yet is China made and sold by a Chinese seller. Money goes back to China, untaxed. This is just one case of the thousands that can be found every day on Amazon.

I live in Spain and every day I come across faked EU products and/or Amazon sellers whose seller name includes "ES", "DE", "EU" and other words that make the customer think they are buying from a European company.

This is not new. When I started selling on Amazon around 2014, Chinese sellers were already gaming the platform.

👉 More shady tactics by Chinese Sellers. This is one of the wildest blackhat cases I’ve ever come across: 10 brands, 14+ seller accounts, 1 bad actor

👉 Amazon launched the “Buy for me” AI agent last week. An agent that allows Amazon to place orders on behalf of some select U.S. shoppers.

👉 Here’s a PO worth $4,707 in tariffs. After "Liberation Day" it'll be $101,673.

👉 Georank and why it matters to Amazon Sellers, especially if you sell in the US

💡 Pro Tip: Prime Day: Timing matters. And so does pricing

⚒️ Software & Services

As we do every week, here are 2 software / services for Amazon Sellers and Vendors.

• A couple of GPTs, this one by Ritu Java AMZ Listings with TOS Compliance

• This one by Andrew Bell: Amazon Optimization GPT, helps diagnose and enhance existing listings using real-time keyword trends.