How Amazon's Rufus is rewriting product discovery; what really killed the aggregators
A new deep dive into Amazon's Rufus argues the assistant is not just a chatbot but a shopping companion that plans, remembers and researches, and that understanding it is now essential for any seller wanting to remain visible.
Seven shifts stand out. Search is becoming mission-based: instead of a list of headphones, Rufus builds a comparison of five brands with prices, reviews and expert opinion. It plans multi-step projects like parties and home offices, so products need to fit bigger stories. It remembers each customer's interests, family and past purchases. It checks facts in real time, pulling current prices, stock and expert coverage from Amazon and major review sites. Context beats keywords. Bundles are prioritized because Rufus plans complete solutions. And outside opinions, such as best-of lists and buying guides, feed its recommendations.
The analysis frames three pillars behind the assistant: reasoning that compares options and explains choices, memory that personalizes suggestions, and real-time sourcing that keeps answers current.
For sellers, the advice centers on Reasoning Path Optimization, which works alongside keyword research rather than replacing it. Keywords tell Rufus what a product is; mission context, such as which backpacks a tent fits or which grinders pair with an espresso machine, keeps it in the recommendation. Listings are urged to signal what comes next, make bundling obvious, work for multiple situations, use lifestyle language and focus on outcomes over features.
New seller statistics show the shifting competitive backdrop: Chinese sellers went from 7 percent of new Amazon.com signups in 2015 to 62 percent in 2024, while the US share fell to 27 percent, and Chinese businesses now account for one in two active sellers globally, though US sellers still generate more revenue per store.
In a candid industry interview, a Thrasio co-founder offered the first insider account of the aggregator collapse. When Thrasio started, brands changed hands at two times EBITDA; by 2021 more than 100 aggregators had flooded the market with 12.3 billion dollars, three quarters of it debt, pushing multiples to seven times for weak assets. Speed made it worse: Thrasio ballooned from four founders to 1,600 staff spread over 10 offices within two and a half years, acquiring 200 brands, and one inventory ordering mistake, in which a sample exercise was mistakenly placed as real purchase orders across nearly the entire catalog, cost hundreds of millions of dollars.
The model itself was not the problem, the account argues: software aggregators such as Threecolts, which raised 200 million dollars, and Carbon6, sold for 210 million, thrived with the same consolidation playbook. With more than 100,000 Amazon sellers now generating over 1 million dollars a year, acquirers are hunting operational excellence rather than raw revenue.
Google, meanwhile, appears close to launching an agentic Buy for Me checkout in which its AI agent completes purchases on merchant websites after a tracked price drops, using saved Google Pay details. New product card layouts, updated track price messaging and a fresh help page signal the launch is imminent, though the price-drop trigger and Google Pay requirement, accepted by only about a tenth of online merchants, will limit early adoption.
Separately, a quick-start guide to the Amazon Influencer Program is circulating: creators post daily on TikTok for 30 days, submit three demo review videos, then produce short product videos for Amazon pages. Standard commissions run about 2 percent, but Creator Connections brand campaigns pay 10 to 50 percent. Realistic earnings are one to two dollars per video per month after three months, making volume the game. And in consumer behavior research, price ranks as the top factor in purchase decisions, ahead of product photos in second and reviews in fourth.
