Seven months in the past, Amazon’s AI assistant figuring out her procuring habits felt uncanny. Commerce media columnist Kiri Masters on what modified — and what it means for each different retail media community.
Seven months in the past, I requested Amazon’s Rufus a easy query: “What sort of particular person am I in actual life?”
Rufus instructed me I play pickleball, have youngsters, personal a cat, preserve a complete skincare routine, and crave costly Japanese tenting gear. I printed the outcomes and watched dozens of individuals in my community strive the identical factor over the following 24 hours. To their dismay, Amazon appeared to crack down didn’t instantly. Rufus began gaslighting clients, claiming he didn’t actually care about them in any respect. A number of connections commented that they could not get Rufus to provide something related. I confirmed it — even I may replicate my very own outcomes inside days of publishing that piece.
I can not declare that my article triggered the pullback. However the timing was uncanny.
Right this moment, if you open Rufus, Amazon truly suggests you ask about your procuring habits. The identical functionality that spooked somebody internally seven months in the past is now a product function, full with person controls. Consdon’t can ask “what have you learnt about me?” and proper something they do not need used for personalization.
That was a strategic selection. And it has implications for each retail media community watching from the sidelines.
From creepy to managed
An Amazon spokesperson instructed me: “Amazon has provided customized procuring suggestions for years, and we proceed to innovate to make the expertise much more useful for patrons. Right this moment, practically each procuring expertise throughout Amazon, together with Rufus, is customized to assist a whole bunch of tens of millions of shoppers discover precisely what they need and wish.”
Amazon has moved Rufus from a device that would floor what it knew about you — when you requested the fitting approach, and caught it in the fitting temper — to at least one that formalizes reminiscence as a function. Customers can examine it. They’ll edit it. And Amazon is betting that sense of management will make folks extra comfy counting on it.
The analysis means that guess is well-placed, although the connection is extra nuanced than “reminiscence equals belief.” A 2025 examine from the College of Basel discovered that when a chatbot addressed customers by title and referred to earlier conversations, members noticed it as extra competent and extra benevolent. They had been extra keen to make use of it and share info. However total belief didn’t rise considerably – integrity and transparency mattered extra. Reminiscence-like habits seems to extend reliance earlier than it totally will increase belief. Amazon appears to have absorbed this lesson.
From recommendation to motion
The larger change is that Rufus now does issues.
Seven months in the past, Rufus was a search device with character. Right this moment, it could actually add objects to your cart, reorder previous purchases, set value alerts, and routinely purchase merchandise after they hit a goal value. I’ve been utilizing auto-buy for a cereal my son loves that’s exhausting to search out via regular channels. Due to Amazon’s market mannequin, the value swings wildly — $5.75 one week, $11.00 the following. Auto-buy watches the value and pulls the set off when it drops under my threshold. It’s a real enchancment over Subscribe & Save, which locks you right into a schedule no matter whether or not the value is sweet. Amazon says clients utilizing auto-buy save a mean of 20% per buy.
Wish to go deeper? Ask The Drum
That is the model of agentic commerce that works. Amazon owns all the stack beneath it. There’s no messy product knowledge to normalize, no gross sales tax to determine, no checkout integration to construct from scratch. Rufus can act as a result of Amazon already controls {the marketplace}, funds, and success that make motion potential.
An information hole stays for advertisers
I acquired curious not too long ago concerning the depth of Rufus’s reminiscence and determined to push additional than my authentic experiment. With some coaxing, Rufus retrieved each buy I’d made since I first opened my Amazon account in 2011. Fourteen years of transaction historical past — and an enchanting journey down reminiscence lane. Once I opened that account, I’d simply moved to New York from Australia. The purchases from these early years confirmed me a very totally different particular person: totally different wants, totally different tastes, totally different life stage. We’re what we purchase, and Rufus can see the complete arc.
I wasn’t shocked that Amazon had this knowledge — it’s tied to precise buy data. However advertisers paying to achieve me via that very same platform are working with a a lot narrower slice.
Amazon’s present new-to-brand definition appears to be like again 12 months. Walmart makes use of an analogous one-year lookback for some reporting like Sponsored Movies. However these are comparatively current expansions, and the trade normal is far and wide. Goal’s Roundel has acknowledged that “new to model” may imply 13 weeks at one community and 13 months at one other.
So the AI assistant sees 14 years of my procuring habits. The advertiser paying to achieve me via that very same platform may see twelve months, at finest. As Rufus turns into the interface for a rising share of Amazon procuring classes, that hole begins to compound. The AI is making suggestions and focusing on choices knowledgeable by years of habits the advertiser can’t see, can’t bid towards, and may’t independently confirm. Advertisers are more and more reliant on Amazon’s AI to search out the fitting shopper on the proper second — with much less visibility into how these choices get made.
What this implies for Amazon Adverts
Amazon has demonstrated, with actual scale, that proudly owning the AI procuring interface creates a brand new promoting floor. Greater than 300 million clients used Rufus in 2025, and the corporate says it contributed practically $12bn in incremental annualized gross merchandise worth. Amazon has launched Sponsored Merchandise and Sponsored Model prompts inside Rufus conversations. I wrote not too long ago about how companies are testing these early placements — discovering notable engagement however virtually no measurement infrastructure but. Amazon can also be pushing Rufus past its personal catalog — a function known as Purchase For Me lets the assistant full purchases on third-party model websites utilizing your Amazon cost particulars. This extends the floor space even additional.
Analysis from the Mars Company discovered that solely 22% of merchandise rating on web page one among Amazon search additionally appeared in Rufus outcomes. That’s recommending totally different merchandise based mostly on totally different logic. If that is the place procuring classes are heading, the key phrase bidding that construction manufacturers spent a decade constructing is geared toward a system that is already being changed.
The place that leaves everybody else
Amazon’s multi-category breadth provides them a bonus in understanding the entire shopper — although they’ve by no means actually cracked grocery, which is a niche. And specialist retailers have grounds to compete, together with assortment depth and area experience thata generalist cannot match. However the sample right here is acquainted: Amazon has an amazing head begin on a shopper expertise that’s quick changing into a traditional a part of how folks store. And as soon as once more, most retailers are behind.
The risk is now two-sided. Third-party AI assistants like ChatGPT and Gemini are intercepting Amazon’s product discovery earlier than customers attain retailer web sites. And Amazon’s personal assistant is locking in customers on the opposite finish, constructing persistent relationships that increase switching prices. The playbook I have been arguing for — offsite activation, knowledge collaboration, bodily retailer funding, product fact as a media asset — will get extra pressing.
Now what?
Seven months in the past, the attention-grabbing query was whether or not Rufus remembered me. Now the query is whether or not the remainder of the trade can construct something that provides Amazon related leverage over the patron relationship — or whether or not Amazon’s head begin in AI-assisted commerce turns into one other strategic benefit that retains compounding.
Most retailer AI assistants I’ve examined previously yr fall someplace between forgettable and irritating. Rufus is neither. And that hole is widening.
Source link
