A shopper looking for a gift used to type "truffle hot sauce" and scroll. Now a growing number of them open the assistant and type a paragraph.
"Hot sauce for my brother as a gift, he likes heat but needs low sodium."
The query has stopped being a keyword and started being a situation, and the response is a shortlist rather than a page of results. If your listing doesn't contain the answer to the constraints inside that question, you aren't on the list, and you'll never see the search that excluded you.
What is Amazon Rufus?
Rufus was Amazon's shopping assistant, built into the app and site. A shopper asks a question in ordinary language and it answers, recommends products, and handles follow-ups the way a conversation does. In May 2026 Amazon folded it into Alexa, and in the Amazon store it now appears as Alexa for Shopping. Most brands we talk to still call it Rufus, and so do I half the time. I'll use both names here.
The mechanic that matters to a brand is the narrowing. Traditional search returns a page and lets the shopper do the filtering. An assistant does the filtering first and returns a handful. Being on page one used to be the goal. Being in the shortlist is a smaller target.
It's also big enough now that it isn't a side channel. On Amazon's third-quarter 2025 earnings call, Andy Jassy said Rufus was "expected to generate over $10 billion in annual incremental sales," and Amazon reported around 250 million shoppers had used it that year. Those are Amazon's numbers about its own product, and I'd read them as Amazon's case for the assistant. They still tell you where the company is putting the shopping experience.
Is Amazon Rufus discontinued?
Amazon retired the Rufus name on May 13, 2026 and moved the assistant into Alexa for Shopping, and the work you'd do on a listing for Rufus carries straight over.
When I did a sweep of what food brands were asking about Amazon this summer, AI search was on the list, and my shorthand for it was "it was Rufus, now it's Alexa." Amazon describes the change as bringing Rufus and Alexa together into one assistant with shared memory. For a shopper, the assistant now remembers conversations across the phone, the laptop and Echo devices, and it can hold onto details like family members, pets, interests and dietary needs.
The surfaces changed more than the name did. According to Amazon's announcement, shoppers can now ask a question straight into the main Amazon search bar instead of opening a separate chat window. The assistant writes AI overviews at the top of search results and on product detail pages. A shopper can select several products from the results and get them compared side by side on features, prices and reviews. And there's a price history view covering up to a year on hundreds of millions of products.
Each of those is a place your listing gets summarized by something other than your own copy. The overview on your detail page is built from what's on it. The comparison pulls your features and your reviews into a table next to a competitor's. If the attribute that makes your product the right answer isn't stated on the page, it can't appear in either one.
What does the AI read?
From what we can see across accounts, it's reading what lives on Amazon.
Your title and bullets, which is where the plain facts about the product have to be. Your A+ content, which is a larger body of text and imagery than most brands treat it as. The text inside your images, which surprises people, because plenty of brands put their most useful claims into an infographic and assume it's decoration. Your reviews, in aggregate and in what they say. Your returns behavior. Your customer question section.
What we haven't seen is a brand get credit on Amazon for something that only lives on its own website. Your blog, your press coverage and your off-Amazon search authority don't appear to transfer. That's worth knowing because a brand with a strong content operation may assume it has an advantage here, and on this surface we haven't seen one.
I'll add one caution to that, because Amazon's own description has moved. Its Alexa for Shopping announcement says the assistant combines product knowledge with "in-depth information from across the web," and its shopping guides compare products "across Amazon and the web." So I won't tell you the web is invisible to it, only that your Amazon page is the part of this you fully control.
The practical consequence: anything true about your product that only exists on your own website is invisible to the assistant. Low sodium, third-party tested, made in a nut-free facility, safe for a specific diet. If it isn't on the page, it can't be matched to the question.
The test I use is simple. Could a customer make the buying decision from the image carousel alone? From the A+ content alone? From the title and bullets alone? Each of those sections gets read on its own, by shoppers who skip the rest and, as far as we can tell, by an assistant pulling one piece to answer one question. Lifestyle photography is the usual gap. A good lifestyle shot makes a great base for an infographic, and on its own it still doesn't tell the customer what they need to know to buy. We go through that work section by section in our guide to fixing an Amazon listing.
The 4.0 problem
We flagged this in the piece on Amazon listings and it's worth the fuller treatment, because it's the one I'd fix first if it applies to you.
We see AI answers skip products sitting under a 4.0 review average with some consistency. Not always, and Amazon doesn't publish a rule, so treat it as an observed pattern rather than a documented threshold. The pattern is consistent enough that I'd plan around it.
Think about what that changes. A 3.8-star product has always had a conversion problem, and brands treat it as one: shoppers see it, hesitate, and some buy anyway. Under an assistant, a 3.8-star product may not get shown, which makes it an eligibility problem. You don't lose some of the traffic. You lose the chance to be considered.
That reframes review work. Getting from 3.8 to 4.3 has always been worth doing. It's now the difference between being in a shortlist and being absent from one, which is a much harder loss to see in your reports, because nothing in Seller Central tells you about the answers you weren't in.
What does the assistant take from your reviews and returns?
It reads what your reviews say as well as the average, and a recurring complaint becomes part of how the assistant describes you.
Amazon's side-by-side comparison weighs reviews alongside features and price, and the overview on a detail page summarizes what buyers report. A complaint that turns up in a dozen reviews ("runs small," "too sweet," "lid leaks") stops being a few unhappy customers and starts being a characteristic of the product in the summary a new shopper reads first.
That's why our audits start with the voice of the customer and the return reasons before anyone rewrites copy. On one listing Josh reviewed this summer, shoppers kept returning the product because the fit wasn't what they expected, and the listing gave a size range without explaining how the product was meant to fit. The fix there is on the page: set the expectation in the images and the A+ content, and the review that was going to complain about it never gets written.
If the problem is the count rather than the content, a newer listing usually needs a review base before any of this matters, and the Amazon Vine program is the fastest honest way to get one.
Where do I find the questions to answer?
Your competitors' listings, in the customer questions section. It's the cheapest research in this article and worth spelling out properly.
Every question a shopper asked there is a question shoppers in your category are asking. Someone wanted to know whether it's shelf stable after opening. Someone asked about the sweetener. Someone asked whether it fits a specific dietary restriction. Those questions are free research, published, sitting on a page you can read in ten minutes.
Take the ten most common across the leading products in your category and answer every one of them on your own page: in a bullet, in an A+ module, in the text on an image. You're doing two things at once. You're becoming the listing that matches a long constrained query. And you're removing the hesitation that costs you conversion from human shoppers, who have those questions too and mostly don't bother asking.
It's the cheapest listing work available and I almost never see it done.
What does personalization change for a natural or better-for-you brand?
A shopper who has told the assistant they're dairy-free may never type "dairy-free" again, and the assistant can only match that need to your product if the attribute is written on your page.
This is my read of what Amazon has described rather than something Amazon has spelled out for sellers. Alexa for Shopping lets shoppers store details like dietary needs and family members, and it carries their past purchases and conversations into new searches. The constraint a shopper used to type into the question is increasingly sitting in their profile instead.
For the brands we work with, that's mostly good news, because the attributes that define natural and better-for-you products are exactly the ones a profile holds: gluten-free, vegan, low sugar, nut-free facility, organic, non-GMO. It's only good news for the brands that state them plainly on Amazon. A certification logo on the front of the bag in the main image is easy to miss for a shopper scrolling, and I wouldn't count on an assistant reading a logo the way it reads a sentence. Write the same claim into the title, bullets or A+ content as well.
The same goes for the language you choose. Write the attribute the way a shopper would describe the need, because that's the vocabulary both the question and the profile use.
Do keywords still matter?
Yes, and I want to be careful here, because "AI changes everything" makes brands abandon work that's still doing the job.
Indexing decides what you're eligible for. If your listing isn't indexed for a term, you can't rank for it, and that machinery hasn't gone anywhere. Your title, your bullets and your back-end search terms are still how Amazon knows what your product is, and a page that hasn't done that groundwork isn't in a position to be surfaced by anything, conversational or not.
So the answer isn't keywords or answers. It's keywords for eligibility, answers for selection. A page that indexes broadly and answers specific questions is well positioned under both systems. A page optimized only for a two-word category term is well positioned under one of them.
The other thing that carries over is that inventing your own vocabulary hurts you in both. If you lead with a term you coined, a keyword search misses you and a situational question misses you too, because a shopper describing their problem uses ordinary words.
Can AI run my Amazon account?
Brands ask us this now, and the honest answer is not yet, for a specific and checkable reason.
I connected Claude to Seller Central to see how far it could get. It returned different sales numbers for the same question depending on whether it pulled a seven-day or a fourteen-day attribution window. Same question, two answers, both defensible, neither flagged as a choice.
That's the problem in miniature. The models are good. The data layer underneath Amazon advertising is full of decisions that look like details and aren't, and a tool that makes those decisions silently will produce confident numbers you can't act on. When an agency or an operator picks an attribution window, they're making a judgment about what they're measuring. When a model picks one because it was first in the response, it isn't.
Some of our own clients have tested this directly by trying to run the account with AI instead of with us. So far every one who tried has come back. I don't say that to be smug about it, because the tools improve every quarter and I expect that answer to change for the simplest accounts first.
I'd extend that beyond reporting. Marketing strategy, copywriting and business strategy are all places where AI is useful and none of them are places where it's finished. Where it has already taken real work is operations: the repetitive account tasks, the pulling and reconciling, the first draft of everything. In a service business the team is the product, so that's where it pays first.
Why generic content doesn't get cited
One more implication, and it runs past Amazon into everything you publish.
If your content says what every other article on the subject says, nothing has a reason to quote it. An AI answer names a source when the source has something the others don't: a number, a method, a position somebody holds. Repeating best practice more fluently doesn't earn a citation, because fluency isn't scarce.
The same logic applies to your product page. A listing that describes a category the way every listing in it does gives an assistant nothing to select on. A listing that answers the specific constrained question gives it a reason to choose you. It's the same test on both surfaces: do you say anything that couldn't be copied from the page next to yours.
What to do this quarter
Pull your review averages by ASIN and rank them. Anything under 4.0 is now a visibility problem, so treat it as urgent rather than as a slow improvement project.
Read the customer questions on the top five competing products in your category and write down every recurring question. Answer all of them on your page, in the bullets, in A+ content and in image text.
Check that every claim that matters about your product exists on Amazon rather than only on your website. Certifications, dietary attributes, use cases, what it's for and who it's for.
Ask the assistant yourself. Open Alexa for Shopping in the Amazon app, type the long, constrained question a real customer of yours would ask, and see which products come back and how it describes them. Then select your product and two competitors and run the side-by-side comparison. It takes ten minutes and shows you what a shopper sees before they ever reach your page.
Read your return reasons and your three most common review complaints, and fix whatever on the page is setting the wrong expectation.
Then leave your keyword and indexing work alone. It's still doing its job.
The bottom line
The shopper asking a paragraph-long question is telling you exactly what they need, in a level of detail a keyword never contained. The brands that win under this are the ones whose pages answer, in plain words, on Amazon, with a review average high enough to be considered.
None of that is a new discipline. It's the listing work that was always worth doing, with a harsher penalty for skipping it.
If you want to know how your listings read to an assistant, send us your ASINs and we'll run the questions your category is asking against your pages and tell you what's missing. We've never regretted giving away the analysis.






