You checked the hero SKU on Sunday night. Sessions were fine. Orders weren't.

Nothing on your end changed. Same price, same images, same keywords ranking right where they always rank and yet the units slipped, and the keyword report gives you nothing to grab onto.

I've watched this exact thing play out in accounts I run. The seller goes hunting for a competitor who undercut them, or a bid that broke, or a suppression flag. Usually it's none of those.

The thing that changed isn't in your listing. It's who's reading it.

The buyer changed, and nobody sent a memo

On May 13, Amazon retired Rufus. The standalone chatbot got folded into a single assistant called Alexa for Shopping, which now sits inside the main search bar. That last part is the part that matters.

Rufus was a side door most shoppers never bothered to open. Alexa for Shopping is the default AI layer for every signed-in U.S. customer on the app and website , no Prime membership, no Echo device required. Same brain, much bigger reach.

Then, as Prime Day kicked off in late June, Amazon showed its next card. It launched Alexa+ Agentic Ads on Echo Show a format Amazon calls the first to take a customer from seeing an ad to completing a purchase entirely inside the conversation, without ever leaving the ad.

So why does it matter to a physical-products seller? Because it's the direction of travel, stated out loud. Amazon is building toward a buyer that researches, compares, and checks out without a human ever eyeballing a product grid.

Keyword matching is giving way to attribute fit. And the engine underneath has a name.

What an agent actually reads , and what it ignores

The engine is COSMO, Amazon's semantic layer. It evaluates listings on how well they address what the customer is trying to accomplish, not just whether they contain a matching string of text. Old world: someone types "running shoes for men," and Amazon finds listings with those words. New world: a customer asks for running shoes for plantar fasciitis on concrete sidewalks, and the system looks for the listing that proves it solves that.

Here's the uncomfortable part. If your listing is built around "running shoes" but never connects to plantar fasciitis or shock absorption or hard surfaces, you don't just rank lower — you're removed from the consideration set, because the AI doesn't believe you're the answer.

That's the shift, and it's a big one. A human buyer gets sold by a hero lifestyle shot and a sharp headline. An agent doesn't feel the vibe of your photography. It parses your attributes, your specs, your bullets, your reviews, your Q&A, and it matches that data against exactly what the shopper said they needed.

Two gates now, not one. A9/A10 still determines whether you even enter the pool of products the AI can pull from; the AI layer determines whether you get recommended out of it. You can clear the first gate and lose at the second, and your keyword report will look completely healthy the whole time.

Where thin data quietly costs you the sale

This is the failure mode nobody catches for weeks, because it doesn't look like a problem. Sessions hold. Rank holds. But the sale gets lost up in the AI answer or the comparison card — the part of the page the shopper acts on before they ever scroll down to your listing.

Your ROAS can look great the entire time. It's just sitting on a base that's quietly shrinking while the agent intercepts generic queries and the share shift stays invisible until BSR moves.

I've seen versions of this in accounts I run, and there's a documented case that shows the shape of it perfectly. A seller had a water bottle the assistant kept describing as plastic — it was stainless steel — a real price-tier and quality difference. They fixed the listing and sales jumped 18% in two weeks. The machine wasn't wrong on purpose. The data was thin.

The pattern shows up by material across a whole catalog too. One kitchen-goods seller saw plastic SKUs drop conversion while stainless held, because the AI steers "best" and "durable" queries toward the material it reads as higher quality.

Miss an attribute and you miss the answer. Ask for the best compact coffee maker for a small kitchen, and if your size and room-type fields are empty, the word "compact" sitting in your title won't surface you.

The three things that carry weight now

So what do you do about it? Three things, in order of leverage.

Attribute completeness

Fill every backend field your category offers. All of them. The discovery attributes especially — subject, target audience, intended use — now feed directly into how COSMO categorizes and surfaces your listing.

It's the least glamorous work on this list and the highest-return. Most catalogs I open have blanks here, and those blanks stay invisible right up until an agent goes looking for the exact thing you left empty.

Comparison-table readiness

Alexa for Shopping builds side-by-side comparison tables across competing ASINs on demand. Picture your product sitting in that table next to your top three competitors.

If they've populated dimensions, material, compatibility, and certifications, and you've got gaps, you lose the row-by-row — even when yours is the better buy. The fix is boring and specific. Pull your top three competitors, and fill every field they've filled that you haven't.

Bullets that answer questions

Stop writing bullets as feature stacks. Write them as answers to the questions buyers actually ask, in the words buyers actually use — mined from your reviews, your Q&A, and the one almost nobody reads, your return-reason text.

Feature, why it matters, specific proof. "Six-millimeter silicone" beats "durable," because the agent can hand a shopper asking whether the case is thick enough a real number instead of a vague adjective.

What this does to your ad spend

Here's where it reaches into the part you're already measuring. When the assistant recommends a competitor inside an answer, that impression never reaches your keyword report and the sale never carries a campaign attribution. You don't see the miss. You just see softer numbers with no cause attached.

Paid is moving onto the same surface. Sponsored Prompts and standard sponsored placements now appear inside conversational results, which means ad dollars increasingly ride on the same listing quality that drives organic surfacing. Thin data drags both at once.

And when the agent can close the purchase itself — the direction Agentic Ads is pointing — creative that only grabs attention starts losing to data that helps the agent decide. One agency exec put the near-term version bluntly to Digiday: if you can't do the organic piece, you're wasting money on the paid piece. I'd say the same to any client. The ad can't rescue a listing the machine can't read.

The same-week audit you can actually run

None of this needs a tool you don't already have. Here's the pass I'd run on any account this week.

Pull your top five to ten SKUs by revenue. Open each one next to your top three competitors, and compare the populated fields, not the copy — every blank you've left that they've filled is a note to self.

Then talk to the assistant like a buyer would. Ask Alexa for Shopping what your product is made of, how it compares to a named competitor, whether it's right for a specific use case. Wrong answers and "I'm not sure" answers are data gaps you can see with your own eyes.

Read your return-reason text and your three- and four-star reviews for the exact language buyers use, then feed those phrases into your bullets and Q&A. Watch two numbers as your early tell: the gap between detail-page sessions and units, and your branded versus non-branded impression share. When an agent starts moving your share, those move before BSR does.

One patience note. COSMO changes propagate over roughly 7 to 14 days, not the 24-hour turnaround an old keyword tweak gave you — treat that as an industry estimate and verify it against your own tracking. Don't rewrite a bullet on Monday and revert it Thursday because nothing moved.

FAQ

What replaced Rufus on Amazon?
Alexa for Shopping. Amazon retired the standalone Rufus chatbot on May 13, 2026 and folded its recommendation features and shopping history into a unified assistant that now lives in the main search bar, the app, and Echo Show.

Do keywords still matter on Amazon in 2026?
Yes , they get you into the retrieval pool. But presence alone no longer wins the recommendation. The system now evaluates how well your listing addresses what the customer is trying to accomplish, and keyword stuffing works against you.

How does Alexa for Shopping decide which products to recommend?
It reads your listing for meaning — attributes, bullets, description, A+ Content, reviews, Q&A — and matches that against the shopper's stated intent using COSMO, Amazon's semantic engine. Rufus and COSMO rely heavily on structured attribute data to match products to queries, so complete and consistent data wins.

What are Alexa+ Agentic Ads, and can I buy them yet?
They're a format that lets a shopper move from seeing an ad to completing a purchase without leaving the Alexa experience, currently available on Echo Show. Right now it's a beta with a closed partner list in food and ticketing, and Amazon hasn't disclosed pricing — not self-serve, and not available for private-label products.

Which product attributes matter most for AI shopping?
The ones buyers filter and ask on: material, size and dimensions, compatibility, intended use, target audience, certifications. Discovery attributes in your product template feed directly into how COSMO categorizes and surfaces your listing.

How long do listing changes take to affect AI recommendations?
Plan on roughly one to two weeks for semantic changes to fully propagate, not the near-instant turnaround you'd expect from a keyword edit. (Estimate — verify against your own data.)

The buyer that isn't human is already in the room

It's reading your data right now, deciding whether you make the answer or sit three scrolls beneath it, and it'll keep making that call whether or not you've handed it anything good to read.

Most sellers won't notice until BSR slips. By then the share has already moved.

If you want a second set of eyes, that's the work I do — pulling a catalog apart attribute by attribute, against the top three competitors, on the SKUs that actually carry the revenue. I run a 2,000-plus SKU FBA catalog across the US and Canada before I ever touch a client account. So I read your listings the way the machine reads them.

Sources

• CNBC — Amazon ditches Rufus AI chatbot in favor of Alexa shopping agent (May 13, 2026): https://www.cnbc.com/2026/05/13/amazon-ditches-rufus-ai-chatbot-in-favor-of-alexa-shopping-agent.html

• Fast Company — These new Amazon ads don't just recommend products — they can make your purchases for you (June 2026): https://www.fastcompany.com/91562971/these-new-amazon-ads-dont-just-recommend-products-they-can-make-the-purchase-for-you

• Search Engine Land — Amazon launches Alexa+ Agentic Ads: https://searchengineland.com/amazon-launches-alexa-agentic-ads-480842

• Digiday — Amazon's latest ad format offers a glimpse of advertising's agentic future: https://digiday.com/marketing/amazons-latest-ad-format-offers-a-glimpse-of-advertisings-agentic-future/

• AdWeek — Alexa Is Becoming a Shopping Agent for Advertisers Outside Amazon: https://www.adweek.com/commerce/alexa-is-becoming-a-shopping-agent-for-advertisers-outside-amazon/

• Amazon Ads — Amazon introduces Alexa+ Agentic Ads: https://advertising.amazon.com/library/news/alexa-agentic-ads

• Seller Labs — How Amazon's COSMO Search Changes SEO for Sellers: https://www.sellerlabs.com/knowledge-base/how-amazons-cosmo-search-changes-seo-for-sellers/

• Canopy Management — Amazon Algorithm 2026: 10 Ranking Factors / Alexa for Shopping Sellers Guide: https://canopymanagement.com/amazon-alexa-for-shopping-sellers-guide/