Amazon Now Has Two Customers. Most Brands Are Only Optimising for One.

When you upload a product listing to Amazon, you are writing for two completely different readers. One is the human shopper - weighing options, scanning bullet points, checking reviews. The other is the AI layer that now sits at the top of the search bar, deciding whether your product is worth surfacing in response to what the shopper typed before the human ever sees a result.

Most brands have spent years learning how to write for the first reader. Almost none are writing for the second.

The short answer: Amazon's traditional search algorithm and the AI shopping layer - now operating as Alexa for Shopping following Amazon's May 2026 transition from Rufus - read your listing very differently. A listing built for keyword-based discovery can actively underperform in AI-mediated search, because an AI extracting product information to answer a conversational query needs clarity, completeness, and natural language - not keyword density. Alexa for Shopping launched as the default search experience for signed-in US customers in May 2026, with UK and EU markets currently in transition. Over 250 million shoppers used Amazon's AI assistant in the year before that US launch, and the share of product discovery flowing through the AI layer - in every market where it is live - is large and growing. Optimising for only one customer means accepting that a significant and increasing share of purchase intent is not finding your product - and that gap will only widen as UK and EU marketplaces complete their own transition.

---

The First Customer: The Human Shopper and the Algorithm That Puts Products in Front of Them

Amazon's search algorithm - commonly referred to as A10 - has shaped how sellers write listings for the better part of a decade. The optimisation logic is well understood: keyword relevance, conversion rate, click-through rate, pricing competitiveness, review velocity, and fulfilment speed all feed signals that determine where a product appears in search results. A product that converts well on a relevant keyword reinforces its own ranking. One that does not gets pushed down.

Listing content built for this system follows a recognisable pattern. Titles are packed with the most searched terms in the category. Bullet points lead with capitalised feature headers followed by a dense clause of keyword variations. Backend search fields are loaded with every term the brand hopes to rank for. The writing is structured around what the algorithm can detect rather than what a human reads naturally.

This approach still matters. The A10 signals are real, and a listing that ignores keyword fundamentals will not rank in traditional search. But building an entire listing strategy around keyword signals alone means building for only one of your two customers - and leaving the other without a coherent strategy.

---

The Second Customer: The AI Layer That Now Operates in the Search Bar

Amazon launched Rufus - its generative AI shopping assistant - in the UK in September 2024. In the year that followed, more than 250 million shoppers used Amazon's AI assistant globally. Users who engaged with the AI during their shopping journey were 60 per cent more likely to complete a purchase than those who did not, and Amazon projected Rufus would drive an additional $10 billion in downstream sales annually.

In May 2026, Amazon retired Rufus and replaced it with Alexa for Shopping - a more capable system that operates not as an optional side panel but as the default search experience at the top of the page. This rollout launched in the US in May 2026. UK and EU marketplaces are currently operating under Rufus or early iterations of the Alexa+ experience while they await the full default search bar transition - but the direction is already set. Where Rufus was something shoppers had to choose to open, Alexa for Shopping intercepts the search before it happens. A shopper who types "which blender handles hot soup" into Amazon's search bar does not get a ranked list of blenders matched by keyword. They get a conversational response - a comparative summary, product recommendation cards, and an AI-generated answer - based on what the system has extracted from listings, reviews, Q&A sections, and attribute data across the entire catalogue.

Your listing is the primary source of information that AI reads before deciding whether to recommend your product. If that information is incomplete, keyword-dense rather than clear, or missing the attributes the system looks for, your product is invisible to that reader - regardless of how well it ranks in traditional search.

---

Why the Same Listing Can Win With One Customer and Lose With the Other

This is the structural tension most brands have not yet addressed. The listing techniques that signal relevance to the A10 algorithm are not the same ones that make a listing useful to an AI extracting information to answer a question.

A bullet point that reads "Heavy Duty Storage Rack - Metal Shelving Unit - Adjustable Garage Shelves - 5 Tier Industrial Shelf - 2000lb Load Capacity - Easy Assembly" performs reasonably well as a keyword document. It contains multiple search terms the algorithm can detect. But an AI asked "what shelving is strong enough to hold welding equipment" needs to extract the weight capacity, the material specification, and context about appropriate use cases - and it needs that information stated clearly rather than buried in a comma-separated phrase list.

The same dynamic runs through every element of the listing. A title dense with keyword variations is harder for a language model to parse than one that states clearly what the product is, who it is for, and what distinguishes it. An A+ content module built as a promotional banner is less useful to an AI than one structured to answer the questions a genuine buyer would ask. A Q&A section with two entries from three years ago is a weaker signal than one with eighty answered questions covering the concerns buyers consistently raise.

When we review client listings ahead of a conversion or compliance audit, the pattern is consistent. The keyword architecture is often in reasonable shape. The attribute fields are half-empty. The Q&A section has not been touched since the listing was created. The A+ content exists but reads as a brand story rather than a source of purchasing information. Against the traditional A10 signals, the listing looks functional. Against the AI customer, it is providing almost nothing to work with - and as AI-mediated discovery becomes the default across US, UK, and EU marketplaces, that proportion will only grow.

---

What the Second Customer Actually Needs From Your Listing

Alexa for Shopping draws on five areas of your listing when deciding whether to include your product in an AI-generated recommendation or comparison.

The first is attribute completeness. Every field in Seller Central - dimensions, weight, material, certifications, compatibility, colour - is data the AI can extract and match against a shopper's query. A field left blank is a question the AI cannot answer on your product's behalf. When a competing product has that field completed and yours does not, the AI will surface theirs. When we review listings, the fields most commonly left blank are material composition, item weight, and product dimensions - the exact specifications an AI needs to answer questions like "will this fit in a standard kitchen cabinet" or "is this strong enough to hold outdoor equipment." This is the most common gap we find, and the one that is fastest to close.

The second is the quality of description and bullet point content. Natural language that states clearly what the product does, who it is for, and under what conditions it performs best is more useful to an AI reader than keyword phrases. Bullets that answer questions buyers genuinely ask - is this compatible with X, how long does it last, is it suitable for outdoor use - give the AI something concrete to work with when matching your product to a conversational query.

The third is Q&A richness. Amazon's AI can draw on the Q&A section in response to shopper questions. A listing with substantial coverage - answers to the common concerns in your category, given in clear and complete language - is significantly better positioned in AI-mediated discovery than one that has been left sparse. This is a consistently under-maintained element of most listings.

The fourth is review content. Reviews are a signal Alexa for Shopping reads when forming recommendations and comparisons. Review quality, recency, and relevance to likely buyer questions all matter. Vine at launch is not a review strategy - it is a starting point. An active, sustained approach to encouraging verified reviews over time feeds the second customer in ways keyword optimisation cannot replicate.

The fifth is A+ content structure. A+ content built to inform rather than to sell - answering practical questions about the product, explaining use cases clearly, providing comparison tables that highlight genuine differences - is more useful to an AI constructing a product comparison than content built purely as a brand awareness asset. If your A+ modules were built to impress at first glance rather than to communicate specific product information, they are doing relatively little for your AI customer.

None of this replaces keyword optimisation. The first customer still needs to be served. The point is that both customers require intentional design, and the vast majority of listing strategies have been built around one.

---

For more on what changed when Amazon retired Rufus and launched Alexa for Shopping - and what to do operationally about the transition - see our earlier piece on what the shift means for sellers.

The gap between what most listings deliver for the A10 customer and what they deliver for the AI customer is almost always visible within the first review. If you want to know where your listings stand against both - and what it would take to close that gap before UK and EU marketplaces complete their own transition to Alexa for Shopping as the default experience - we can start with a straightforward audit of your core catalogue. Talk to us here.

---

Frequently Asked Questions

What is Alexa for Shopping and how is it different from Amazon Rufus?

Rufus was Amazon's generative AI shopping assistant, launched in the UK in September 2024 and used by over 250 million shoppers globally in 2025. In May 2026, Amazon retired Rufus and replaced it with Alexa for Shopping - a more capable system that operates as the default search experience for all signed-in customers rather than an opt-in side panel. The critical difference is reach: where Rufus was something shoppers had to actively choose to open, Alexa for Shopping intercepts the search for every signed-in customer automatically. For sellers, this means AI-mediated product discovery is now the dominant pathway for a large share of shopping journeys, not a feature only some users ever encountered.

Does optimising for AI search mean abandoning my keyword strategy?

No. Keyword optimisation for the A10 algorithm remains relevant - organic rank in traditional search still drives significant volume and should not be neglected. The goal is to design your listing so it works for both customers rather than only one. In practice this usually means completing attribute fields that have been left empty, restructuring bullet points to read more naturally while retaining relevant terms, building out Q&A coverage, and ensuring A+ content answers real purchase questions rather than functioning purely as brand promotion.

Which specific listing elements does Alexa for Shopping prioritise?

The five most significant areas are attribute completeness (every Seller Central field that applies to your product category), the clarity and specificity of bullet points and description content, Q&A section depth and recency, review quality and sustained review velocity, and the informational value of A+ content. Incomplete attribute fields are the most common gap we find in listing reviews - they represent questions the AI cannot answer on a product's behalf when a shopper's query would otherwise be a strong match.

How do I know if my listing is underperforming in AI-mediated search?

A practical test is to read your own listing as an AI reader would. Can you clearly extract what the product is made from? Who it is designed for? What it weighs and measures? What specific questions it answers? Can you find substantial, relevant answers in the Q&A section to the questions buyers in your category commonly raise? If those answers are difficult to find or simply absent, the AI is encountering the same problem - and recommending your product in response to conversational queries is correspondingly harder.

Does this apply to Amazon UK and EU marketplaces as well as the US?

Rufus launched in the UK in September 2024 and rolled out across EU marketplaces in late 2024. The transition to Alexa for Shopping is being rolled out from the US, with international markets following. If you are selling on Amazon.co.uk or EU marketplaces, the AI discovery layer is already active - and the same listing principles apply across all markets where the feature is live. The direction of travel is clear regardless of precise rollout timing: AI-mediated discovery is the default going forward, and optimising for it is no longer optional.