We asked Amazon's AI shopping assistant the same 50 questions every hour for 11 days and recorded who it recommended. The results explain why some brands are far more visible on Alexa for Shopping (formerly Rufus) than their search ranking would suggest, and what the rest can do about it. This article covers the main findings from our webinar with David Halls and Filip Żok. The highlights reel is embedded below.
Watch the highlights reel here
Want to see the full session. The link you need is here.
Why this matters now
Amazon reported in its 2025 earnings calls that 300 million customers have used its AI shopping assistant, that it has invested $12 billion in it, and that shoppers who use it are 60% more likely to complete a purchase. The assistant is branded Alexa for Shopping in the US, Australia and the UAE, and still appears as Rufus across Europe, Canada, India and Latin America. The mechanics are the same.
The shopping journey is different from traditional search. A shopper types "coffee machines", refines to "coffee machines with milk frother", then opens product pages to read bullets and reviews. With the assistant, they ask "what is the best coffee machine for lattes under £100", then ask follow-up questions about pod costs and set-up without ever opening the page. Fewer clicks, fewer products shown, and a different set of winners.
What we measured
We ran 50 shopper questions across two UK categories, coffee machines and sunscreen, every hour for 261 consecutive hours. That produced 11,000 answers, 71,000 recommendation slots, and just over 1,000 distinct products from 286 brands.
Four things we found
The assistant draws from a pool rather than ranking a list. Ask the same question twice in a row and you can get a different top product, because the assistant re-interprets the question each time (a budget of under £500 on one pass, under £400 on the next). The odds of appearing are stable and predictable. The outcome of any single draw is not. That is why measuring once tells you very little, and measuring a thousand times tells you a lot.
The pool is wide but the odds inside it are very uneven. A typical question drew from around 94 distinct products, yet 67% of all 71,000 slots went to just 50 of them. In the categories we studied, 65% of products never appeared once. An average answer shows six products, against 30 to 60 on a traditional search results page.
Completed attributes were the strongest content signal. Products with no structured attributes filled in did not appear at all. Visibility rose sharply between zero and four completed attributes, and products with seven, eight or nine completed took most of the slots. Specific attributes carried real weight for specific questions: age range for questions about children's sunscreen, recommended uses, scent name and skin tone. Categories differ. 80% of coffee machine listings already had eight or more attributes completed. For sun protection it was 43%.
Best seller rank and the Buy Box still decide a lot. Products in the top 10% of best seller rank had a 90% likelihood of appearing. Products with a suppressed Buy Box were far less likely to appear. Who won the Buy Box, first party or third party, made no detectable difference.
What mattered less than expected
Review volume did not correlate directly with visibility. Some products with very large review counts never appeared because they did not fit the question well enough. Star rating mattered less than expected too: the assistant recommended 52 products rated under four stars, including some at 2.9. Reviews still drive sales, and sales drive best seller rank, so they matter indirectly. They are not the lever most people assumed.
What to do with this
Fill in every attribute field on every product, starting with your top sellers. Then check that your content answers the questions shoppers ask in your category, rather than the pitch you want to make. Traditional search optimisation still matters because it drives the best seller rank that Alexa for Shopping relies on, so the two need to be worked together.
The window for organic results is closing. Sponsored placements are already live on Amazon.com and are expected to reach European markets in the second half of 2026. Brands that know their organic share of voice, and where they are absent, will be in a much better position to decide where paid support is worth it.
How the eStore platform tracks it
In the webinar Filip showed how Alexa for Shopping tracking works in the eStore platform: share of voice against named competitors, the share of category questions where at least one of your products appears, which brands win when you are absent, and a breakdown by around 20 customer needs per category so you can see where you are strong and where you are not. From there the platform identifies products with a strong best seller rank that the assistant is not recommending, and generates title, description and attribute suggestions for a chosen customer need, each with a justification so your team can check it against brand guidelines before anything changes.
If you would like to see this running against your own categories, use the ‘Speak to our Team’ button at the top of this page, or contact your customer success manager.
