Amazon Rufus is Amazon’s generative AI shopping assistant: the chat box that answers questions like “is this tennis racket suitable for beginners?” using the product catalogue, customer reviews, community Q&A and information from the web. It has been live in the UK since September 2024. In the US, Amazon folded it into a new brand, Alexa for Shopping, in May 2026.
Most of what’s written about it is multipliers nobody can source. This post sticks to what Amazon has published, how the system works, and the listing changes we’d make.
What is Amazon Rufus?
Rufus is a conversational assistant inside the Amazon app and website that answers shopping questions and recommends products, using Amazon’s own data plus the open web. Amazon announced it on 1 February 2024 and describes it as “trained on Amazon’s extensive product catalog, customer reviews, community Q&As, and information from across the web”. It sits alongside search, not in place of it.
Amazon’s own UK examples: “What do I need for climbing?”, “Are these durable?”, “Should I get trail shoes or running shoes?” None of them is a keyword. They’re questions about use, fit and suitability. That’s the whole point of this post.
Here’s how the product has changed, as of September 2026:
| Date | What happened | Where |
|---|---|---|
| 1 Feb 2024 | Rufus announced, beta in the US app | US |
| 4 Sep 2024 | Beta for selected UK customers in the Amazon app, wider rollout promised “in the coming weeks” | UK |
| 29 Oct 2024 | Beta in Germany, France, Italy and Spain | EU |
| Apr 2025 | “Buy for Me” beta: Amazon buys items from other brands’ own websites for the shopper | US |
| 18 Nov 2025 | Account memory, price history, auto-buy at a target price, “Help me decide”, agentic add-to-basket | US |
| Nov 2025 → 25 Mar 2026 | Sponsored Products and Sponsored Brands prompts: free test, then billable on CPC | US only |
| 19 Mar 2026 | Alexa+ early access in the UK (£19.99 a month after early access, free with Prime) | UK |
| 13 May 2026 | Rufus rebranded as Alexa for Shopping, merged with Alexa+ | US rollout |
Is Rufus in the UK, and is it still called Rufus?
Yes, Rufus UK has been available since a September 2024 app beta, later extended to desktop. Amazon’s May 2026 Alexa for Shopping announcement covered US customers only. Amazon said Rufus “will continue to power parts of the experience behind the scenes”, so the name changed in the US while the engine didn’t.
For UK sellers the branding matters less than you’d think. Whatever the button says, the answers come from the same sources: your listing, your reviews, questions about your product, and the web. Some agency blogs claim Alexa for Shopping is already live across the UK and Europe. We couldn’t find an Amazon announcement confirming that as of 22 September 2026, so check the Amazon.co.uk app yourself.
Alexa+ did reach the UK in early access on 19 March 2026, with shopping features such as price-drop alerts, comparisons and reordering “your preferred brand at the best price”. That’s a voice route into the same catalogue.
What does Rufus read on your listing?
Rufus answers from four published sources: the product catalogue (your listing data), customer reviews, community Q&A, and the wider web. Since November 2025 in the US, it also uses the shopper’s own history and conversations. Amazon hasn’t published how it weights any of them. So treat each one as a place your answer might come from, and make sure the right answer is there.
| Source Amazon names | What it means for your ASIN | How much you control it |
|---|---|---|
| Product catalogue | Title, Item Highlights, bullets, description, attributes, A+ text | High |
| Customer reviews | What buyers say about fit, durability, smell, noise, ease of use | Indirect (product quality, inserts within policy) |
| Community Q&A | Questions shoppers asked and the answers given | Partial (answer accurately and quickly where the module still appears) |
| Information from the web | Your brand site, reviews on other sites, forums, press | Partial |
| Shopper’s own history (US, since Nov 2025) | What they bought, what they told Alexa or Rufus | None |
First, attributes and text beat images for anything factual. Amazon hasn’t said Rufus reads the text printed on your infographics. If “fits a 15-inch laptop” only appears on image four, put it in a bullet or attribute too.
Second, your reviews are now part of your listing copy. If 40 reviews say the lid leaks, that’s how Rufus will answer “does this leak?”. No bullet rewrite fixes that.
What is Amazon COSMO?
Amazon COSMO is a commonsense knowledge system that links products to the reasons people buy them: the function, the audience, the occasion, the place. Amazon Science described it in a May 2024 blog post and a paper at SIGMOD 2024. It isn’t Rufus; it’s one layer of how Amazon understands what a query means.
A shopper searches “shoes for pregnant women”. The catalogue knows sizes and colours, not that pregnant women tend to want slip-ons with wide fits and good grip. COSMO fills that gap.
How it works, per Amazon Science: mine which queries led to which purchases and which products are bought together; ask a large language model to explain each link; filter weak explanations; have humans rate a sample for plausibility; then train a smaller model (COSMO-LM) to generate that knowledge at scale.
The relations are the useful bit for sellers. The paper lists 15, including used for function (“dry face”), used for event (“walk the dog”), used for audience (“daycare worker”), used in location (“bedroom”), used in body (“sensitive skin”) and used with (complementary products).
The numbers Amazon published, as far as we could verify them:
- 18 major product categories covered, built from about 30,000 annotated instructions (Amazon Science publication page).
- Offline, a COSMO-based model beat the baseline by up to 60% on macro F1 (Amazon Science blog, 10 May 2024).
- Online A/B tests in search navigation, as reported from the paper: a 0.7% relative increase in product sales in the test segment and an 8% increase in navigation engagement.
That 0.7% is a relative lift in one test segment of search navigation. It isn’t “COSMO boosts your sales”, and anyone quoting it as a seller benefit is stretching it.
What has Amazon actually published about Rufus’s results?
Amazon has published company-wide figures in earnings calls and press releases: user counts, purchase likelihood and “incremental annualized sales”. It has not published Rufus’s share of sales, any per-category data, or any UK-specific figure. Here’s the full list we could source:
| Figure | Source and date |
|---|---|
| 250 million customers used Rufus “this year” | Andy Jassy, Q3 2025 earnings call, 30 Oct 2025 |
| Monthly users up 140% and interactions up 210% year on year | Same call |
| Customers who use Rufus during a shopping trip are 60% more likely to complete a purchase | Same call |
| “On track to deliver over $10 billion in incremental annualized sales” | Same call |
| 300 million+ customers in 2025; “nearly $12 billion in incremental annualized sales” | Q4 2025 results release, 5 Feb 2026 |
| Monthly active users up 115% and engagement up nearly 400% year on year | Alexa for Shopping announcement, 13 May 2026, as reported by GeekWire |
The stats you should ignore
You’ll see claims like “Rufus drives 30% of Amazon sales”, “Rufus-ready listings convert 3x better” or “80% of shoppers now use Rufus”. We couldn’t find any of them in anything Amazon has published. If a slide can’t name the earnings call or the About Amazon post, bin it.
Even the real numbers need reading carefully:
- “60% more likely to complete a purchase” compares shoppers who used Rufus with those who didn’t. People who stop to ask detailed questions are often closer to buying anyway. Amazon hasn’t said it controlled for that.
- “Incremental annualized sales” is Amazon’s own attribution model. It hasn’t published the method, and it’s a global, all-category figure. It tells you nothing about your ASIN.
- User counts measure anyone who touched Rufus once. That’s reach, not reliance.
What the numbers do support: hundreds of millions of people use it, and it now carries ads. That’s reason enough to make your listing answerable.
Are there ads inside Rufus?
Yes, in the US. Amazon Ads calls them Sponsored Products prompts and Sponsored Brands prompts. They became generally available and billable on 25 March 2026, US only. Amazon’s description: an “AI-powered enhancement to your existing campaigns” that appears in shopping results and on detail pages, and can open a Rufus conversation or answer on the page.
The mechanics, as of September 2026:
- Automatic enrolment. Eligible SP and SB campaigns are opted in with no setup.
- Generated, not written. Amazon builds the prompts from “first-party signals from your detail pages, Brand Store, campaign data, and more”. You don’t write the question.
- Billed on your existing CPC. There’s no separate bid.
- Pausable. You can pause individual prompts in the Ad Console.
For UK advertisers there’s nothing to switch on yet, and Amazon hasn’t announced a UK date. The lesson carries over: the prompt is written from your detail page, so a thin listing gives it nothing to say. If you run US campaigns, review the prompts data monthly alongside your other PPC channel decisions.
Rufus Amazon listing optimisation: what should you change?
Answer the questions shoppers ask in plain text, in the fields Amazon reads, and stop relying on images and keyword lists to carry facts. None of this is a Rufus trick. It’s the same work that helps search, conversion and returns; Rufus just makes the gaps visible. Our checklist, in order:
1. Collect the real questions first
Before you edit anything, list the 15–20 questions shoppers actually ask. Pull them from community Q&A on your ASIN and three competitors’, one- to three-star reviews, buyer messages, return reasons, the suggested questions Rufus shows on your own page, and use-case queries (“for small dogs”, “for eczema”) in your Search Query Performance data. Tag each with a COSMO-style relation: function, audience, occasion, location, body, compatibility.
2. Map each question to a field
| Question type | Example | Where the answer should live |
|---|---|---|
| Who it’s for | “Is it OK for a 5-year-old?” | Bullet + age attribute |
| What it does | “Will it keep coffee hot all day?” | Bullet with a number (hours, °C) |
| Where it’s used | “Can I use it outdoors?” | Bullet + A+ module text |
| Compatibility | “Does it fit a Nespresso Vertuo?” | Title or Item Highlights if it’s the main use, bullet otherwise, plus attribute |
| Comparison | “What’s the difference between the 500 ml and 750 ml?” | A+ comparison chart, with the key facts in text |
3. Keep the title short and literal
Since 27 July 2026, titles in every category except Media are capped at 75 characters including spaces, with up to 125 characters of Item Highlights underneath. Our 75-character title guide has the detail. For Rufus, the rule is: brand, what it is, and the one attribute that most defines who it’s for. “For runners”, “for sensitive skin”, “for 13-inch laptops”. Cryptic model names don’t help anyone.
4. Rewrite bullets as answers
“PREMIUM QUALITY — built to last” answers nothing. “Stainless steel body, 5-year warranty, dishwasher-safe lid” answers two questions in one line. Our heuristic, not an Amazon rule: every bullet carries at least one number or one named use case, and the five bullets between them answer your top ten questions.
5. Put image text into real text
If a spec only exists as pixels, assume the assistant may miss it. Every claim in your image stack (dimensions, capacity, what’s in the box, compatibility) should also appear in a bullet, an attribute or A+ text. Where A+ modules let you add alt text, write a factual sentence, not a keyword list.
6. Fill attributes completely
Attributes are catalogue data, the first source Amazon names. Empty fields for material, age range, capacity or compatibility leave the assistant guessing. Keep them consistent with the bullets: 1 L in the title and 900 ml in the attributes is the fastest route to an unhelpful answer.
7. Work the reviews and Q&A
You can’t write reviews, but you can fix the recurring complaint and say so on the listing, answer community questions where that module still appears (on many pages, AI-generated answers now sit where Q&A used to be), and use Request a Review or Vine within policy on new ASINs.
8. Test it yourself
Give changes 48 hours, then ask Rufus your top ten questions about your own ASIN, plus the category questions (“best X for Y”). Answers that are wrong, vague or drawn from a bad review are your next edits. Repeat quarterly.
We can’t give you an uplift figure for this, and nobody else can either. What you can measure is conversion in your own Business Reports.
Mistakes we keep seeing
Stuffing “Rufus keywords”. There’s no such list. Question phrases jammed into backend search terms don’t make the assistant recommend you.
Writing for the bot. Bullets like “Q: Is it waterproof? A: Yes” read oddly to people. Answer in a normal sentence.
Ignoring the product. If reviews contradict your listing, the assistant has both. Fix the product or change the claim.
If your listings need a full rebuild for the 75-character limit anyway, that’s the moment to do all of this at once. It’s what our listing optimisation service covers.
FAQ
What is Amazon Rufus?
Amazon Rufus is Amazon’s generative AI shopping assistant, built into the Amazon app and website. It answers product and category questions and recommends products using Amazon’s catalogue, customer reviews, community Q&A and information from the web. In the US it was rebranded as Alexa for Shopping in May 2026, with Rufus still powering parts of the experience.
When did Rufus launch in the UK?
Amazon opened a Rufus beta to selected UK customers in the Amazon mobile app in September 2024, with a wider rollout following in the weeks after. It later became available on desktop. As of 22 September 2026, Amazon’s Alexa for Shopping announcement covered US customers only.
How is Amazon COSMO different from Rufus?
COSMO is a knowledge system that links products to why people buy them: function, audience, occasion and location. It was published by Amazon Science in 2024 and deployed in search navigation. Rufus is the chat assistant shoppers talk to. COSMO is about understanding intent; Rufus is one place where that understanding shows up.
Does Rufus affect organic ranking on Amazon?
Amazon hasn’t said that it does. Rufus recommends products inside conversations, which is a separate surface from the search results page. The listing changes that help Rufus answer questions — clear use cases, complete attributes, accurate specs — also help conversion, and conversion does feed organic rank over time.
Can you advertise in Rufus in the UK?
Not yet, as of September 2026. Sponsored Products and Sponsored Brands prompts became generally available and billable on 25 March 2026 in the US only. Eligible US campaigns are enrolled automatically, charged on existing CPC bids, and prompts can be paused in the Ad Console.
How many people use the Amazon AI shopping assistant?
Amazon said more than 300 million customers used Rufus in 2025 and that it delivered nearly $12 billion in incremental annualized sales, in its Q4 2025 results on 5 February 2026. Amazon hasn’t published UK user numbers or what share of sales Rufus influences.
Sami Sultan runs Ecom Enable, an Amazon growth partner for DTC brands. Rufus announcements, earnings figures, COSMO research and Sponsored Prompts availability checked 22 September 2026. Want to know how your listings answer the questions shoppers ask? It’s part of our free Amazon audit.
Sources: About Amazon: Amazon announces Rufus (Feb 2024); About Amazon UK: Rufus launch in the UK; Retail Insight Network: Rufus UK beta, 4 Sep 2024; About Amazon EU: Rufus beta across Europe; About Amazon: Rufus gets smarter and more personal (Nov 2025); About Amazon: Buy for Me; Motley Fool: Amazon Q3 2025 earnings call transcript; About Amazon: Q4 2025 results; About Amazon: Meet Alexa for Shopping (May 2026); GeekWire: Amazon unifies Alexa+ and Rufus; About Amazon UK: Introducing Alexa+ in the UK; Amazon Ads: Sponsored Products and Sponsored Brands prompts; Amazon Science: Building commonsense knowledge graphs to aid product recommendation; Amazon Science: COSMO publication (SIGMOD 2024); ACM Digital Library: COSMO paper; ByteByteGo summary of COSMO A/B results. Title limit: Ecom Enable’s 75-character title post, sourced to Seller Central.