How to Optimise Amazon Listings for Rufus, Amazon's AI Shopping Assistant (2026)

Amazon shoppers increasingly ask questions instead of typing keywords: “which of these is quieter?”, “will this fit a 60 cm worktop?”, “is this safe for a 3-year-old?”. Rufus, Amazon’s AI shopping assistant, answers from whatever your listing, reviews and Q&As say. If the answer isn’t there, or is buried in an image, your product is harder to recommend. This guide covers what Rufus reads, what to change and how to check it’s working.

What is Rufus, and is it still called Rufus?

Rufus is Amazon’s generative AI shopping assistant. Amazon describes it as trained on its product catalogue, customer reviews, community Q&As and information from across the web. It launched in beta in the US in 2024, reached UK shoppers in beta on 4 September 2024 and Germany, France, Italy and Spain in October 2024.

On 13 May 2026 Amazon renamed Rufus to Alexa for Shopping in the US, merging it with Alexa+. Coverage of the launch describes it as US-only at launch, with AI overviews on search results and product pages, side-by-side comparisons and buying guides. UK and EU shoppers may still see the Rufus name. The listing data behind both is the same, so this guide uses “Rufus” throughout.

Why it matters, in Amazon’s own numbers:

Figure	Source
More than 250 million customers have used Rufus, Amazon says	Amazon
More than 300 million customers used Rufus in 2025 (full year)	Andy Jassy, Q4 2025 results, via Digital Commerce 360
Shoppers who use it are over 60% more likely to buy during that trip	Amazon
Expected to generate over $10 billion in annual incremental sales	Andy Jassy, Q3 2025 call, via Fortune

How does Rufus read an Amazon listing?

Rufus builds answers from several sources at once, and you control most of them. The table shows what it draws on and what that means for your listing.

Source	What Rufus uses it for	What you control
Title, bullets, description, attributes	Facts about the product: size, material, compatibility, use	All of it
A+ content and Brand Story	Use cases, comparisons, brand claims (treated as a knowledge source in third-party testing)	All of it
Customer reviews	What real buyers say about quality, fit and problems	Indirectly: the product, the packaging and your expectations
Community Q&As	Answers to specific questions	Your answers, where you can still respond
Images	Amazon hasn’t said; testing suggests text overlays can be read	All of it
The wider web	Buying advice from publishers such as Good Housekeeping and Vogue (Amazon)	Little, except your brand’s reputation
The shopper’s own history	Personalised answers through account memory	Nothing

Reviews matter in a second way too. Amazon’s “Customers say” review highlights are AI summaries built only from verified-purchase reviews, and they’re live in the UK. A complaint that shows up in many reviews, such as “runs small”, becomes part of what shoppers see and what an assistant can repeat.

Amazon’s research points the same way. Its COSMO project uses AI to infer the intent behind a search: someone searching “shoes for pregnant women” probably wants slip-resistant shoes. Amazon hasn’t said that COSMO powers Rufus. But it shows the direction: products get matched to needs and situations, not just to matching words.

Keyword search vs Rufus: what changes for sellers?

Keyword SEO still decides whether you appear in normal search, and most shoppers still type keywords. Rufus adds a second test: can an AI answer a specific question about your product with confidence?

	Keyword search	Rufus and AI answers
Shopper input	“insulated water bottle 750ml”	“Which bottle keeps drinks cold all day and fits a car cup holder?”
What wins	Relevance and sales history for the keyword	Clear, consistent facts that answer the question
Typical failure	Missing the keyword	Missing the fact, or facts that contradict each other
Where facts live	Title, bullets, backend search terms	Title, bullets, attributes, A+, reviews, Q&A

What to change on your listing for Rufus

Work through these in order. The first three give most of the benefit.

1. Fill every relevant attribute

Structured fields (material, dimensions, capacity, age range, compatibility, care instructions, power source) are the easiest facts for any system to read. Many listings leave half of them blank or use a different value from the bullets. Download your listings, fill the gaps and check that every value matches the copy.

2. Rewrite bullets so each one answers a question

Start each bullet with what it answers, then give the specific fact and the proof. Keep claims you can back up.

Worked example (illustrative product: an insulated 750 ml bottle)

Before	After
PREMIUM QUALITY: made from the best materials for all your needs	KEEPS DRINKS COLD FOR 24 HOURS: double-walled 18/8 stainless steel; keeps drinks hot for 12 hours
PERFECT FOR TRAVEL: take it anywhere	FITS MOST CAR CUP HOLDERS: 7.4 cm base, 26 cm tall, 380 g empty
EASY TO CLEAN	DISHWASHER-SAFE LID: hand-wash the bottle; wide 5 cm mouth fits ice cubes and a brush

The “after” version answers three questions Rufus is likely to get (“how long does it stay cold?”, “will it fit my car?”, “can it go in the dishwasher?”). The “before” version answers none.

3. Make the title clear before clever

Amazon’s title requirements cap titles at 200 characters including spaces and, since January 2025, block characters such as ! $ ? _ { } ^ unless they’re part of the brand name (summary). Lead with brand, product type and the attributes that decide a purchase (size, quantity, compatibility), not a string of keywords.

4. Put facts in text, then show them in images

Amazon’s image requirements don’t allow text on the main image, but secondary images can carry infographics. Use them for dimensions, what’s in the box and comparisons, and repeat the same facts in the bullets or attributes. Never leave a fact only in an image.

Images also decide whether shoppers click at all. In one Horison test, a redesigned main image lifted click-through rate by 68% and conversion rate by 47%.

5. Use A+ content for comparisons and use cases

Rufus gets comparison questions (“what’s the difference between the 500 ml and 750 ml?”). A+ comparison charts and “who it’s for” modules answer them directly. Keep A+ consistent with the bullets: contradictions are how an assistant ends up giving a wrong answer, and wrong answers become returns.

Accuracy pays even without AI. For one brand, clearer specifications, bullets and A+ content cut the refund rate from 12.55% to 7.59%, a 40% drop, because buyers knew what they were getting.

6. Fix what reviews and Q&As say

Read your 1–3 star reviews and returns comments for repeated themes. If a theme is about the product, fix the product or the packaging. If it’s about expectations (“smaller than I thought”), fix the listing. Answer outstanding customer questions where Seller Central still lets you. Never incentivise or filter reviews; that breaks Amazon’s policies.

7. Keep doing keyword SEO

Rufus doesn’t replace search. Keep your keyword research, title and backend search terms current. For a new supplement brand with no established keywords, two rounds of listing SEO plus targeted campaigns grew sessions by 230% and sales by 132%.

How do you find the questions shoppers ask?

You already have most of them. Pull questions from five places:

1. Reviews: every complaint is an unanswered question (“does it fit…?”, “is it loud?”).
2. Returns reasons in Seller Central and the Voice of the Customer dashboard.
3. Community Q&As on your listings and your top three competitors’ listings.
4. Brand Analytics Search Query Performance: long queries such as “for small hands” or “without BPA” are questions in keyword form.
5. Rufus itself: ask it about your category and note what it says buyers should consider.

Turn the list into 10–20 questions and check that each one is answered somewhere in your listing text.

How to test whether your listing works for Rufus

There’s no Rufus ranking report, so build your own simple test and pair it with Amazon’s split-testing tool.

A fixed question panel. Each month, ask the assistant the same questions on a phone, on the same marketplace. Record whether your product is named, and whether the answer about it is correct. Answers are personalised through account memory, so use the same account each time and treat it as a trend, not a precise score.

Example log (illustrative numbers):

Month	Questions asked	Your product named	Answers with a wrong fact
June	15	3	2 (wrong capacity)
August, after attribute and bullet fixes	15	7	0

Manage Your Experiments. Brand-registered sellers can A/B test images, titles, bullet points, descriptions and A+ content. Amazon recommends running tests for 8 to 10 weeks if you set the duration yourself, and says optimised content can increase sales by up to 20%. Test one change at a time.

Business metrics. Watch sessions, unit session percentage (conversion rate) and refund rate by ASIN. If AI answers are sending better-matched shoppers, conversion should rise and returns should fall.

Should you advertise inside Rufus?

Only once the listing is right. In the US, Sponsored Prompts show ads as AI-generated questions next to the assistant’s answers, and moved out of beta on 25 March 2026. Amazon generates the prompts from your product page content, so a thin listing produces weak prompts. Fix the content first, then test the ads as part of your wider Amazon PPC strategy.

UK and EU sellers: what’s different?

The principles are the same, but each marketplace reads its own listing in its own language. A German shopper asking about a product on Amazon.de gets answers from your German listing, so machine-translated bullets with wrong units or mistranslated materials lead to wrong answers. Check every marketplace’s attributes separately, and use metric units and local sizing. For the VAT, EPR and fulfilment side of reaching those stores, see our guide to selling on Amazon Europe from the UK.

Common mistakes

• Pasting keyword lists into bullets. They answer no questions.
• Key facts only in infographics.
• Titles, bullets, attributes and A+ that disagree on size, quantity or materials.
• Claims you can’t support, such as “best” or medical-sounding benefits.
• Ignoring a review theme that has been repeated for months.

How Horison does it

Our Amazon SEO team writes listings for keyword search and for AI questions at the same time: complete attributes, question-answering bullets, and images and A+ content from our creative team that match the text. Horison is an Amazon Ads Verified Partner and has managed more than 200 brands.

To see where your listing stands, run our free AI listing audit, which includes an AI shopping readiness score, or book a free strategy call. For more on how Amazon ranks products, listen to The Secrets to Success on Amazon with Amanda Smith on The Seller’s Journey podcast.