How Amazon Rufus reads your listing, and which fields to fill for it

Amazon's shopping assistant answers questions by reading listings, not keywords. Which fields it draws on, why structured attributes and the description matter more than they used to, and a practical order for filling them.

By Vlad Bazarov · Updated 18 September 2026 · 7 min read

Rufus is the assistant built into the Amazon app and site. Shoppers ask it questions in plain language: is this bottle dishwasher safe, which of these is better for a beginner, will this fit a 2019 Subaru. It answers by reading listings and reviews, and it recommends products in its answers. That changes what a listing is for: it is no longer only a page a person skims, it is also a source an assistant quotes.

Amazon has not published a ranking formula for Rufus, and any article that claims exact weights is guessing. What is known comes from Amazon's own descriptions of the assistant, from its patents on reading product data, and from what sellers see change when they fill fields that used to be ignored. This guide sticks to that.

What Rufus reads

Rufus works from the same data a shopper can see plus the structured data behind it:

  • Structured attributes: material, size, count, age range, compatibility, special features, care instructions, certifications. These are the easiest data for a model to trust, because each value sits in a named field.
  • Title and Item Highlights: what the product is and its main reasons to buy.
  • Bullet points and description: the details, the use cases, who it is for.
  • A+ Content for brand-registered sellers, including the text in its modules.
  • Text on images. Amazon's patents describe reading text inside product images, so an infographic's claims are readable, not decorative.
  • Reviews and Q&A: what buyers say, which the assistant uses to answer "is it any good for X".

The practical point: Rufus answers questions, so the listing has to contain answers. A keyword list has no answers in it.

Why the description matters again

For years the description was the field sellers wrote last, because bullets and images did the selling and A+ Content replaced it on the page. An assistant reads differently. A description that explains the routine, the fit, the sizing, what is in the box and what the product is not for gives Rufus material to quote. A description that repeats the bullets gives it nothing new.

Write the description as the answer to the questions a shopper would ask a knowledgeable friend. If you have a support inbox, the most common questions are already in it.

Why attributes matter more than the title now

Amazon's filters, its comparison tables and Rufus all read the attribute fields in Seller Central before the prose. A listing with an empty material or compatible devices attribute never appears in a filter for that material and cannot be recommended for that device, however good the bullets are.

Fill every attribute your category offers, even the ones that look optional. Use the values Amazon lists where it gives a choice; free-text values that match nothing are ignored. Where a fact appears in the bullets and there is an attribute for it, put it in both.

An order for filling a listing

  1. Attributes first. Everything measurable and every yes/no: dimensions, material, count, compatibility, certifications, age range.
  2. Title and Item Highlights: what it is, then the top reasons to choose it, within 75 and 125 characters.
  3. Bullets as answers: each one a question a shopper asks, answered with a specific, checkable fact. The rules still apply: no emoji, no promotions, no guarantees.
  4. Description for what the bullets could not hold: the routine, the fit, the caveats, the box contents.
  5. Images with text that matches the copy. Image 2 as an infographic with three to five benefits stated as facts.
  6. Backend keywords last, for the search phrases that fit nowhere visible.

Writing for a reader that checks

Two habits make a listing easier for an assistant to use, and easier for a person too:

  • Specific over superlative. Keeps drinks cold for 24 hours answers a question. Amazing insulation does not. Superlatives are also what Amazon's checker removes.
  • Consistent across fields. If the attribute says 32 oz, the title says 32 oz and the infographic says 32 oz. Contradictions between fields are the fastest way to get a wrong answer given about your product.

What not to do

  • Do not write bullets as questions to "match" what shoppers ask. Bullets that read Looking for a bottle that keeps drinks cold? waste the first ten words on nothing.
  • Do not stuff attributes with keywords. An attribute holds one value; a wrong value is worse than an empty one.
  • Do not chase rumours about Rufus weights. What is stable is simple: complete, specific, consistent, true.

Checklist

  • Every attribute your category offers is filled with a valid value
  • Title says what it is; Item Highlights say why
  • Each bullet answers a different question with a checkable fact
  • Description adds what the bullets could not: routine, fit, caveats, box contents
  • Image text matches the copy and the current pack
  • No contradictions between fields

Where Cobalt12 fits

Cobalt12 fills the structured attributes together with the title, highlights, bullets, description and backend keywords, from one photo and a competitor audit, and keeps the facts consistent across all of them. You can try it on your own product; the first listing is free.

Amazon changes Rufus continuously and publishes no ranking details. This article reflects what Amazon and its patents describe as of September 2026.

Keep reading