Amazon Rufus answers questions rather than returning a list, and the products it names are the ones whose listings answer those questions in plain text. Here is what that changes about listings, keywords and measurement.
Strategy
Amazon Rufus answers a question rather than returning a list, and the products it names are the ones whose listings answer that question in plain, checkable text. The ranking machine underneath has not changed. What changed is the shape of the search, and that is enough to change what your listing needs to say.
There is a lot of noise about this. Most of it is either panic or a sales pitch. Here is the practical version: what actually changed, what to do about it, and how to tell whether it worked.
What is on this page
- What changed, and what did not
- What an assistant can actually read
- Rewriting a listing to be quotable
- What it means for keywords
- What it means for advertising
- Reviews are doing more work than they used to
- Three ways brands overreact
- How to measure something with no report
- The same principle off Amazon
What changed, and what did not
Underneath the assistant, the store works exactly as it did. Relevance decides whether you are eligible to appear for a query. Conversion rate and sales velocity decide who appears first. Nothing in that has been replaced, which is worth saying because plenty of people are selling the opposite.
What has genuinely changed is the query. Shoppers ask longer, more specific questions, and they expect an answer rather than a list. That widens the tail enormously: the same intent now arrives in dozens of phrasings, most of which you would never have written into a keyword list. The mechanics of ranking that still apply are in how Amazon organic rank actually works.
What an assistant can actually read
An assistant recommending a product needs facts it can state confidently. That means structured attributes, plain sentences, and claims a shopper could check. It cannot read the text you put inside a graphic, however beautifully designed, which is where a lot of brands keep their best information.
The practical consequence is unglamorous. Fill every attribute in your listing properly, including the ones that feel tedious: dimensions, materials, compatibility, count, care instructions. Those fields are exactly the sort of thing a question is asked about, and an empty field is a question you cannot be the answer to.
A quick test. Take the five questions your customer service inbox answers most often. Search your own listing text for the answers. If any of them only exist inside an image, or nowhere at all, you have just found the work.
Want to know how your listings read to an assistant?
We check the whole listing, attributes included, and tell you what is missing. Free.
Rewriting a listing to be quotable
Three changes do most of the work.
Write bullets as facts, not adjectives. “Premium quality construction” cannot be quoted by anything. “Stainless steel body, dishwasher safe, fits 58mm portafilters” can. Adjectives are what brands write when they have not decided what the product actually offers.
Answer objections in the text. Whatever reviews and questions keep raising, put the answer plainly in the listing. Size and fit, compatibility, what is in the box, what it does not do. That last one earns more trust than it costs.
Move information out of images. Keep the graphic if it helps a human, and put the same information in text as well. Nothing is lost by saying it twice in two formats, and the alternative is being invisible to half the shopping journey. The wider version of this is in why your A plus content is not converting.
All three help conversion for human shoppers too, which is the useful part. This is not a separate optimisation track competing for your time. It is the same clarity work, with a second reason to do it.
What it means for keywords
Exact match alone covers less of the demand than it used to, because the demand is spread across more phrasings. That does not make exact match less useful for the terms you have proven, it makes discovery more important alongside it.
Automatic campaigns become the research engine. Run them on modest bids, read the search terms weekly, and promote anything that converts into its own exact keyword with a deliberate bid. Accounts doing that are finding real volume in phrasings no tool suggested. Accounts that set automatics up two years ago and never looked are not. The routine is in the search term report guide, and the wider approach to building a list is in Amazon keyword research.
Write the answer, not the keyword. The keyword was always a proxy for the question somebody was asking.
What it means for advertising
Ad placements in and around assisted shopping keep moving, and anybody telling you they have a settled strategy for them is ahead of the evidence. Two things are worth doing regardless of what the formats do next.
Keep your discovery campaigns funded. They are how you find the new phrasings, and they are the part of the account most likely to be cut when budgets get tight.
Make the listing worth recommending. Every format, paid or organic, eventually sends a shopper to your product page. A page that answers questions clearly converts better whoever sent the traffic, which is the point made at length in your listing is a mobile listing now.
Reviews are doing more work than they used to
Assistants read reviews as well as listings, which makes them source material rather than social proof alone.
That changes what a good review is worth. A review saying it arrived quickly tells an assistant nothing about the product. A review saying it fitted a particular model, survived a particular use, or solved a particular problem is exactly the kind of specific, checkable statement that gets a product named in an answer.
You cannot write reviews and should not try. You can influence what they talk about. Follow up messaging that asks a specific question, packaging inserts that mention the use case, and Vine units sent to reviewers who will actually test the thing all shift the language of your review set over months. So does answering customer questions properly, since those answers are public text on the listing.
The same logic applies to the questions section. Every unanswered question is a gap in the record, and every answered one adds a plain sentence about your product that an assistant can use.
Three ways brands overreact to this
Rewriting everything at once. The listing changes described here improve human conversion too, so treat them as normal listing work with a second benefit, not an emergency programme.
Buying tools that promise AI visibility scores. There is no reliable measurement of this inside Amazon yet. A score with no underlying data is a number somebody invented.
Abandoning keyword targeting. Exact match on proven terms still carries most accounts. The tail is an addition, not a replacement, and moving budget wholesale into discovery is a fast way to make your costs worse.
How to measure something with no report
No report will tell you an assistant recommended your product. So watch the indirect signals as a trend rather than a number.
Search term length and variety. Are the terms in your reports getting longer and more varied over months? That is the shift arriving in your account.
Impression share on long phrases. Are you appearing on the four and five word searches, or only the two word ones?
Conversion rate on that longer tail traffic. It should be higher than your average, because the intent is more specific. If it is not, the listing is not answering what those shoppers asked.
Branded search volume. Slower moving, and the clearest sign that people are arriving already knowing who you are.
Track those quarterly. Anybody promising a precise measurement of AI visibility inside Amazon is selling something that does not exist yet.
The same principle off Amazon
Assistants outside Amazon work on the same logic and read different sources: your website, your pricing page, and whatever third parties say about you. What gets a brand recommended in either place is the same thing, which is clearly stated, checkable facts that nobody has to infer.
That is why we publish our own prices openly rather than quoting on request, and it is a decision worth considering in your own category. A brand that states what it charges, what it includes and what it does not is a brand an assistant can describe accurately. A brand that hides everything behind a contact form can only be described vaguely, and vague descriptions do not get recommended.
Want your listings checked for what a person and an assistant both read?
Twenty minutes on a call and we will show you where the gaps are.
Send us the listings and the reports and we will do the check for you
Free, no card, written answer in three working days, and yours to keep either way.
Frequently asked questions
Amazon’s shopping assistant. Instead of returning a list of products for a keyword, it answers a question in sentences and names specific products, drawing on listing content, reviews, questions and category data. It changes the shape of the search, not the underlying store.
The ranking system underneath is the same: relevance decides eligibility, conversion and sales velocity decide order. What changes is the shape of the query. Longer, more conversational searches mean relevance now has to cover phrasings you would never have written into a keyword list.
Answer the questions shoppers actually ask, in plain text, inside your listing. Attributes filled properly, bullets that state facts rather than adjectives, and a customer questions section that covers the real objections. Text inside images cannot be read, so anything important trapped in a graphic is invisible.
Ad placements have been appearing in and around these experiences, and the picture keeps moving. The useful position is not to chase a placement but to make sure the listing itself is the sort of thing an assistant can confidently recommend, since that holds whatever the ad formats do next.
Yes. Exact match alone covers less of the demand than it used to, because the same intent arrives in many more phrasings. Automatic campaigns become your research engine rather than a tidying exercise, and they need reading weekly.
Watch the length and variety of search terms in your reports, your impression share on longer phrases, and conversion rate on that traffic. You will not get a report saying an assistant recommended you, so the signals are indirect and worth tracking as a trend.
The principle is the same, which is that assistants recommend what is clearly stated and verifiable. The sources differ: Rufus reads your Amazon listing, while assistants outside Amazon read your website and whatever third parties say about you.
Send the right reports. Get a straight answer back.
We read the account properly and write back with where the money is going, what it is buying, and the three things we would change first. You keep it either way.
No obligation · no card · a person replies, usually the same working day