Rufus answers 250 million shoppers now. Here is what that does to a fee-focused listing.

Key takeaways

• The scale is real: Rufus is fully rolled out in the US with over 250 million users and interactions up more than 210% year over year; one analysis counts 274 million daily queries by late 2024 and Rufus present in 38% of Black Friday 2025 shopping sessions.
• The proportion matters more: traditional search still drives roughly 80-85%, by the estimates available, of discovery traffic as of early 2026. Rufus layers on top of classic ranking — it does not replace it, so keyword work keeps most of its value.
• The mechanism shift: where classic ranking asks whether your listing contains the searched words, the COSMO layer is asking whether your product solves the problem the shopper described. Amazon has not officially confirmed COSMO powers Rufus; the seller community treats the link as probable, and so do we — flagged as such.
• The cheapest wins are structural, not creative: filling blank backend attribute columns with standard taxonomy values, answering the questions Rufus itself displays on your listing page, and writing descriptive A+ image alt text — all free, all reported to feed the AI layer directly.
• Measurement exists but is partial: brand-registered sellers will find incomplete Rufus attribution data inside Brand Analytics — directional, not definitive.

Rufus is in front of more than 250 million shoppers and classic search still drives four out of five discoveries. Both numbers are true, and together they tell you how much of your week Rufus deserves: some, not all. Every algorithm change on Amazon produces the same two crowds — sellers who ignore it and sellers who panic-buy tools — and Rufus deserves neither response. The numbers say it is big and growing; the same numbers say classic search still carries four fifths of discovery. What follows is the fee-desk version of the story: what is, what is confirmed, and what a listing owner should actually change — ordered by cost, starting at free.

01The scale, dated and attributed

The countervailing number belongs in the same paragraph: as of early 2026, traditional search is still at roughly 80-85% of discovery traffic, with Rufus growing from about 13% of searches in late 2024 toward projections of 35%. Both things are true at once — which is why the right response is re-prioritizing listing work, not rebuilding it.

One honesty note this site owes you: the COSMO connection — the “commonsense knowledge graph” widely described as the intelligence under Rufus — is not officially confirmed by Amazon. Practitioners infer it from patent filings and observed behavior. We treat it as probable and label it accordingly; be suspicious of anyone selling certainty here.

02What actually changes in how a listing is read

Classic ranking matches words. The AI layer is matching intent: a query like “quiet vacuum for a small apartment with a shedding dog” gets decomposed into use cases and constraints, then matched against what your listing demonstrates, not just what it repeats. Three reported consequences follow:

Behavior	Practical meaning
Structured backend fields are preferred over bullet claims — bullets read as “marketing fluff”, fields as verified data	The Category Listing Report’s blank columns are unclaimed ranking surface
A+ image alt text and on-image text are read as data by the AI layer	“Woman crushing ice with tamper tool” tells the machine compatibility facts a pretty lifestyle shot never will
Listings that justify value — durability, lifespan, cost-per-use — are more likely to be recommended	Price justification is no longer just conversion copy; it is machine-readable evidence

03The action list, ordered by cost

Free, this week — backend attributes. Download your Category Listing Report from Seller Central and fill every relevant blank column with standard taxonomy values — the the rule is blunt: if the dropdown says Navy Blue, do not write Midnight Ocean. Pay special attention to the “Specific Uses for Product” field: specific use cases (“searing”, “induction cooking”) over generic room names, because these are hard-coding the product into intent buckets.

Free, this week — answer Rufus’s own questions. Open your listing as a shopper and read the questions Rufus prompts under your product. That list is Amazon telling you what the AI wants to answer about your product. Make sure bullets, A+ and Q&A answer them directly — it is the cheapest market research on the platform.

Free, next content pass — images as data. Rewrite A+ alt text descriptively and put key noun phrases on your infographic images. Keep the main image clean for humans and compliance; make the second image carry the machine-readable text.

Already-planned money, redirected. Nothing above requires a subscription. Your existing listing and conversion work still does most of the lifting — conversion rate remains the lever that moves your cost per ad order more than any bid change, and the AI layer reads the same improved listing.

Measurement, with expectations set. Brand-registered sellers will find partial Rufus attribution inside Brand Analytics. Incomplete and directional — read it monthly for trend, not weekly for decisions.

04What not to do

Do not delete keyword work: four fifths of discovery is still classic search. Do not buy “AI readiness” tooling before exhausting the free structural list above — the vendors quoting the scariest Rufus numbers are, without exception, selling the antidote. And do not chase certainty about COSMO internals that Amazon has not published; optimize for the observable behavior, dated, and revisit when the reporting moves.

Open your own listing as a shopper and write down the three questions Rufus shows under it. Then check whether your bullets answer them. That is the whole audit, and it is free.

Frequently asked

Is Rufus replacing the classic search algorithm?

No. Reporting as of early 2026 puts traditional search at roughly 80-85% of discovery traffic, with Rufus layered on top and growing. Keyword-driven work retains most of its value; the AI layer re-orders priorities rather than replacing them.

Is it confirmed that COSMO powers Rufus?

Amazon has not officially confirmed it. The connection is the widely held understanding among practitioners, inferred from patent filings and observed behavior. This guide treats it as probable and labels claims accordingly.

What is the single highest-value change?

Filling blank backend attribute columns in the Category Listing Report with standard taxonomy values. It is free, feeds the AI layer directly as verified data, and most catalogs have significant blanks.

Can I measure Rufus traffic to my listings?

Partially. Brand-registered sellers will find incomplete Rufus attribution data in Brand Analytics — useful as a directional signal, not as campaign-grade measurement.

Sources

1. Rufus, Amazon’s generative-AI shopping assistant: launch and expansion announcements (US rollout; questions surfaced on product detail pages), About Amazon, company newsroom accessed 2026-09-04
2. Understanding Amazon Rufus AI: the 2026 guide (full US rollout; 250M+ users; interactions +210% YoY; COSMO described as powering contextual intelligence), ZonGuru Secondary accessed 2026-09-04
3. How to optimize your Amazon listing for Rufus: the 2026 playbook (COSMO-Rufus link not officially confirmed but widely held; intent understanding vs keyword matching), ZonGuru Secondary accessed 2026-09-04
4. Amazon Rufus and COSMO: is your listing ready for AI-mediated search (274M daily queries late 2024; 38% of Black Friday 2025 sessions; 60% purchase-likelihood lift; $10B projection; traditional search still 80-85% of discovery; 13% to 35% growth path; partial Rufus attribution in Brand Analytics), Advertpreneur Secondary accessed 2026-09-04
5. The 2026 guide to optimizing Amazon listings for Rufus (structured fields preferred over bullets; Category Listing Report blanks; Specific Uses intent buckets; A+ image alt text and on-image noun phrases as AI data), SellerMetrics Secondary accessed 2026-09-04
6. Amazon listing optimization: the complete guide for brands 2026 (Rufus-prompted questions on the listing page as research input; value-justification reported to improve recommendation odds), Incrementum Digital Secondary accessed 2026-09-04
7. Amazon Rufus and Cosmo explained (introduced early 2024; positioning within Amazon's generative-AI direction), BellaVix Secondary accessed 2026-09-04