Computers Are Smart Now: Cracking Amazon's Rufus & Cosmo AEO with Jon Tilley

Amazon’s search results page is quietly being replaced by an answer. About 300 million people, roughly 20% of Amazon shoppers, now use Rufus, and Amazon says a shopper who gets a Rufus recommendation is about 60% more likely to buy. So the question every seller should be asking has changed from “am I ranking?” to “is the AI recommending me?” If it isn’t, you are not making the sale. This is not a next-year problem. Rufus is already surfacing ahead of the search results for some queries, and the traditional results page is on its way out.

Here is why it changed. The old A9 algorithm counted keywords, which is why listings turned into clunky keyword soup. A9 is gone. What drives Rufus, conversational search, and increasingly the organic keyword ranking too is Cosmo, an intelligent LLM that connects the dots instead of matching strings. Jon Tilley, who built ZonGuru by reverse-engineering Amazon search, calls the shift SEO to AEO, AI Engine Optimization. The good news is that it rewards better copy, not worse. You can finally let the listing breathe, write for the customer, talk about your brand and your benefits, and stop stuffing.

The New AI Model: two systems

Jon’s model is simple. Listing engineering now splits into two systems, and you can tackle them in any order.

System 1: Semantic Mapping (Cosmo) answers what the product is, who it’s for, when it’s used, and what makes it different. System 2: Question Engineering (Rufus) answers what questions customers ask, which of those drive the decision, and whether your listing actually answers them.

You’re not optimizing content. You’re engineering understanding and decisions.

System 1: How Cosmo maps your product

To recommend you, Cosmo first has to understand what you are and how you fit into a shopper’s world. It does that through a Product Knowledge Graph, and Amazon has published the exact relationship types it maps. This is straight from the source:

Relationship	Maps to	Example
USED_FOR_FUNC	Function / usage	dry face
USED_FOR_EVE	Event / activity	walk the dog
USED_FOR_AUD	Audience	daycare worker
CAPABLE_OF	Function / usage	hold books
USED_TO	Function / usage	build a fence
IS_A	Concept / product type	smart watch
USED_ON	Concept / product type	normal suit
USED_IN	Time / season / event	late winter
USED_IN_LOC	Location / facility	bedroom
USED_IN_BODY	Body part	sensitive skin
USED_WITH	Complementary	surface cover
USED_BY	Audience	cat owner
xINTERESTED_IN	Interest	herbal medicine
xIS_A	Audience	pregnant women
xWANT	Activity	play tennis

The row almost every listing forgets is USED_WITH, the complementary products yours pairs with. Tell Cosmo an espresso maker is used with Italian espresso cups and fresh coffee beans, and it can recommend you to a shopper who never searched for an espresso maker at all. That is net-new demand most sellers leave on the table.

System 2: How Jon reverse-engineers the questions

We don’t wait for questions. We engineer them before they’re asked.

The Rufus side is not guesswork. Jon’s team builds the question set from real inputs, filters it, then ranks it by how much each question moves the sale.

Start from multiple sources: your current listing content, customer reviews and Q&A, category specs, and external web sources.

Filter every candidate question through three tests:

• Relevance — would roughly half of buyers actually ask this?
• Decision impact — does the answer change “buy” versus “don’t buy”?
• Answerability — can it be answered clearly inside the listing?

Then weight what survives into three tiers:

Tier	What it is	Why it matters
Tier 1 — Decision-critical	The core questions driving the purchase	Win or lose the sale
Tier 2 — Confidence builders	Reduce hesitation, add detail	Close the doubt
Tier 3 — Supporting details	Nice-to-have clarity	Round out the picture

The goal is simple: maximize answer coverage across the highest-impact buyer questions. Answer the Tier 1 questions and Cosmo is smart enough to surface you for the long tail of related conversations too.

Where does all of this live? Your title, bullets, description, back-end search terms, and attributes, plus your image overlays and A+ content. You do not have to force it into the title. Fill the content with the right signals in a way that still sounds great to a human, because Cosmo reads both your words and your images.

Here is why this matters, in one story. Jon ran a real audit on a standing desk that was electric and had presets, the exact attributes a shopper was looking for. Rufus still did not recommend it, because the listing never mapped those attributes clearly. The product was right. The signaling was missing. That gap is invisible until you lose the sale, and every listing has some version of it.

See where you stand, for free. Jon’s team built a free AI Readiness report. Drop in an ASIN and it scores your listing out of 100, where anything over 70 is decent, on exactly these two systems, semantic mapping and question coverage. Then it shows you the specific relationships you are missing and the questions going unanswered, with how to fix each one. Run yours here: ZonGuru AI Readiness Score.

Then prove it before you commit the whole catalog. Optimize one listing, push it live, and measure a month against the old version. Watch three numbers with price and PPC held equal: sessions (discoverability), conversion rate, and niche rank versus competitors. When those lift, roll it out catalog-wide, and you get the jump on your category before everyone else catches on.

This is the Impressions lever of the Amazon Formula. Getting found used to mean ranking for keywords. Now it means being the answer the AI hands back, on Amazon today and, because the same structured signals feed ChatGPT and Google, everywhere shopping is headed next. Computers used to be dumb. They are smart now. Get ready.