Andrew Bell deeply analyzed Amazon Patent 12,561,383 and he says it reveals something most sellers miss: Amazon doesn't just match keywords anymore. It reads what shoppers do between searches and predicts what they'll type next.

Two systems run on top of each other. Keywords retrieve products. Intent classification steers shoppers toward what to search next.

Every time a shopper changes their search, Amazon classifies the move into one of six intent types:

Specification - Adding words to narrow down ("running shoes" → "Nike marathon running shoes"). Highest purchase intent. Win this with detailed titles, tiered backend keywords, and benefit-rich bullets.

Generalization - Removing words because results disappointed ("KitchenAid stand mixer" → "stand mixer"). Category leaders win here. Make sure you rank for the broad category term, not just long-tail.

Equivalence - Searching by exact model or brand name. They've decided. Make your product name distinctive and searchable.

Substitution - Swapping one brand for another. Amazon proactively suggests competitors. Include competitor brand names in backend keywords. Add competitive comparison modules in A+ Content.

Complement - Searching for accessories after a primary purchase ("camera" → "camera bag"). The window opens the second a related product gets bought. Target "with," "for," "alongside" keywords.

Irrelevant - Topic abandonment. Resets the entire session. Make titles and bullets immediately clear about what your product is and isn't.

Three Patterns That Change How You Sell

The Saturation Cliff. After roughly four narrowing searches without buying, Amazon starts predicting the shopper will switch brands. If you dominated the specification funnel but didn't convert, the system actively pushes competitors next. Solution: convert faster, use price anchoring, reduce friction.

The Post-Purchase Inversion. The moment a shopper buys a primary product, Amazon flips to predicting accessories. If you sell complementary products, your biggest visibility window opens at competitor checkouts, not your own. Optimize for "[primary product name] + accessory" queries.

The Session Reset. One unrelated search wipes the slate clean. No carryover. Every product earns relevance from scratch.

The Personalization Stack

Same query, different results, depending on five layers Amazon attaches to every search: language, geography, demographics, device, and network location. There's no single search results page. There are millions.

Action Items for Your Listing

Title: Lead with core category (positions 1-3 capture Generalization), layer in primary modifier (captures Specification), include filtering words like "professional" or "heavy-duty" (filters out Irrelevant traffic).

Bullets: Bullet 1 confirms what the product is. Bullets 2-4 hit specification benefits. Bullet 5 mentions complement ecosystem ("pairs well with…").

Backend Keywords (starting allocation): 60% specification long-tail, 15% broad category, 15% competitor and equivalence terms, 10% complement queries.

A+ Content: Four modules. Relevance confirmation. Specification depth. Competitive positioning. Complement ecosystem.

Most sellers optimize for one layer: keyword retrieval. The second layer (intent prediction) decides what shoppers search next.

Sellers who engineer their listings around intent transitions, not just keywords, capture customers Amazon is already steering toward them.

Voice, image, and video searches all feed the same intent engine. Optimize once for intent, win across every input type.

Geek out on Andrew’s full breakdown here.

Scott Needham at SmartScout dropped a fascinating analysis breaking down major CPG acquisitions and what slice of their revenue actually came from Amazon.

The headline finding: Amazon usually isn't the main driver of these valuations, but in certain cases it's pulling way more weight than people realize.

The Acquisition List (Sale Price | Amazon % of Revenue)

Two Playbooks, Same Outcome

The data reveals two distinct paths to a big exit:

Retail/omnichannel brands (Poppi, Siete, Rhode, Coterie) treat Amazon as a support channel. Their valuations are built on retail shelf presence, DTC, and brand heat. Amazon is the cherry, not the cake.

Amazon-leveraged brands (Color Wow, Dr. Squatch, GHOST, Grüns, Medik8) lean into the marketplace as a core growth engine. For Color Wow, Amazon is nearly 42% of the business. That's not a side channel, that's the engine.

Effect on Multiples and Sale Price

The pattern worth noting: the brands commanding the highest exit values lean less on Amazon. Poppi at 7.5% sold for $500M. Rhode at 4.7% sold for $100M (with monster multiples on limited revenue). Meanwhile Color Wow's 42% Amazon mix landed at $300M.

This doesn't mean Amazon hurts valuation. It means acquirers pay premium multiples for brand equity, retail distribution, and category dominance that exists independent of any single channel.

An Amazon-heavy brand carries platform risk in the eyes of strategic buyers, fee changes, algorithm shifts, suspension exposure, all real concerns that compress multiples.

But Amazon-leveraged brands still hit nine-figure exits. Dr. Squatch at 30.7% Amazon revenue still landed $200M. The marketplace can absolutely build a sellable business.

The Real Question for Founders

If you're building CPG today, stop asking whether you should be on Amazon. Ask what role Amazon should play in your exit story. Are you building a brand that uses Amazon, or an Amazon business with a brand wrapper? Both can win. They just attract different buyers at different multiples.

Amazon, Meta, Microsoft, Salesforce, and Stripe just joined the Universal Commerce Protocol (UCP) Tech Council, joining founding members Google, Shopify, Etsy, Target, and Wayfair.

What is UCP?

An open standard that lets AI agents (like Gemini, ChatGPT, Copilot) talk to retailers across the entire shopping journey: discovery, cart, checkout, and post-purchase. Think of it as a universal language so every platform doesn't speak a different dialect.

Why Amazon joining is a big deal

Amazon spent 2025 blocking AI bots from OpenAI, Anthropic, Google, and Meta while quietly building Rufus and Buy for Me. Now they're sitting on the council that shapes how outside agents interact with every retailer. They didn't abandon their own playbook. They just decided being inside the room beats being outside it.

Why this matters for Amazon sellers

1. AI agents are becoming the new shelf. Rufus hit 300M users and drove $12B in incremental sales last year. Rufus narrows picks to about 5 products, filtered by structured backend attributes.

2. AI bot traffic to retail sites jumped 5.4x in 2025. OpenAI alone crawls 198 times for every single visit it sends. Your product data needs to be machine-readable, not just shopper-readable.

3. The funnel is collapsing. When an agent picks the product and checks out, there's no landing page, no click, no last-click attribution. Traditional PPC thinking starts to break.

4. Structured data is now table stakes. Titles, bullets, A+ content, and backend attributes need to feed AI systems clearly. Sloppy listings get filtered out before a human ever sees them.

Five of the biggest names in tech and payments are now steering the standard for how AI agents buy things. Amazon is in the room but still running its own walled garden with Rufus.

For sellers, the real takeaway: optimize for AI discovery now. The agents are picking the products, and the criteria they use is structured data, not pretty photos.

Shoppers can now set a rule: when a product hits their target price, Amazon buys it for them automatically.

Think about what this means for sellers.

A customer lands on your listing. Likes the product. Price isn't quite right. Today they bounce. You probably never see them again.

With Auto Buy, they set a target and stay tied to your ASIN. You get a second shot at the sale without spending another dollar on ads.

Here's where it gets interesting.

If Amazon gives sellers access to this data (how many Auto Buy rules are set on your ASIN and at what price points) you'd know exactly how many sales you're leaving on the table. And what price unlocks them.

That's demand curve data straight from the source.

No guessing. No A/B testing prices for weeks. Just real purchase intent tied to real dollar amounts.

We'll see how Amazon rolls this out. But this could be a game-changer for pricing strategy.

✌🏼 See you again Thursday …

The answer to today’s STUMP BEZOS is
Stripe processed $1.9 trillion in payments in 2025.